# METACOGNITIVE THERAPY: SCIENCE AND PRACTICE OF A PARADIGM

EDITED BY : Adrian Wells, Lora Capobianco, Gerald Matthews and Hans M. Nordahl PUBLISHED IN : Frontiers in Psychology

#### Frontiers eBook Copyright Statement

The copyright in the text of individual articles in this eBook is the property of their respective authors or their respective institutions or funders. The copyright in graphics and images within each article may be subject to copyright of other parties. In both cases this is subject to a license granted to Frontiers. The compilation of articles constituting this eBook is the property of Frontiers.

Each article within this eBook, and the eBook itself, are published under the most recent version of the Creative Commons CC-BY licence. The version current at the date of publication of this eBook is CC-BY 4.0. If the CC-BY licence is updated, the licence granted by Frontiers is automatically updated to the new version.

When exercising any right under the CC-BY licence, Frontiers must be attributed as the original publisher of the article or eBook, as applicable.

Authors have the responsibility of ensuring that any graphics or other materials which are the property of others may be included in the CC-BY licence, but this should be checked before relying on the CC-BY licence to reproduce those materials. Any copyright notices relating to those materials must be complied with.

Copyright and source acknowledgement notices may not be removed and must be displayed in any copy, derivative work or partial copy which includes the elements in question.

All copyright, and all rights therein, are protected by national and international copyright laws. The above represents a summary only. For further information please read Frontiers' Conditions for Website Use and Copyright Statement, and the applicable CC-BY licence.

ISSN 1664-8714 ISBN 978-2-88966-244-9 DOI 10.3389/978-2-88966-244-9

#### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

#### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

#### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# METACOGNITIVE THERAPY: SCIENCE AND PRACTICE OF A PARADIGM

Topic Editors:

Adrian Wells, The University of Manchester, United Kingdom Lora Capobianco, Manchester Mental Health and Social Care Trust, United Kingdom

Gerald Matthews, University of Central Florida, United States Hans M. Nordahl, Norwegian University of Science and Technology, Norway

Citation: Wells, A., Capobianco, L., Matthews, G., Nordahl, H. M., eds. (2020). Metacognitive Therapy: Science and Practice of a Paradigm. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88966-244-9

# Table of Contents

# PART 1

# THEORY AND MECHANISMS RESEARCH


Henrik Nordahl, Odin Hjemdal, Roger Hagen, Hans M. Nordahl and Adrian Wells

*34 What Comes First Metacognition or Negative Emotion? A Test of Temporal Precedence*

Lora Capobianco, Calvin Heal, Measha Bright and Adrian Wells


Marie Louise Reinholdt-Dunne, Andreas Blicher, Henrik Nordahl, Nicoline Normann, Barbara Hoff Esbjørn and Adrian Wells

*96 Metacognitive Beliefs Predict Greater Mental Contamination Severity After an Evoking Source*

Thomas A. Fergus, Kelsi A. Clayson and Sara L. Dolan


Cintia L. Faija, David Reeves, Calvin Heal, Lora Capobianco, Rebecca Anderson and Adrian Wells

*127 Neural Correlates of Cognitive-Attentional Syndrome: An fMRI Study on Repetitive Negative Thinking Induction and Resting State Functional Connectivity*

Joachim Kowalski, Marek Wypych, Artur Marchewka and Małgorzata Dragan

*143 Single Dose of the Attention Training Technique Increases Resting Alpha and Beta-Oscillations in Frontoparietal Brain Networks: A Randomized Controlled Comparison*

Mark M. Knowles and Adrian Wells

*152 Shifting Instead of Drifting – Improving Attentional Performance by Means of the Attention Training Technique*

Vincent Barth, Ivo Heitland, Tillmann H. C. Kruger, Kai G. Kahl, Christopher Sinke and Lotta Winter

*160 One Step Ahead—Attention Control Capabilities at Baseline are Associated With the Effectiveness of the Attention Training Technique* Ivo Heitland, Vincent Barth, Lotta Winter, Niklas Jahn, Alev Burak, Christopher Sinke, Tillmann H. C. Krüger and Kai G. Kahl

# PART 2

# CLINICAL OUTCOME RESEARCH

*170 The Efficacy of Metacognitive Therapy: A Systematic Review and Meta-Analysis*

Nicoline Normann and Nexhmedin Morina


Michael Simons and Anna-Lena Kursawe


Peter L. Fisher, Angela Byrne, Louise Fairburn, Helen Ullmer, Gareth Abbey and Peter Salmon

*237 Can the Attention Training Technique Reduce Stress in Students? A Controlled Study of Stress Appraisals and Meta-Worry*

Peter Myhr, Timo Hursti, Katarina Emanuelsson, Elina Löfgren and Odin Hjemdal


Costas Papageorgiou, Karen Carlile, Sue Thorgaard, Howard Waring, Justin Haslam, Louise Horne and Adrian Wells

*263 Group Metacognitive Therapy for Generalized Anxiety Disorder: A Pilot Feasibility Trial*

Svein Haseth, Stian Solem, Grethe Baardsen Sørø, Eirin Bjørnstad, Torun Grøtte and Peter Fisher

*273 A Randomized Controlled Trial of Metacognitive Therapy for Depression: Analysis of 1-Year Follow-Up*

Odin Hjemdal, Stian Solem, Roger Hagen, Leif Edward Ottesen Kennair, Hans M. Nordahl and Adrian Wells

*282 A Comparison of Metacognitive Therapy in Current Versus Persistent Depressive Disorder – A Pilot Outpatient Study*

Lotta Winter, Julia Gottschalk, Janina Nielsen, Adrian Wells, Ulrich Schweiger and Kai G. Kahl

*290 Neurobiological Mechanisms of Metacognitive Therapy – An Experimental Paradigm*

Lotta Winter, Mesbah Alam, Hans E. Heissler, Assel Saryyeva, Denny Milakara, Xingxing Jin, Ivo Heitland, Kerstin Schwabe, Joachim K. Krauss and Kai G. Kahl

*300 Metacognitive Therapy Versus Cognitive Behavioral Therapy: A Network Approach*

Sverre Urnes Johnson and Asle Hoffart

*310 Metacognitive Beliefs as Predictors of Return to Work After Intensive Return-to-Work Rehabilitation in Patients With Chronic Pain, Chronic Fatigue and Common Psychological Disorders: Results From a Prospective Trial*

Henrik B. Jacobsen, Mari Glette, Karen W. Hara and Tore C. Stiles

*318 Innovation in Psychotherapy, Challenges, and Opportunities: An Opinion Paper*

Janina Isabel Schweiger, Kai G. Kahl, Jan Philipp Klein, Valerija Sipos and Ulrich Schweiger

# Editorial: Metacognitive Therapy: Science and Practice of a Paradigm

#### Adrian Wells 1,2 \*, Lora Capobianco<sup>2</sup> , Gerald Matthews <sup>3</sup> and Hans M. Nordahl <sup>4</sup>

*<sup>1</sup> Faculty of Biology Medicine and Health, School of Psychological Sciences, University of Manchester, Manchester, United Kingdom, <sup>2</sup> Research and Innovation, Greater Manchester Mental Health National Health Service Foundation Trust, Manchester, United Kingdom, <sup>3</sup> Institute for Simulation and Training, University of Central Florida, Orlando, FL, United States, <sup>4</sup> Department of Mental Health, Norwegian University of Science and Technology, Trondheim, Norway*

Keywords: metacognitive therapy, S-REF model, metacognition, psychological disorders, causal mechanisms

#### **Editorial on the Research Topic**

#### **Metacognitive Therapy: Science and Practice of a Paradigm**

One of the greatest challenges facing mental health research is the development and testing of bone-fide causal theories of psychopathology that inform the development of more effective treatments. Unfortunately, apart from the major progress offered by cognitive-behavior therapy over 40 years ago, there have been few advances in models and treatments that have improved outcomes. Developments are hindered by the prevailing clinical research strategy that has attempted to innovate by combining therapeutic techniques taken from a wide range of existing sources, but in the absence of an understanding of causal mechanisms. This raises crucial questions: how can the researcher or practitioner know which of the plethora of techniques to choose, should they be combined or used in the absence of a theoretical rationale and are they compatible?

It is evident that progress could be made by developing a more rigorous, scientifically grounded theory of causal mechanisms, and devising treatment techniques ground-up from this theoretical platform. This approach was used by Wells and Matthews (1994), (see also Wells and Matthews, 2015) in the development of their S-REF model, which offered the early foundations of metacognitive therapy (MCT) (Wells, 2009, 2019) based on the cognitive science of emotion. The present Research Topic aims to capture the breadth of current ideas and studies in MCT and bring together active researchers at the forefront of the field. The objectives are to demonstrate the universal influence of MCT, present data probing theoretical mechanisms and to offer a grounding from which ideas can spring that will support future investigations.

There are 30 articles in this issue covering advanced theory, evaluation of mechanisms, clinical evaluations of treatment efficacy, feasibility of novel applications of treatment and studies of assessment tools. The articles consist of clinical, non-clinical, cognitive and neuroscience studies, research in adults and children, and studies of personality, stress, psychosis, alcohol abuse, anxiety, trauma, obsessions and depression. The articles are grouped into two clusters. First, work on theory and mechanisms is presented and this is followed by studies of the clinical effects of MCT.

MCT is based on some basic principles central to the S-REF model: (1) most disorders are caused by a common or transdiagnostic set of processes made up of difficult to control extended negative thinking, (2) psychological distress is prone to self-correct but is thwarted in doing so by maladaptive self-regulatory strategies, (3) metacognitions are key to adaptive and maladaptive self-regulatory processes.

Wells elaborates on the original S-REF model and makes important and more detailed distinctions between cognitive and metacognitive structures and processes, drawing out the necessary components and hypothesized circuits in formulating adaptive and maladaptive selfregulation. The paper describes a metacognitive control system involved in psychological disorders

Edited and reviewed by: *Nuno Barbosa Rocha, Polytechnic of Porto, Portugal*

\*Correspondence: *Adrian Wells adrian.wells@manchester.ac.uk*

#### Specialty section:

*This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology*

Received: *25 June 2020* Accepted: *17 August 2020* Published: *18 September 2020*

#### Citation:

*Wells A, Capobianco L, Matthews G and Nordahl HM (2020) Editorial: Metacognitive Therapy: Science and Practice of a Paradigm. Front. Psychol. 11:576210. doi: 10.3389/fpsyg.2020.576210*

**6**

and elucidates different types of metacognitive information that influence the way cognition is experienced. Repetitive and extended negative processing that maintains psychological distress is a process normally prone to decay but this is thwarted by maladaption in the metacognitive system leading to persistence of negative processing. The model leads to predictions of the existence of several important mechanisms and types of metacognitive information including "cybernetic code" generated by the metacognitive system that impact on neural networks and contribute to emotional recovery or sustained processing and psychological disorder. The model is broadened to consider how metacognitive information and its flow between systems helps to create embodiment, self-awareness and meta-representational states that provide resources for selfregulation. The paper concludes by exploring how the model has shaped the development and focus of MCT and explores the implications for future treatment development and advances in theory and research.

Concepts of neuroticism and trait-anxiety are widely used to measure psychological vulnerability. Nevertheless, they can be limiting because they do not identify the underlying mechanisms of disorder; instead, they focus on the likelihood of experiencing symptoms. Nordahl et al. show that negative and positive beliefs about worry are both cross sectional and prospective predictors of trait-anxiety, suggesting that dysfunction in metacognition might be the underlying mechanism that is captured by emotion-vulnerability measures. The direction of causality in metacognition-emotion relationships is addressed by Capobianco et al. using cross-lagged structural equation modeling. They found that metacognitive beliefs predicted subsequent anxiety and anxiety predicted subsequent metacognition over different time-courses suggesting mutual causal links that might (if measured over a longer time-frame) constitute a dysfunctional metacognition-emotion cycle. One important way to examine emotional vulnerability is to assess multiple traits that contribute not only to dysfunction but also those that may confer the opposite; psychological resilience. Matthews et al. examined the effects of metaworry, worry and resilience traits on the performance of a complex task under two types of stressor, differing in self-reference. Meta-worry was associated with subjective stress and EEG responses to the more self-referent stressor (negative feedback). Moderator effects on associations between state worry, performance and EEG measures suggested that high trait meta-worry blocks adaptation to stress through compensatory effort.

The Metacognitions Questionnaire (MCQ) is the most commonly used measure of adult metacognitive beliefs linked to disorder in the metacognitive theory. The MCQ has also been adjusted for use in children and adolescents as reviewed in this special issue by Myers et al. These authors examined the psychometric properties of variants of the MCQ, demonstrating a similar latent structure, reliability, and validity estimates in child versions to those obtained in the original scale. Furthermore, theoretically expected relationships between metacognitions and emotion disorder symptoms are evident, resembling those found in adults.

Utilizing the child version of the MCQ, Reinholdt-Dunne et al. demonstrated elevated dysfunctional metacognitions and lower self-report attentional control in a clinical compared with a community sample of 7–14 year-olds. In the community but not the clinical sample, MCQ-total interacted with attentional control in explaining symptoms of anxiety. The result is consistent with the idea that detrimental effects of metacognitive beliefs might be remediated by high-levels of perceived attention control.

Fergus et al. specifically examined the effect of metacognitive beliefs on mental contamination (feelings of internal dirtiness) in women who had experienced sexual trauma. Following exposure to an evoking stimulus, metacognitions concerning uncontrollability and danger, low cognitive confidence, and need to control thoughts positively correlated with the severity of mental contamination. The strength of relationship between specific metacognitive beliefs and symptoms of psychological disorder is likely to be subject to a range of other metacognitive influences as specified in the MCT model. Bardeen and Fergus examined this issue in the context of PTSD symptoms. They found that amongst adults exposed to trauma, deficits in executive control strengthened the positive association between positive metacognitive beliefs (e.g., "worrying will keep me safe") and PTSD symptom scores. The positive relationship between negative metacognitive beliefs and symptoms was not moderated by executive control.

Two articles in this special issue specifically examine the Cognitive Attentional Syndrome (CAS) defined as a combination of repetitive negative thinking, unhelpful coping strategies and underlying dysfunctional metacognitions. In one of these studies, Faija et al. report on the Cognitive Attentional Syndrome Scale-1 (CAS-1) adapted for research in cardiac rehabilitation patients reporting anxiety and depression. A three-factor solution was supported by confirmatory factor analysis composed of coping strategies, negative, and positive metacognitive beliefs. Each subscale independently contributed to anxiety while coping strategies independently contributed to depression symptoms.

One way to objectively validate the effects of the CAS as proposed in the S-REF model is by testing for specific neural correlates of this syndrome, a task undertaken by Kowalski et al. Their study explored the neural correlates of the CAS using fMRI during induced negative thinking. Low- and high-CAS groups differed in functional connectivity during induced negative and abstract thinking and also in resting state fMRI. The results suggest disrupted self-referential processing in individuals who score high on self-report CAS dimensions.

Three articles report on laboratory-based effects of an individual MCT treatment technique; the Attention Training (ATT). ATT was developed to attenuate the CAS, by reducing self-focused processing and strengthening knowledge concerning flexible control of thinking. Knowles and Wells demonstrated that a single session of the ATT increased resting alpha and beta oscillations in front-parietal brain regions when compared with a control condition. The signature and location of effects is consistent with the ATT affecting executive control processes for which it was designed. Continuing with the evaluation of objective ATT effects. Barth et al. tested effects on attentional performance across attention bias, inhibition, working memory and disengagement tasks. The results showed specific effects on attention bias suggesting that ATT might promote greater attention flexibility in healthy subjects. In a related paper, Heitland et al. tested whether pre-treatment attentional control was related to these effects of ATT. Individuals who scored high in self-report attentional control at pre-intervention showed the largest improvements in attention task performance. The data imply that pre-existing metacognitions concerning attention control might moderate the effectiveness of the ATT.

The effectiveness of metacognitive therapy has been tested with a range of methodologies in clinical and non-clinical participants. In their paper, Normann and Morina present a systematic review and meta-analysis of randomized trials of MCT for anxiety and depression disorders. The data appear to show that MCT is highly effective in reducing primary symptoms of anxiety or depression, secondary symptoms and hypothesized causal variables. The magnitude of effects reported seem larger than those of comparison treatments classified as cognitivebehavioral therapies. Direct comparisons of MCT with CBT in disorders such as generalized anxiety (Nordahl et al., 2018) and major depression (Callesen et al., 2020), published elsewhere, add particular weight to these results.

The special issue incorporates a series of papers reporting novel applications of MCT. Some of these are small scale or non-randomized treatment-related feasibility and acceptability papers. They are of course limited by lack of control for nonspecific factors and low generalizability, but they are crucial early steps in generalizing treatment applications and offer proof of principle prior to investment in large -scale efficacy research.

Nordahl and Wells apply MCT to treating traumatized patients with Borderline Personality Disorder. This is the first evaluation of MCT with this client group, who suffer from emotion dysregulation, self-harm and impulse regulation difficulties. The study suggests that MCT is a feasible and acceptable treatment in this context. The within-group effect sizes seemed to compare favorably with the outcomes observed in other forms of therapy. Maintaining the trauma theme Simons and Kursawe conducted a feasibility study of MCT for PTSD in children and adolescents (ages 8–19 years). Treatment was associated with large improvements in PTSD symptoms and high recovery rates. The results show MCT is feasible and acceptable in traumatized youth as young as 8 years of age and justify larger scale studies.

Parker et al. explored the feasibility and acceptability of MCT in Individuals at high risk of developing psychosis. The majority of patients were able to complete treatment and gains on psychosis symptoms and secondary measures were observed. Retention at 6-month follow-up was lower and this is an area future studies should consider in the planning phase. The result is consistent with an earlier study on medication resistant patients with schizophrenia, suggesting that treatment with MCT is feasible and might be associated with significant benefit (Morrison et al., 2014). In the study reported by Caselli et al., single-case methodology was used to replicate treatment-related effects across five individuals with alcohol abuse. All patients showed clinically meaningful reductions in weekly alcohol use and number of binge drinking episodes. Winter, Naumann et al. applied MCT to adjustment disorder in a patient suffering from pulmonary arterial hypertension with noticeable gains during treatment in psychological and behavioral outcomes. The application of MCT in medical conditions is also the theme in the paper by Fisher et al.. In their study, anxiety and depression symptoms were treated in 27 cancer survivors across 6 treatment sessions. MCT appeared feasible and acceptable with 75% of patients completing the full course. Treatment appeared to be associated with large improvements in symptoms.

Exploration of the novel application of MCT not only addresses diagnoseable problems but is applied to modifying stress-related processes in the study by Myhr et al.. Here, college students who received the attention training technique (ATT) showed significant improvement in meta-worry and perceived stress compared to those that did not. The outcome indicates that ATT may reduce negative appraisal processes at both metacognitive and cognitive levels within the context of academic stress symptoms.

Most often, treatment is delivered in a one-to-one interaction between patient and therapist, but the nature of MCT, focusing on universal mechanisms, means it should be well-suited to application in groups and trans-diagnostically. These topics are addressed in three papers presented in the special issue. Callesen et al. report an uncontrolled evaluation of MCT when applied to a group of patients with a range of different diagnoses. Large pre to post treatment improvements in symptoms were observed during treatment sessions, and treatment gains appeared to be stable over follow-up.

The study by Papageorgiou et al. examined group treatment of obsessive-compulsive disorder and compared group delivered MCT with group delivered CBT. The study provides additional interest because MCT was introduced within a particular service as an attempt to improve patient outcomes beyond CBT that was traditionally offered. Whilst there was no randomization, the study is based on a benchmarking of effects of each treatment in large samples as a pragmatic evaluation of service change. CBT was associated with large improvements in OCD and related symptomatology and the effects compared favorably with those reported in the literature, but MCT was associated with better outcomes.

Several published studies have tested the effects of MCT in the treatment of GAD, and MCT is recognized in NHS NICE guidelines as a treatment option. Most often the treatment is delivered on a one-to-one-basis. Haseth et al. contribute to the group treatment literature in their feasibility study of group MCT applied to patients suffering from generalized anxiety disorder. Out of 23 consecutively referred patients 19% declined group MCT in favor of individual MCT. The group intervention was associated with a 65% recovery rate at post treatment and 78% at 3 month follow-up.

Depression is the second largest cause of global disability and a major contributor to risk through self-harm and suicide. MCT is proving that it might be a highly effective treatment for depression as shown in recent studies. Hjemdal et al. present 1-year follow-up data on their randomized trial of MCT for major depression. The results suggest a high level of maintenance of positive treatment effects following MCT, with 67% (intention to treat) and 75% of those who completed treatment classified as recovered. This is encouraging in a condition that normally has high rates of relapse. An important issue in depression treatment centers on the management of recurrent or persistent depression cases. It has been suggested that such cases require special considerations and a different treatment approach. Winter, Gottschalk et al. compared the response to MCT of patients with major depression or persistent depression. All of the persistent depression group had failed to benefit from antidepressant treatment and most of them had also received previous psychotherapy. Both sets of patients showed large and similar levels of improvement in symptoms and rates of remission during MCT and at follow-up.

Two studies in the special issue examine the question of mechanisms of change in MCT. Another study by Winter, Alam et al. capitalized on the opportunity to directly read neuronal local field signals from implanted brain electrodes in a patient with OCD during a series of MCT treatment techniques. OCD symptoms decreased after treatment and increases in alpha, beta and gamma bands and reduced theta were detected. In a different study, Johnson and Hoffart analyzed mechanism data from their earlier trial where they reported that transdiagnostic MCT was more effective than disorderspecific CBT for anxiety disorders. They found that both MCT and CBT shared some mechanisms of change; worry and attention, but additionally central to MCT was change in metacognitive beliefs about uncontrollability. Interestingly, the set of change mechanisms (in both treatments) that seemed important would be better captured by the S-REF model than by a CBT model.

# REFERENCES


Jacobsen et al. present evidence of metacognition predicting response to treatment. They delivered a brief return-to-work rehabilitation package. Whilst it was not an MCT based intervention the authors did assess metacognitive beliefs at pretreatment and their change during treatment. Pre-treatment metacognitions were not related to return to work, but reduction in metacognitive beliefs about the need to control thoughts gave 20% greater odds of returning to work over 1 year.

Innovation in psychotherapy through the systematic use of theory-driven empiricism has been the guiding principle behind MCT development, and is a process amply demonstrated in the array of papers in this issue. The process of MCT development has eschewed the integrative and eclectic technique-driven approach in favor of developing strong theory, grounded in cognitive psychology that can inform the discovery of mechanisms of disorder and the design of specific treatment techniques. This theme is discussed in the opinion paper by Schweiger et al., in a wider context of innovation in psychotherapy. They raise important discussion questions that invite a retrospective evaluation of the barriers that have existed (and still exist in many areas) in psychotherapy evolution. They show how the process of development used in MCT offers a model that might be adopted more widely in improving psychotherapy research and treatment outcomes in the future.

# AUTHOR CONTRIBUTIONS

AW acted as chief guest editor and conceived the Research Topic. All authors contributed to reviewing manuscripts for the special issue, acted in an editorial capacity and contributed to drafts of the editorial.


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Wells, Capobianco, Matthews and Nordahl. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Breaking the Cybernetic Code: Understanding and Treating the Human Metacognitive Control System to Enhance Mental Health

#### *Adrian Wells1,2 \**

*1School of Psychological Sciences, Faculty of Biology, Medicine and Health, The University of Manchester, Manchester, United Kingdom, 2Greater Manchester Mental Health NHS Foundation Trust, Manchester, United Kingdom*

The self-regulatory executive function (S-REF) model explains the role of strategic processes and metacognition in psychological disorder and was a major influence on the development of metacognitive therapy. The model identifies a universal style of perseverative negative processing termed the cognitive attentional syndrome (CAS), comprised of worry, rumination, and threat monitoring in the development of disorder. The CAS is linked to dysfunctional metacognitions that include beliefs and plans for regulating cognition. In this paper, I extend the theoretical foundations necessary to support further research on mechanisms linking metacognition to cognitive regulation and effective treatment. I propose a metacognitive control system (MCS) of the S-REF that can be usefully distinguished from cognition and is comprised of multiple structures, information, and processes. The MCS monitors and controls activity of the cognitive system and regulates the behavior of neural networks whose activities bias the way cognition is experienced. Metacognitive information involved in the regulation of on-line processing includes metacognitive beliefs, metacognitive procedural commands, and more transient cybernetic code. Separation of the cognitive and metacognitive systems and modeling their relationship presents major implications concerning what should be done in therapy and how it should be done. The paper concludes with an in-depth consideration of methods that strengthen the psychological basis of psychotherapy and aid in understanding and applying metacognitive therapy in particular. Finally, limitations of the model and implications for future research on self-awareness, self-regulation, and metacognition are discussed.

Keywords: metacognitive therapy, metacognition, self-awareness, transdiagnostic mechanisms, cognitive behavior therapy, neural networks, embodiment, attention

# INTRODUCTION

Throughout the last 25 years, the Self-Regulatory Executive Function (S-REF) model (Wells and Matthews, 1994, 1996) has stimulated a large volume of research on cognitive control processes in psychological disorder and is the grounding of an effective psychological treatment: metacognitive therapy (MCT: Wells, 1995, 2009). In this paper, I consider the central principles of the model in light of recent evidence and expand on the functional components of its metacognitive control system. The aim is to provide a theoretical framework to stimulate and

#### *Edited by:*

*Changiz Mohiyeddini, Northeastern University, United States*

#### *Reviewed by:*

*Giancarlo Dimaggio, Centro di Terapia Metacognitiva Interpersonale (CTMI), Italy Gabriele Caselli, Sigmund Freud University Vienna, Austria*

*\*Correspondence: Adrian Wells adrian.wells@manchester.ac.uk*

#### *Specialty section:*

*This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology*

*Received: 21 June 2019 Accepted: 06 November 2019 Published: 12 December 2019*

#### *Citation:*

*Wells A (2019) Breaking the Cybernetic Code: Understanding and Treating the Human Metacognitive Control System to Enhance Mental Health. Front. Psychol. 10:2621. doi: 10.3389/fpsyg.2019.02621*

**10**

advance future research on varieties of metacognitive information, processes, and structures in psychological disorder, self-awareness, and treatment.

# HISTORICAL CONTEXT OF THE SELF-REGULATORY EXECUTIVE FUNCTION MODEL

Our initial aim in the work leading to the S-REF was to take a robust scientific approach that was deeply rooted in cognitive psychology to develop an explanation of the mechanisms behind psychological disorder. That aim culminated in our book, *Attention and Emotion: A Clinical Perspective*; first published in 1994 and since re-published (Wells and Matthews, 1994, 2015). Our goal was to generate testable theory-based predictions that would lead to clinical innovation.

The S-REF model aimed to explain laboratory-based data on attention bias, individual differences in stress responses, and the cause of psychological disorder. This did not turn out to be a simple task, but it was a controversial one. The prevailing view at the time was that psychological disorder was largely an effect of bottom-up (automatic) stimulus-driven biases in processing resulting from schemas or associative networks. We questioned this view, setting out a model based on alternative mechanisms, involving maladaptation in top-down volitional cognitive control, arguing that clinical disorder is associated with a reduction in dynamic control and adaptability.

The application of cognitive psychology principles in the field of psychopathology and treatment was limited when we began. Innovative research on attention in anxiety (Mathews and MacLeod, 1985, 1986; Williams et al., 1988; Mathews et al., 1990; MacLeod, 1991) demonstrated that patients are characterized by a bias toward information with negative content. Our initial goal was to attempt to explain such selective processing. What might lead the emotional disordered patient to focus on negative information? We began by evaluating the success of existing theory in accounting for biased attention and its success in accommodating important attention factors; capacity limitation and distinctions between voluntary and involuntary (automatic) processes.

Influential models of psychological disorders centered on memory structures (e.g. schemas or associative networks) as key causes of disorder and the major treatment approaches focused primarily on the content of these structures and related cognitions. For example, Beck's cognitive theory (Beck, 1976; Beck et al., 1985) of emotional disorders assigned a prominent role to the content of beliefs and interpretations in disorder, identifying the negative triad in depression and a preponderance of thoughts about danger in anxiety (e.g. "I'm going to physically collapse"). In contrast, we argued that maladaptation occurs principally due to volitional biases in executive control, in the selection of self-regulation strategies; the emotionally vulnerable person selecting those strategies that prolonged rather than terminated negative processing. Increasingly, we became aware of limitations of the schema and "automaticity" concepts as an explanation of these features of processing. In particular, they failed to account for the individuals influence over whether or not to continue with current processing. For instance, the content of self-knowledge or schemas (e.g. "I'm a failure as a mother") does not explain bias in attention or cognitive regulation because the individual retains choice in whether or not to continue analyzing their failures. In effect, the role of top-down or executive processes in the regulation of processing necessitated elaboration. Therefore, our model aimed to explain how voluntary (executive processes) and involuntary processes interacted with stored knowledge, especially *metacognition* in the regulation of processing.

Metacognition refers to the structures, content, and processes involved in the monitoring, appraisal, and control of cognition. Sometimes loosely defined as that part of cognition that is turned onto itself, this simple definition may be misleading, because it suggests a single structure of cognition responsible for cognition and metacognition. Seminal work on metacognition prior to the S-REF model was predominantly in developmental, educational, and memory psychology with defining contributions of Flavell (1979), Nelson and Narens (1990), and colleagues.

In order to develop a comprehensive model of cognitive control and the prioritizing of negative processing, we predicted a central contribution of dysfunctional metacognition and attentional control plans stored in long term memory. Subsequently, the metacognitive component of the model was elaborated as the basis for metacognitive therapy (Wells, 1995, 2000, 2009), and the model was extended with greater detail of features of its architecture and metacognitive components (especially metacognitive beliefs). However, the central tenets of the theory and its implications, emphasizing universal top-down influences, remain the same.

The S-REF model has influenced the development of other treatment approaches. For example, Clark and Wells (1995) advanced a model and treatment of social phobia that has proven effective (Clark et al., 2006; Nordahl et al., 2016) and is a recommended intervention in health guidelines (NCCMH, 2013). Wider influences of the S-REF on psychotherapy are apparent as extensions of CBT, for example, "emotional schema" theory and treatment (Leahy, 2015). While in a separate line of work, metacognition has been formulated differently by Dimaggio et al. (2015) in their therapeutic approach of interpersonal therapy in personality disorder and by Moritz and Woodward (2007) in metacognitive training for schizophrenia.

# OUTLINE OF THE SELF-REGULATORY EXECUTIVE FUNCTION MODEL

The S-REF model is based on the principle that most psychological disorders are the result of a universal style of cognition and behavior termed the Cognitive Attentional Syndrome (CAS). The CAS is a state of processing where negative self-relevant information is prioritized and becomes perseverative (i.e. extended and repetitive). The most common types of perseveration include worrying or ruminating (brooding) on negative and threatening events such as how to deal with future threats or trying to understand past events and feelings. In addition to worry and ruminations, the CAS is also comprised

of attentional strategies of "threat-monitoring" such as checking for symptoms or thoughts or scanning the environment for specific signs of danger (e.g. contamination or personal rejection). Added to these elements are other forms of problematic behavior such as avoidance, inactivity, thought suppression, or substance use. These strategies intensify and extend negative processing. They also reduce direct experiences of discontinuation of processing by the mind itself.

An illustration of the CAS and its effects can be seen in a depressed patient who when questioned about feelings of lethargy reported: "I don't have the strength to cope" and described how subsequently he responded to this cognition by analyzing why he lacked energy, compared himself with other people, repeatedly questioned why he felt depressed, closely monitored his feelings of fatigue, engaged in self-criticism in an attempt to increase motivation, and reduced activity levels in order to conserve strength. This constellation of responses prolonged negative self-focused processing and undermined his subjective ability to deal with situations.

In the S-REF model, the CAS is caused by the individual's metacognitive knowledge (Wells and Matthews, 1994, 1996), and such knowledge is formulated as a major target in metacognitive therapy (Wells, 1995, 2000). A distinction is made between declarative and procedural metacognitive knowledge. The declarative can be expressed verbally as beliefs about thinking (e.g. "worrying is harmful"), whilst procedural knowledge exists as implicit instructional information (i.e. commands or "plans") that inform the cognitive system how to operate (e.g. the instructions behind generating worry or rumination).

The declarative metacognitive beliefs in psychopathology can be further divided into those that are positive or negative. The positives concern the usefulness of CAS strategies such as worry, rumination, and attending to threat (e.g. "Worrying means I'm always prepared"), while the negatives concern the uncontrollability and harmfulness of cognition (e.g. "I have lost control of my thinking" and "Some thoughts can harm me"). The latter are considered of greater causal significance in disorder because beliefs concerning the uncontrollability and danger of cognition interfere with effective control and lead to omnipresent threat from an internal process; cognition itself (Wells, 1995).

It is evident in the S-REF analysis that the cognitive and neural architecture accommodates strategic processes such as worry, rumination, and threat monitoring that are conceptualized as serving personal self-regulatory goals and are linked to metacognition. However, many of the constructs in our model were new and therefore a research program was needed to develop tools for measuring metacognitive beliefs (Cartwright-Hatton and Wells, 1997), thought control strategies (Wells and Davies, 1994), and types of worry (Wells, 1994, 2005a) to facilitate model testing.

A significant proportion of work in this domain was enabled by developing the metacognitions questionnaire (MCQ; Cartwright-Hatton and Wells, 1997, Wells and Cartwright-Hatton, 2004), a measure of beliefs about thinking. The MCQ measures five domains of metacognitive knowledge each on a separate subscale: negative beliefs about thoughts concerning uncontrollability and danger (e.g. "When I start worrying I cannot stop"); positive beliefs about worrying (e.g. "Worrying helps me to avoid problems in the future"); cognitive confidence (e.g. "I have a poor memory"); need for mental control (e.g. It is bad to think certain thoughts"); and cognitive selfconsciousness (e.g. "I constantly examine my thoughts"). These domains represent the declarative knowledge or information that individuals hold about thinking and are considered linked to the procedural knowledge or the commands of the S-REF that influence processing.

# SCIENTIFIC STATUS OF THE SELF-REGULATORY EXECUTIVE FUNCTION MODEL

The S-REF model emphasized common processes in psychological disorder, predicting universal, or transdiagnostic abnormalities in attention (e.g. threat monitoring), metacognition and perseveration. Consistent with this prediction, attentional bias has been demonstrated across different traits and disorders (Bar-Haim et al., 2007; Cisler and Koster, 2010; Staugaard, 2010; Techmann et al., 2010; Epp et al., 2012), and universal dysfunction in metacognitive beliefs has been shown across pathologies (e.g. Sun et al., 2017). In the next section, data on metacognitions and the CAS will be considered. Several extensive reviews of biased attention can be found in the literature elsewhere (e.g. Bar-Haim et al., 2007; Cisler and Koster, 2010; Epp et al., 2012).

# Metacognitive Beliefs

It is now reliably established that metacognitions are elevated across psychological disorders and are associated meaningfully with perseverative styles of negative thinking (e.g. worry, rumination) and emotional vulnerability as our model predicted (Cartwright-Hatton and Wells, 1997; Wells and Cartwright-Hatton, 2004; Spada et al., 2008; Nordahl et al., 2019). In a meta-analysis of 45 studies including 3,772 patients and 3,376 healthy individuals, Sun et al. (2017) showed elevated dysfunctional metacognitions across patients, with large and robust effects for beliefs concerning the uncontrollability and danger of worry and beliefs about the need to control thoughts. Of particular note, researchers have demonstrated that the metacognitions of the S-REF model appear to be stronger and more reliable predictors of psychological vulnerability and symptoms of disorder than the content of cognition (Gwilliam et al., 2004; Myers and Wells, 2005; Spada et al., 2007; Myers et al., 2009; Bennett and Wells, 2010; Bailey and Wells, 2016; Nordahl and Wells, 2017). Furthermore, change in metacognitions during treatment appears to predict positive outcome better than change in cognition (Solem et al., 2009; Nordahl et al., 2017), while pre-treatment metacognition may also impact on outcomes (e.g. Spada et al., 2009). Development of more specific metacognitive belief measures for depressive rumination, alcohol use, and health anxiety add further evidence of positive relationships between metacognitive knowledge, problematic affect, and behaviors (Papageorgiou and Wells, 2003, 2009; Spada and Wells, 2008; Bailey and Wells, 2015a). In addition, prospective studies support the role of elevated metacognition as a precedent to elevated emotion disorder symptoms (Myers et al., 2009; Yilmaz et al., 2011; Capobianco et al., 2019) and as a moderator of the effects of cognition on anxiety (Bailey and Wells, 2015b).

Experimental studies have sought to manipulate metacognitive beliefs directly to test their causal impact on symptoms. Rassin et al. (1999) tested the effect on obsessional thoughts in a non-clinical sample. Participants were led to believe that an EEG apparatus to which they were connected would detect the occurrence of the thought: "apple" and on doing so would deliver an electric shock to another participant they had just met. The participants were informed that they could interrupt the electric shock by pressing a button within 2 s after the word "apple" had surfaced in their consciousness. In a comparison condition, participants were told that the EEG could detect the thought "apple," but no information about shocks was given. Thus, the experimental condition can be interpreted as inducing metacognitive beliefs about the power of the thought "apple" to cause an electric shock unless the participant acts to prevent it. The experimental condition resulted in more intrusive thoughts, greater discomfort, more internally directed anger, and greater effort to avoid thinking.

In an extension and modification of this paradigm, Myers and Wells (2013) selected non-patients who scored high and low on a measure of obsessional symptoms and randomly allocated them to a metacognitive belief induction or control condition. All participants were connected to a fake EEG apparatus and asked to watch a video about drinking water. Following the video, participants in the experimental group were led to believe that having thoughts about drinking would be detected by the EEG apparatus and if so a burst of white noise sufficient to startle them might be generated through headphones. The control group were informed that the EEG apparatus could detect thoughts about drinking, and they may receive a random burst of white noise sufficient to startle them. Therefore, only the experimental group were led to believe the aversive loud noise could be caused by their thoughts. Consistent with study hypotheses, participants high in obsessions in the experimental group reported significantly more intrusions about drinking, more time thinking about them and greater discomfort than high obsession participants in the control group.

Capobianco et al. (2018b) used the fake EEG paradigm to induce negative metacognitive beliefs about the importance of thoughts and explore their effects on stress responses. Participants were led to believe that an EEG device could detect negative thoughts and in the experimental condition this might lead to a burst of white noise. In the control condition, the noise was introduced as possibly occurring at random (there was no actual noise exposure in any condition). All subjects underwent the Trier Social Stress Test to induce stress symptoms that were measured across the study and during a 10-min recovery period. On physiological measures (skin conductance), no differences were observed between groups. But on self-report outcomes, participants in the experimental condition reported greater negative affect and lower positive affect in response to the stressor and maintained lower positive affect at recovery than control participants.

# The Cognitive Attentional Syndrome

Turning to data on the CAS, a substantial body of research supports negative effects of worry (see Davey and Wells, 2006) and rumination (see Papageorgiou and Wells, 2004) on stress responses, emotion recovery, and psychological vulnerability. Matthews et al. (1999) showed that test-anxiety measured at a trait level was positively related to maladaptive metacognition and worry (which together loaded on a general factor) and to style of coping. Furthermore, the effects of worrying appear to be influenced by metacognition in some contexts. In a study of performance under evaluative stress, the effects of high worry states on performance and psychophysiological outcomes were moderated by metacognition (i.e. meta-worry), perhaps reflecting the impact of metacognition on compensatory effort or resource allocation (Matthews et al., 2019). The impact of the CAS on symptoms of psychopathology has additional metacognitive moderators; high perceived attention control appears to reduce the strength of association between the CAS and disorder symptoms (Fergus et al., 2012).

Studies of individual differences in the control of distressing thoughts provide reliable support for the predicted negative effects of using CAS-related strategies and the ubiquity of strategies such as worry across different disorders and symptoms. A large number of studies have used the thought control questionnaire (TCQ: Wells and Davies, 1994). The TCQ separately assesses the use of worry and self-punishment, and other occasionally more adaptive strategies of distraction, social control, and reappraisal. As predicted, worry, and self-punishment are positively associated with psychological disorder symptoms (Amir et al., 1997; Warda and Bryant, 1998; Morrison et al., 2000; Roussis and Wells, 2006). The results of longitudinal analyses of traumatic stress symptoms suggest that they may have a causal role (Holeva et al., 2001; Roussis and Wells, 2008). While these data show that CAS is reliably correlated with symptoms of psychological disorder, the CAS is also distinguishable from other constructs such as psychological flexibility that are emphasized in other approaches such as relational frame theory (Fergus et al., 2013). Symptom correlates of the CAS observed in stress and emotional disorder generalize to psychosis confirming the universality of these relationships. In their systematic review, Sellers et al. (2017) identified 51 eligible studies among which findings confirmed specific positive relationships between central elements of the CAS and experiences of psychosis and psychological distress.

Experimental manipulations of CAS processes demonstrate effects on emotional outcomes and recovery from stress that are consistent with the S-REF. The induction of worry or rumination under laboratory settings maintains cognitive and emotional symptoms following stress exposure. In early work, pre-dating the S-REF model, Borkovec et al. (1983) showed that a brief period of induced worry led to greater intrusive thoughts during a subsequent non-worry task. Subsequently, Wells and Papagerogiou (1995) and Butler et al. (1995) studied the effects of induced brief worry and other forms of mentation after exposure to a stressful film and showed that worry increased the frequency of intrusive images most over a subsequent 3-day period. Reviews by Nolen-Hoeksema (1991, 2000) and Lyubomirsky and Tkach (2004) describe experimental and correlational studies demonstrating that ruminative thinking about the implications of depressive symptoms maintains those symptoms, impairs problem solving, and is associated with worse emotional outcomes after stressful life

events. Capobianco et al. (2018a) tested whether specific CAS responses delayed recovery from stress. Participants were randomly assigned to CAS conditions or a distraction control condition and exposed to the Trier social stress test. The rate of recovery from self-report negative affect and physiological stress (Galvanic Skin Conductance) was monitored. Compared to a distraction condition, rumination appeared to impact on skin conductance indicating a prolonged recovery on this index, while worry subjects reported more immediate delayed recovery marked by an initial elevation in self-reported negative affect scores.

# REVISITING THE CONTROL OF COGNITION

Schneider and Shiffrin (1977) contrast *automatic processing* that is fast and reflexively triggered by inputs and runs with little or no conscious involvement with *controlled or "strategic" processing*, which requires varying quantities of attention resources, is partially accessible to consciousness and malleable. The cognitive system is configured such that stimuli continually trigger off circuits of automatic processing, but controlled processing is called when the system indicates a failure of performance or a situation involving novelty or personal importance. It is conceivable that abnormality in automatic or controlled processing could contribute to different degrees to the CAS such as selective focusing on threat or the persistence of worrying. For example, exposure to repeated traumas might sensitize processing assemblies for the initial detection of threat giving it an automatic nature. However, it seems this in itself would not explain the failure to disengage negative processing which is identified in the S-REF model as central to disorder. In the S-REF model sustained processing such as worry, rumination and threat monitoring is attributed to executive or strategic factors with metacognitions playing a key role.

Although both controlled and automatic processing are likely to operate in disorder (Matthews and Wells, 2000), evidence supporting the S-REF emphasis on strategic factors has grown. For example, Phaf and Kan's (2007) review concluded: "the emotional Stroop effect seems to rely more on a slow disengagement process than on a fast, automatic bias" (p. 184). This conclusion fits neatly with a central hypothesis of the S-REF that psychological disorder is linked with strategic factors that are the cause of perseverative or extended negative processing. It also fits with the impact of effective treatment strategies derived from the S-REF, such as the attention training technique(Wells, 1990), which demonstrably enhance self-reported attention flexibility (Nassif and Wells, 2014), objectively measured attention disengagement (Callinan et al., 2015), and neurophysiological markers of executive control (Knowles and Wells, 2018; Rosenbaum et al., 2018).

The S-REF model elucidates an advanced "architecture" of control that involves two sets of distinctions; one between automatic and controlled processing and the other between cognitive and metacognitive systems. The distinction between cognitive and metacognitive systems is supported not only by self-report as reviewed above but also by neuro-imaging data.

In particular, a meta-analysis of 193 functional neuroimaging studies of executive functioning tasks (i.e. flexibility, inhibition, working memory, initiation, planning, vigilance) in 2,832 healthy individuals demonstrated that these tasks share a super-ordinate network involving the pre-frontal, dorsal anterior cingulate, and parietal cortices (Niendam et al., 2012). Additionally, imaging of neural activity during cognitive tasks such as decision making suggests a neural system located in the pre-frontal cortex mainly involved in metacognition and independent of a cognitive system (Qiu et al., 2018).

It is evident from these parallel developments in metacognitive and neuropsychological research that a more detailed modeling of the metacognitive and cognitive architectures supporting self-regulatory processing is needed to advance the field. Such a model must explain the dynamic relationship between metacognition and cognition and the nature of the structures, circuits, and information involved in the perseveration or disengagement of negative processing.

In the remaining sections of this paper, I outline a model of a metacognitive control system of the S-REF specifying the nature and influences of metacognitive processes that contribute to the CAS and maladaptation. I then explore the implications of the model for metacognitive therapy and for future theory and research in the area.

# THE METACOGNITIVE CONTROL SYSTEM

The Metacognitive Control System Model (MCS) introduces novel concepts\* alongside those that already feature in the S-REF. In **Table 1** they are defined, and their functional characteristics are summarized to aid understanding.

A simplified schematic of the metacognitive control system (MCS) and its relationship with the cognitive system (CS) is depicted in **Figure 1**. Three overall sets of components are differentiated in the figure: (1) cognitive system (where automatic and on-line strategic processing are further distinguished), (2) metacognitive system, and (3) neural networks. It should be noted that this tri-partite separation simplifies the architecture and overlap and sharing of some structures and processes is expected. In particular, both cognitive and metacognitive processing are likely to consist of automatic and strategic processes but for simplicity this is not shown. The model is intended to represent features of standard architecture and processes for cognitive control, but as depicted the cognitive

TABLE 1 | Definitions and functional characteristics of constructs in the MCS model.


system (CS) is populated with the type of on-line processing (i.e. the CAS) that gives rise to psychological disorder.

The MCS is comprised of a comparator mechanism, metacognitive information in the form of declarative knowledge (D), procedural knowledge (P), and cybernetic code. There are also temporary memory registers. Different types of on-line processing are directed by the MCS, not just the style of extended negative processing that constitutes the CAS.

The function of the MCS is to monitor (M) and control (C) the activities of the cognitive system in pursuit of processing goals. It achieves this through direct and indirect effects involving the flow of information *via* the circuits depicted.

The cognitive system, shown in the left-hand side of **Figure 1**, is comprised of low-level automatic processing and on-line (strategic) processing that includes the limited capacity "thinking space." The output illustrated is labeled "psychological disorder" and is considered the consequence of the cognitive attentional syndrome (CAS) dominating on-line processing as depicted. Under different on-line processing configurations, where, for example, inhibition of worry under control of the MCS is specified, internal psychological events will be transitory and therefore not constitute "disorder."

Some features of metacognitive control are attentionally demanding and require conscious involvement and therefore draw on limited capacity processing which may compete with CS on-line processing. The operations of the MCS depend on temporary and longer-term memory stores, with some specialized memory structures (i.e. memory registers) among other dimensions (e.g. those involved in comparator function) likely to be specific to the MCS.

Centrally, the MCS continuously monitors and tests through the comparator mechanism the current state of processing in the CS against an internal model. The model represents a reference standard for the present and future/expected state of cognition. After a discrepancy or mismatch (error) is detected, instructions are issued to control mechanisms to bring CS processing in-line with goals. To accomplish this control function, it is hypothesized that the MCS has a capability to translate the current status (e.g. a discrepancy) into information; a *cybernetic code* that can be used to influence the behavior of cognitive and neural systems, biasing activity toward, for example, discrepancy reduction. It is therefore hypothesized that an important function of the MCS is generating, storing and using cybernetic information in the control of processing.

Code can influence processing across different neural networks that are recruited to bias the CS. For example, the code may be used to send commands to interoceptive networks leading to a "felt-sense" or "gut-feeling" that is recruited to bias or maintain a particular processing routine. As a means of illustration, consider an experience familiar to most people; the "tip-of the tongue" effect. When an item cannot currently be retrieved from memory (a discrepancy), this is accompanied by a strong somatic feeling and repetitive and sustained retrieval attempts that are often strategic but can also continue autonomously long after the individual has given up trying to remember. Thus, in this example, production of interoceptive responses and changes in arousal linked to receiving a signal of discrepancy (code), bias retrieval (perhaps a type of statedependency effect), maintain implementation of retrieval instructions and increase motivation for sustained strategic memory search.

Because the comparator is consistently transitioning to the next set of processes, the system must protect against the loss of earlier code when the goal of processing remains unmet. A solution is for code to be stored temporarily in *memory registers*. It is then available to the system for repeating processing sequences – *cybernetic looping* – in pursuit of goals. Cybernetic looping, or repetition of a set of processes, like in the example

of sustained memory search in the "tip-of-the tongue" experience is usually adaptive. Looping increases the probability of goal attainment (e.g. memory retrieval).

An important question relating to self-regulation concerns the determinant of number of repetitions of a cognitive process (i.e. adaptive perseveration) in an attempt to reach processing goals, especially when goals are unattainable. Several possible solutions to this issue need to be explored. It seems most probable that there are in-built system limits to iterations of processing, which may continue until neuronal or biological states (e.g. level of arousal) change. Plausibly, the memory registers holding cybernetic code may be temporary with decay being the norm. These proposed characteristics may be an important feature of psychological recovery or adaptation that naturally ensues over time. Nevertheless, this process could be adversely affected by dysfunctional metacognitive knowledge (e.g. "I must worry about all negative possibilities" or "I have lost control over thinking"). Under these influences choice of self-regulation strategy is dominated by the CAS (e.g. worry), which perpetuates processing and contributes to discrepancies (e.g. a sustained sense of threat).

This and other important implications emerge from the cybernetic code hypothesis. Under the direction of commands presented in procedural knowledge, cybernetic code could be used to control processing at different destinations in the neural network. For example, when specific commands activate or bias interoceptive processors it becomes *via*ble to "somatize" or feel the status of cognition. Feasibly, through this function the "sensing" of discrepancies and perhaps other mental processes can be implemented by the procedures of the MCS. In consequence, this allows for more complex internal representation and communication of the events occurring within the CS. A "sensing" of cognition may be a building block of the *embodiment of thinking* and a process likely to be important in the construction of self-awareness, to which I will return later.

As I have already proposed a range of memory structures are required to make *internal cybernetic communication* possible and are depicted as part of the MCS in **Figure 1**. There must be temporary storage (i.e. memory registers), long-term stores of metacognitive declarative (D-knowledge), and procedural (P-knowledge). While the memory registers act as a temporary buffer to protect against cybernetic code loss, the long-term memory stores provide metacognitive information and the instructions or commands for the model, the comparator process, and control of other neural systems.

# Embodiment and Self-Awareness

The theoretical structures and inter-relationships described above provide an architecture, set of functions, and feedback systems that could have several useful properties. They enable real-time information about cognitive activity to pass *via* monitoring into the MCS. In turn, under the commands of procedural knowledge, cybernetic code about cognition can be generated and influence processing in specific networks. Depending on the networks involved a combination of interoceptive (arousal), visual, or auditory processing activity linked to the code can arise. This raises the possibility that metacognitive commands (procedural knowledge) could specify that processing activity in particular networks is used as data (D in **Figure 1**) to create a context or *meta-representation* for the events in on-line processing. A system of such configuration could be directed by its procedural knowledge to compute in on-line processing a particular *meta-representation* consisting of a subjective stance in relation to cognition as objectifiable, separate from external events and within (i.e. tangible, felt, or embodied). Such a mechanism might provide a basis for states of objective meta-awareness (i.e. a "sense of cognition" e.g. a *feeling* that an item of knowledge is stored in memory). Furthermore, if procedural knowledge or system commands specify that objective meta-awareness (i.e. the "sense-ofcognition") is processed symbolically as "I" or "me" within on-line processing, objective *meta*-awareness is transformed into *self*-awareness. Thus, self-awareness as conceived may require as a building block a basic metacognitive system configuration within which the commands generate a sensorial response to cybernetic information which is subject to "on-line" (i.e. conscious) symbolic processing.

A propensity to experience meta-awareness, to objectify thoughts and memory and label the observer as "self " creates enablers and barriers to cognitive control. Self as a construction or context for cognition provides for greater flexibility and development of control because it permits cognition to become the object of focal attention and the subject of an individual's motivations and goals. For example, a person's explicit goals can be to improve problem solving, concentration or memory ability, or to become more optimistic. What is more, it means that the private content of cognition can be shared and modified through language or other forms of expression. Ironically, it also means that private cognition can be hijacked and underlying metacognitions corrupted by, for example religious and social systems that sanctify or punish the possession of certain thoughts and beliefs.

# TREATMENT IMPLICATIONS

The ideas developed in this paper are the basis of metacognitive therapy (MCT), which focuses on reducing the CAS and modifying metacognition so that recovery can occur. Full MCT treatment was first developed for generalized anxiety disorder (Wells, 1995, 1997) and subsequently other disorders (Wells, 2000, 2009). In meta-analyses, MCT demonstrates large treatment effects and appears potentially more effective or more efficient than cognitive behavioral approaches (Normann et al., 2014; Normann and Morina, 2018). In a direct test of transdiagnostic MCT against disorder-specific CBT across anxiety disorders, outcomes favoring MCT were reported (Johnson et al., 2017) and potential mechanisms of change could be distinguished (Johnson and Hoffart, 2018). Several trials have evaluated the effects of MCT against CBT for generalized anxiety. In each case MCT was superior (Van der Heiden et al., 2010; Wells et al., 2010; Nordahl et al., 2018). More naturalistic studies of less highly selected patients also support positive treatment effects of the full MCT package (e.g. Hagen et al., 2017; Papageorgiou et al., 2018; Callesen et al., 2019) and of individual treatment techniques (e.g. Knowles et al., 2016). The majority of treatment outcome studies have been conducted in anxiety and depression, but preliminary feasibility data suggest that the treatment can be implemented in psychosis (Morrison et al., 2014; Carter and Wells, 2018), transdiagnostic group settings (Capobianco et al., 2018c), comorbidity (Hjemdal et al., 2017), treatment resistant cases (Wells et al., 2012; Winter et al., 2019), alcohol abuse (Caselli et al., 2018), and traumatized borderline personality (Nordhal and Wells, 2019).

# Advanced Treatment Considerations

What is the impact of the MCS model for clinicians and researchers aiming to develop a better understanding of the mechanisms and processes of MCT and its effective practise?

A consequence of separating the cognitive system from the MCS in conceptualizing information processing is the following: worry, rumination, appraisals, and the execution of behaviors are all processes occurring within the cognitive system (CS). However, control, executive processes, knowledge supporting control and information on the current status of cognition are properties of the MCS. In psychological disorder it is chiefly the MCS that is the cause of bias observed in the cognitive system (CS). Maladaptation in the MCS is the major internal source of extended negative processing (the CAS) occurring in the CS. An implication of the distinction is that treatment should focus on formulating and modifying the content, strategies, and regulatory influence of the MCS as the most important source of disorder. Thus, treatment does not as a matter of emphasis focus on changing the properties of the CS such as the content of thoughts, general beliefs, memories or images or aim to change reflexive (automatic) networks of the CS through prolonged exposure techniques.

The conceptualization of procedural metacognition located in the MCS and its separation from cognition (the CS) presents an important implication concerning how treatment is conducted. It means that MCS knowledge; not only declarative but also the procedural commands that direct the comparator and bias the activities of CS must be extracted from the MCS and processed (e.g. modified) in the CS on-line before being returned to the MCS or sent to another location in the network. Crucially, this means that the appropriate parcel of procedural knowledge must be extracted; that which is the source of the CAS. Since the CAS can take a variety of forms the therapist must accurately identify it on a case by case basis. Furthermore, excessive CAS activity in the CS must be moderated early in therapy, so that the limited capacity "thinking space" can be liberated and used for MCS modification.

Metacognitive therapy contains techniques designed for the above purpose that explicitly induce and "hold" the patient in a "metacognitive mode" of processing during sessions with the aim to modify both declarative and procedural meta-knowledge while governing CS processing load. These techniques include among others: meta-level discourse, the attention training technique, the free-association and tiger tasks, rumination postponement, metacognitive focused exposure, metacognitive experiments, and worry-modulation procedures. The therapist must use direct *metacognitive experiences* and a *discourse* that transforms processing styles in the CS before reassigning the knowledge supporting them to the MCS. In this manner, the techniques used increase the range, choices, and flexibility with which the individual controls and can relate to their CS. These techniques are described in detail elsewhere (Wells, 2005b, 2009).

The model highlights clear differences between metacognitive therapy and other treatment approaches in the intended target of change. In MCT, the therapist retrieves and modifies the validity of declarative metacognitions and also retrieves and re-writes the commands (procedures) for regulating processing with the purpose of modifying those involved in the CAS. In contrast, other treatments either do not aim to work on metacognitions or they do so without maintaining a clear structural and functional distinction between systems. But such a distinction could be facilitative in the design of more advanced theory-grounded treatment techniques. For example, if we consider the treatment of low self-esteem, a cognitive therapist will aim to identify and challenge negative beliefs about the self by asking questions such as: "What is the evidence you are a failure, is there another way to view the situation?" but the metacognitive therapist would ask: "What's the point in analyzing your failures?" and follows with techniques that allow the individual to directly step-back and abandon the perseverative thought processes that extend the idea. Of particular importance, in MCT, the client discovers that processing remains malleable and subject to control in spite of the dominant cognition (belief) "I'm a failure," thus creating an alternative model of processing rather than an alternative model of the social self (the latter considered a secondary topographic event).

Good metacognitive therapy, the model suggests, is that which modifies the procedural knowledge base. It should enable the individual to: (1) directly alter the relationship or "stance" they have with products of cognition; (2) directly manipulate the control of cognition (e.g. delay worry and inhibit perseverative thinking); and (3) separate metacognition (i.e. mechanisms of control) from the strong influence of internal (e.g. thoughts and feelings) and external events (as per Attention Training Technique protocol). The systematic regulation of attention using a framework of discovery that shows attention remains flexible irrespective of mental events supports the development of generalpurpose strong metacognitive control procedures of this kind.

An implication of the MCS as described is that it can (under commands of procedural knowledge) initiate and hold in the moment different *meta-representations* of internal cognition. A meta-representation is influenced by the effect of the current cybernetic code on other processors that provide input to on-line processing. This creates flexibility and the possibility of choosing how to relate spatially and sensorially (or emotionally) to inner thoughts, memories and mental events. In *object mode,* thoughts are experienced as direct perceptions and treated as facts (the individual is in the thought), but in *metacognitive mode,* they are experienced as events or stimuli in the mind and the individual steps outside of them (Wells and Matthews, 1994). The model directs us toward developing techniques that change the *meta-representational state*. For example practise of "flipping" between modes or of co-joint experiencing of incongruent thoughts (e.g. negative thought plus positive memory) or of experiencing a negative thought and coupling it with a positive feeling. In each case the meta-representation might be changed by shifting "stance" or coupling cybernetic code with new and incongruous bodily and affective states.

Since a goal of MCT is to reduce over-reliance on thinking, it is usually better to shift into a metacognitive mode and disengage further conceptual processing rather than analyze and interrogate negative thoughts as a means of change. However, the model suggests that an exception must occur when a negative metacognitive appraisal or meta-belief is present (e.g. "Worrying will cause cancer"). Since this is primarily a property of the MCS (it reflects maladaptive metacognitive knowledge), it should be evaluated and replaced with more adaptive information because it will continue to impact on cognitive control and the stance in relation to cognition. To summarize, in metacognitive therapy challenging of the validity of metacognitions is supported, but challenging the validity of cognitions is not.

## Metacognitive Focused Exposure

Simply engaging the CS in activities of cognitive-behavior therapy such as evaluating the validity of thoughts or repeated exposure to fear stimuli present imprecise and coincidental ways of modifying the control system. Exposure is considered to facilitate habituation or "emotional processing," which is defined as: "a process whereby emotional disturbances are absorbed and decline to the extent that other experiences and behavior can proceed without disruption" (Rachman, 1980, p. 51). This has typically been viewed as a mechanism whereby information about declining arousal is automatically incorporated in fear networks (e.g. Foa and Kozak, 1986) such that pre-existing links between stimulus-response nodes and negative meanings attached to anxiety are weakened. This conception of emotional processing relates most closely to automatic processing and neglects the involvement of upper-level cognitive structures, including the metacognitive control system. For example, it is possible to think about an emotional event in an unemotional way. Furthermore, the network approach does not address questions concerning the factors that determine the cessation of emotional processing or how the goals of emotional processing are represented and monitored?

The MCS model invites the clinician to concentrate treatment on top-down influences on extended processing such as the use of worry, over-analysis of memory or threat-monitoring that lead to repeated or sustained activation of fear networks. The MCS model also implies that emotion networks may respond to cybernetic code and the impact of code on the network may be moderated by metacognitive knowledge. For instance, the ability to think about an emotional event in an un-emotive way is resolved, because the MCS can change the nature of the relationship (meta-representation) with thoughts. In addition, theoretical questions about the cessation and representation of the goals of emotional processing are dealt with by hypothesizing that the MCS can monitor and control emotional networks partly through its comparator and cybernetic code functions. Emotional processing stops when the goal of processing is met or when the cybernetic code decays. The ability to achieve such exit signals is potentially reduced by the CAS and dysfunctional metacognitions, leading to psychological maladaption.

There are implications of the model for developing more efficient and effective exposure therapy techniques. This can be achieved by inhibiting the CAS during exposure and by configuring exposure to explicitly modify maladaptive metacognitive knowledge; both declarative and procedural. Such an approach of *metacognitively focused exposure* has been previously introduced (Wells, 2000).

In a simple form, the combination of exposure with attention instructions designed to reduce threat monitoring and increase access to non-threat related information will be helpful. But more unexpected applications are indicated. For instance, the MCS model presents an idea that runs counter to the traditional approach to exposure treatments that emphasize the need to eliminate avoidance. If we take as an example the treatment of obsessive-compulsive disorder, exposure and prevention of covert and overt rituals (forms of avoidance) such as repeated washing is an effective and recommended treatment. In contrast to this approach, in MCT, the patient can be permitted to use rituals in response to thoughts provided they hold the thought in mind, because the goal is to change the metarepresentation of the thought in the MCS and not the associative links at a fear network level through habituation. The aim in MCT is to change the nature of the person's relationship with negative cognitions so that *thoughts are experienced* as unimportant and transient events in the mind.

A small number of pilot studies have experimented with forms of metacognitive focused exposure. Fisher and Wells (2005) examined the effects of brief exposure when it was presented as an experiment to explicitly test metacognitive beliefs in OCD. In this study, patients with OCD were asked to listen for 5 min to their obsessional thoughts recorded on a loop-tape under two contrasting conditions. In one condition, a habituation instruction was used with the goal of staying with the feelings of anxiety and stopping any rituals. In the metacognitive condition, the instruction was also to stop any rituals but with the goal of discovering that the thoughts were unimportant. While both rationales were seen as equally credible by participants, the metacognitive condition was associated with significantly greater reductions in anxiety, metacognitive beliefs and urge to neutralize. In another study, Wells and Papageorgiou (1998) exposed social phobia patients to feared social situations under a habituation rationale or external attention focusing rational that counteracted threat monitoring. The latter condition produced superior effects after a single brief exposure.

# Resistance to Change

The present model offers a means of understanding and dealing with resistance to change in psychotherapy. It implies that metacognition can act against a person "changing their mind." The model draws the clinician to the paradoxes in cognitive control such as holding both positive and negative metacognitive beliefs concerning sustained processing. In generalized anxiety disorder (GAD), the client believes that worrying will help anticipate and avoid threat but in conjunction with this there is the belief that worrying is uncontrollable and harmful (Wells and Carter, 2001). In health anxiety, there is a belief that negative misinterpretation of symptoms will facilitate illness detection and also that thoughts can cause illness (Bailey and Wells, 2015a). In depression that analyzing why one feels depressed will lead to feeling better but might also cause selfharm (Papageorgiou and Wells, 2001, 2003). Each of these examples presents potential ambivalence, uncertainty, or vacillation in abandoning the CAS. A belief in the uncontrollability or pure "biological basis" of negative cognition contributes to a sense of hopelessness, reduced effort invested in control or a reliance on extraneous forms of control. This acts against the client using their own internal control, which might otherwise enhance MCS capacity to create change.

We have seen how a proposed normal in-built mechanism; cybernetic looping, contributes to perseveration of processing. This could explain persistent but relatively normal affective and motivational states such as longing, desire, grief, craving, anger, regret, shame, and remorse among others. In these instances and in stress and adjustment reactions, we would expect spontaneous recovery over time. However, when an individual uses the CAS as a coping strategy it maintains the sense of threat and disrupts the normal exit conditions for the cybernetic loop, leading the individual to become "gripped" by their feelings. Furthermore, worrying and ruminating consume processing resources that are required for metacognitive control such as switching between goals for processing, consequently negative processing is less flexible and persists. In each of these cases, the treatment aim should be to remove the barriers (i.e. CAS) to exit and effective internal control conditions. Usually, perseverative processes appear to have an in-built limited and system determined repetition that we might conceptualize as a normal psychological recovery period. This concept is used in treating post-traumatic stress disorder, where the explicit goal shared with clients in MCT is to remove the CAS so that in-built *reflexive adaptation processes* run their natural course (Wells, 2009; Wells and Colbear, 2012; Wells et al., 2015). An important implication is that restructuring thoughts about trauma, modifying trauma memory and reliving methods are not necessary for effective treatment. Treatment should only be introduced after recovery processes have been given an opportunity to run naturally.

Cognition is not supplied with a user manual or a schematic that allows the owner to understand how it works or how best to operate it. However, we rely on information and procedures (knowledge) of how our memory and attention works, we learn to compensate for tiredness or a noisy environment by increasing effort or concentration, we learn what a thought is, what a dream is, that we have a good memory for places, and that cognition is harmless and not prone to loss of control. We might reasonably assume that metacognitive knowledge about cognitive control has a special place and powerful influence on how we construe our own experiences and how much we allow our own mental events to impact and shape our lives. The impact can be profound. For instance, consider how some approaches to mental illness might contribute to a disabling and unhelpful knowledge of metacognitive control that solidifies a sense of helplessness and mental brokenness. This is not very useful to the individual, but the discovery of control and a belief that recovery is a matter of letting some thoughts go is likely to be more beneficial. More broadly, the MCS model encourages us to examine the messages carried by existing approaches to mental health diagnosis and treatment. Treatment delivery programs should ensure that unhelpful metacognitions are not created but those that already exist are modified.

# The Process of Recovery

Implicit in all that I have described above is a fundamental idea. The MCS is involved in the perpetuation of negative psychological experiences, and it is also involved in their cessation; it plays a role in recovery. Under typical circumstances, we might consider the cybernetic code functions as a "code for recovery" because it supports continued processing toward goal attainment and any repetition of processing is usually limited. However, when metacognitions specify the CAS and when they give rise to a sense of uncontrollability and threat from cognition itself, errors or deviations from reference internal states persist and the code is constantly refreshed. The process of recovery in psychological therapies is one in which decay of the code and exit conditions for cybernetic looping are made accessible. In MCT, this is achieved through modifying maladaptive metacognitive knowledge, by enhancing flexible control and by disengaging the coping strategies that depend on extended processing.

# LIMITATIONS AND FUTURE RESEARCH

It must be borne in mind that the model is rudimentary and a project in development. For example, in the interests of simplicity I have shown "automatic processing" as a separate cell in **Figure 1**. However, a dichotomy between automatic and controlled processing is simplistic, and it may be better to view processing along a continuum of automaticity across multiple systems. Some automatic processes in the CS may prime specific procedural knowledge within the MCS, so the CS has some limited influence over the MCS, which is not explored. The CS is controlled by its own "hard-wiring" and in a more flexible and extended way by the procedural knowledge and codes of the MCS. The processes of the MCS, such as activities of the comparator and the priming of procedural knowledge are unconscious and the processes reflexively "run-off" in response to stimuli.

Unanswered questions surface concerning the reliance of both metacognition and cognition on shared and domainspecific structures and processes, among them memory. In particular, depiction of the memory registers is not intended to imply that these are structurally equivalent to long-term memory or working memory. Instead, the model points to the importance of exploring and separating multiple components of memory including the hypothesized memory registers and processes that temporarily represent discrepancies in processing. The prediction that activity in such structures and related processes is moderated by cybernetic code offers a potential means to distinguish them from other memory processes using paradigms that induce code (i.e. cause discrepancies such as violations of expectancy and induction of performance errors).

There are clear limitations in the current database, including a paucity of information concerning the antecedents of dysfunctional metacognitive knowledge, such as the possible role of stressful early life experiences (e.g. Myers and Wells, 2015). Furthermore, while preliminary evidence suggests that different components of metacognitive knowledge may interact in explaining distress, this remains to be explored in detail. For instance, interaction between knowledge about attention and beliefs about uncontrollability of thoughts appears to provide additional nuanced effects (at least in children) that may prove important (e.g. Reinholdt-Dunne et al., 2019).

So far in this account I have intentionally avoided any detailed consideration of the detrimental effects of metacognition on performance of cognitive tasks. The detrimental effects of anxiety on performance are well established (e.g. Eysenck, 1992). Anxious mood appears to be a stronger determinant of impaired performance than trait-anxiety, with worry predicting poorer performance better than emotional and physiological aspects of anxiety (e.g. Morris et al., 1981). Eysenck and Calvo (1992) proposed that anxiety impairs the efficiency of the central executive which appears much like working memory as proposed by Baddeley (1986). Their theory assumed that task-irrelevant processing such as worry does not always have a negative impact on the effectiveness of performance. Finding oneself worrying may in fact enhance motivation to overcome the negative performance effects by using additional processing resources. This appears to be at odds with the idea of a CAS that causes problems. However, it remains consistent with the MCS model because the ability to compensate will depend on characteristics of the MCS. In particular, metacognitive beliefs of lack of control should negatively influence the level of compensatory resources used. For example, in a study by Matthews et al. (2019), the effects of high worry on performance and neurophysiology under social-evaluative stress was dependent on the level of meta-worry (i.e. negative appraisals of the uncontrollability and danger of worrying).

It remains to be determined how the MCS might relate to a wider range of executive functions, to concepts such as working memory (Baddeley, 1986, 1996) and inhibition and attention shifting functions hypothesized by Eysenck et al. (2007) in attention control theory. But the model points to the importance of examining the influence of metacognitions on these dimensions.

While there is strong evidence of dysfunctional metacognitive knowledge across psychopathologies, most of the evidence is at the level of self-report. Self-report can be criticized, but it is a mistake to dismiss it as it provides important clues to the consciously accessible aspects of information processing such as goals and choice of strategy. But this area of research needs to be strengthened by investigating further the effect of self-report metacognitions on attentional responses at a performance and neural level. Such efforts should seek to explore the cybernetic code hypothesis and map the neural structures, circuits and dynamic effects involved. Usefully, the MCS model suggests the development of laboratory paradigms to probe and isolate such effects by using the induction of discrepancies between actual and desired processing states, such as violating cognitive expectancies. If a trace of the cybernetic code in such paradigms can be detected in the form of activity or temporary change at a cellular or network level this might be used as proof. It may be possible to adapt this, using speed of decay of such activity produced in discrepancy induction paradigms to measure inherent psychological resilience. For example, greater resilience might be associated with faster loss of the cybernetic code from memory registers.

Finally, the model presents important questions and research directions concerning childhood development of the MCS; when and what are the influences on the development of beliefs about inner-thought? Is there a sequence of development of attention control skills and is there an optimal set pattern? We might hypothesize that it is possible to identify *protometacognitive* states and stages that track the transition from early attention fixation and limited control through to acquired attention flexibility and the later development of higher-order knowledge of control necessary in consolidating a MCS. Exploration of levels of complexity and degree of interconnectedness of the CS and MCS presents major trajectories for future cognitive and neuropsychological research.

# CONCLUSION

The S-REF model has influenced research on cognitive control in psychological disorder, placed top-down processes and metacognition in a prominent role and informed the development of metacognitive and other therapies. But an important challenge remains: to strengthen the theoretical foundations necessary to advance the study of metacognition in self-awareness and mental health. One means is by exploring and describing in detail the components, architecture and functions of the metacognitive control system of the S-REF and how it relates to disorder; my goal in this paper. In particular, the field can benefit from consideration of the types and effects of metacognitive information generated and used by the system in pursuit of cognitive regulation. This has become more justified as evidence from neuropsychological and S-REF based research supports a neural system separate from cognition and involved in metacognition as the S-REF predicted.

Psychological disorder from the position of the S-REF model is conceptualized as a state of persistence of negative processing that is difficult to control. In most cases, negative ideas and feelings are transitory but in psychologically vulnerable individuals they become extended and "fixed" due to a transdiagnostic style of thinking: Cognitive Attentional Syndrome (CAS). The CAS is largely a consequence of the impact of biased metacognitions on cognitive regulation. Persistence of processing is influenced by different features of the MCS; repetition of processing is normally a feature of cybernetic looping when discrepancies or errors are detected. But in psychological disorder this effect is disrupted by choice of strategies linked to metacognitive knowledge that interfere with exit conditions for looping, diminish inhibitory control attempts (e.g. "I have lost control of my thoughts") or sanction extended processing (e.g. "I must analyze all my failures until I become a success").

An architecture replete with metacognitive information (i.e. declarative and procedural knowledge, mental models, cybernetic code and metacognitive experiences) has emergent properties that contribute to cognitive control. It is a framework for the development through meta-representational states of within-ness (embodiment), self-awareness, and a subjective ownership of cognition. Such effects normally increase flexibility, a sense of stability, and self-control of thoughts. They also facilitate the social communication of thought, but they can as described present a wider range of potential loci for bias that contributes to disorder. At the most basic of applied levels, health systems and clinicians working with service users must begin to consider the potential negative effects on metacognition of the information and treatment techniques they provide.

In the future, it may be possible to describe the proposed psychological structures and processes with greater precision. But for now the model points to the potential in isolating a discrete metacognitive control system that is separate from cognition, studying the impact of its components and content on psychopathology, self-awareness, and self-regulation. I have described how strengthening this separation can continue to provide a basis for theoretically derived treatment techniques in MCT that target specific causal mechanisms in a particular way. The MCS model opens up a substantial set of new avenues for research addressing issues that include: mapping the role of different neural systems in cognitive control; testing the effects of discrepancies or violations of expectancies (i.e. production of cybernetic code) on interactions between systems; testing the co-dependence of metacognitive and cognitive operations on limited capacity; examining the multiple memory requirements and processes of metacognition; testing the interactive effects of metacognitive knowledge and attention control on symptoms; exploring the relationship between metacognition and selfawareness; and in a broad context examining untoward effects of healthcare delivery and social systems on metacognitive functioning. It provides a framework for a more unified cognitive, social and neurobiological theory of awareness, self-regulation and mental wellbeing.

Advances in psychotherapy require a paradigm shift; stronger information processing theory that can successfully explain the control of cognition and the negative subjective changes in perceived control and sense of self that are central features of disorder. Psychological wellbeing is not a matter of what we think. It is an issue of how we regulate the cognitive processes that prioritize and extend thoughts. It is the stance

# REFERENCES


taken in relation to the content of the limited capacity "thinking space." It is above all, the nature and effect of *metacognitive information* generated, held and used by processing systems.

# AUTHOR CONTRIBUTIONS

The author confirms being the sole contributor of this work and has approved it for publication.

patient sample: a randomised feasibility trial. *Psychiatry Res.* 259, 554–561. doi: 10.1016/j.psychres.2017.11.045


meta-cognitive model. *Clin. Psychol. Psychother.* 11, 137–144. doi: 10.1002/ cpp.402


Leahy, R. L. (2015). *Emotional schema therapy*. New York: Guilford Press.


depressive disorder: a pilot outpatient study. *Front. Psychol.* 10:1714. doi: 10.3389/fpsyg.2019.01714

Yilmaz, A. E., Gencoz, T., and Wells, A. (2011). The temporal precedence of metacognitions in the development of anxiety and depression symptoms in the context of life-stress: a prospective study. *J. Anxiety Disord.* 25, 389–396. doi: 10.1016/j.janxdis.2010.11.001

**Conflict of Interest:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer GC declared a past co-authorship with the author AW to the handling editor.

*Copyright © 2019 Wells. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# What Lies Beneath Trait-Anxiety? Testing the Self-Regulatory Executive Function Model of Vulnerability

Henrik Nordahl1,2 \*, Odin Hjemdal<sup>1</sup> , Roger Hagen<sup>1</sup> , Hans M. Nordahl2,3 and Adrian Wells4,5

<sup>1</sup> Department of Psychology, Norwegian University of Science and Technology, Trondheim, Norway, <sup>2</sup> Nidaros District Psychiatric Center, St. Olav's University Hospital, Trondheim, Norway, <sup>3</sup> Department of Mental Health, Norwegian University of Science and Technology, Trondheim, Norway, <sup>4</sup> Division of Clinical and Health Psychology, School of Health Sciences, Faculty of Biology, Medicine and Health, Manchester Academic Health Science Centre, The University of Manchester, Manchester, United Kingdom, <sup>5</sup> Greater Manchester Mental Health NHS Foundation Trust, Prestwich, United Kingdom

#### Edited by:

Changiz Mohiyeddini, Northeastern University, United States

#### Reviewed by:

Deana Davalos, Colorado State University, United States Juan Ramos-Cejudo, Complutense University of Madrid, Spain

> \*Correspondence: Henrik Nordahl henrik.nordahl@ntnu.no

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 09 October 2018 Accepted: 14 January 2019 Published: 30 January 2019

#### Citation:

Nordahl H, Hjemdal O, Hagen R, Nordahl HM and Wells A (2019) What Lies Beneath Trait-Anxiety? Testing the Self-Regulatory Executive Function Model of Vulnerability. Front. Psychol. 10:122. doi: 10.3389/fpsyg.2019.00122 Vulnerability to psychological disorder can be assessed with constructs such as trait anxiety and neuroticism which among others are transdiagnostic risk factors. However, trait-anxiety and related concepts have been criticised because they don't illuminate the etiological mechanisms of psychopathology. In contrast, the metacognitive (S-REF) model offers a framework in which metacognitive knowledge conceptualised in trait terms is part of a core mechanism underlying trait-anxiety and related constructs. The present study therefore set out to explore metacognitions as potential underlying factors in trait-anxiety (the propensity to depression and anxiety). Nine hundred and eighty two participants completed self-report measures of metacognitions and traitanxiety at time 1, and 425 individuals completed the same measures 8 weeks later. At the cross-sectional level, metacognitions accounted for 83% of the variance in anxiety- and 64% of depression propensity. Furthermore, despite both domains of traitanxiety showing high stability over time, negative- and positive metacognitive beliefs were significant prospective predictors of both domains of vulnerability. These findings suggests that metacognitive beliefs may be an underlying mechanism of vulnerability attributed to trait-anxiety with the implication that the metacognitive (S-REF) model informs conceptualization of psychological vulnerability, and that metacognitive therapy applications might be employed to enhance psychological resilience.

Keywords: metacognitive beliefs, trait-anxiety, risk factors, anxiety, depression, resilience

# INTRODUCTION

Founded in personality research, the concept of psychological vulnerability can be assessed by a variety of trait constructs such as Trait-Anxiety, Neuroticism and negative affectivity (Eysenck and Eysenck, 1975; Spielberger et al., 1983; Watson and Clark, 1984). These constructs are positively linked with psychopathology and are considered to be a general tendency to experience negative emotions that is genetically influenced (e.g., Rosenström et al., 2018). They are reliably associated

with psychological disorders (Clark and Watson, 1991; Brown et al., 1998; Kotov et al., 2010; Mincic, 2015), broader aspects of physical health and illness, subjective well-being, relationship satisfaction, and social and occupational impairment (Lengel et al., 2016). It has been argued that trait theory has been underutilised in clinical settings (e.g., Barlow et al., 2014; Lengel et al., 2016; Watson et al., 2016), and that formulating and targeting traits such as negative affectivity could potentially advance our understanding of psychopathology (Sauer-Zavala et al., 2017).

One of the most frequently used measures of negative affectivity in psychological research is the State-Trait Anxiety Inventory (STAI: Spielberger et al., 1983). The STAI was designed by Spielberger et al. (1983) to measure anxiety as a state at a given point in time (state anxiety), and as a trait reflecting proneness to react with anxiety under stressful circumstances (trait anxiety). Trait anxiety is a dimension along which people vary, and can be invoked to explain individual differences in the frequency, intensity, and duration of episodes of state anxiety and negative affect. More recent studies employing factor analyses have suggested that the STAI-T consist of two interrelated factors and that its items measure propensity to both anxiety and depression (Bieling et al., 1998; Grös et al., 2007; Bados et al., 2010; Balsamo et al., 2013). Hence, rather than being considered a measure of specific proneness to anxiety as originally proposed, trait-anxiety should be considered a measure of general vulnerability to emotional disorder and distress.

Although the trait-anxiety construct has proven useful in the assessment of vulnerability and prediction of emotion disorder symptoms, critics have argued that personality dispositions such as negative affectivity or trait-anxiety do not yield useful information on the etiological mechanisms of psychopathology (Claridge and Davis, 2001; Ormel et al., 2004). Furthermore, the mechanisms underlying them must be elucidated in conceptualising these traits as central vulnerability factors (see e.g., Cuijpers et al., 2010; Ormel et al., 2013). One possibility is that there is overlap in vulnerability to both anxiety and depression and related constructs such as negative affect and these might be related to some common set of underlying psychological processes.

In the Self-Regulatory Executive Function (S-REF) model, Wells and Matthews (1994) argue that the differences between disorders are less important than the similarities, and that underlying transdiagnostic mechanisms of distress rather than topographical differences should become a greater focus in psychopathology research. In this approach, emotional disorders are viewed as caused by a common negative and perseverative thinking style, called the cognitive attentional syndrome (CAS; Wells, 2009). The CAS consist of worry and rumination, threat monitoring and maladaptive coping strategies that impair selfregulation. Furthermore, the CAS is regulated by underlying metacognitive beliefs conceptualised in trait terms, which includes knowledge about thinking, memory and attention (Wells and Matthews, 1994). Thus, metacognitive knowledge (i.e., metacognitive beliefs) are formulated as a central factor in both state and trait emotion, and might therefore be a core underlying mechanism in trait-anxiety and related constructs. For example, negative metacognitive beliefs about the uncontrollability and danger of worry in particular are likely to predict depression and anxiety proneness by contributing to reduced investment in controlling thinking and also to negative interpretations of internal experience, compromising choice of effective coping strategies when exposed to stress (Wells and Matthews, 1994).

Based on the S-REF model, there are two main measures which have been developed to assess generic metacognitive beliefs: the Metacognitions Questionnaire (MCQ; Cartwright-Hatton and Wells, 1997) and a briefer version, the Metacognitions Questionnaire-30 (Wells and Cartwright-Hatton, 2004). These trait measures consists of five factors assessing positive beliefs about worry, negative beliefs about uncontrollability and danger of worry, confidence in memory/attention, beliefs about the need to control thoughts, and cognitive self-consciousness. The five factor structure has been reported as reliable (Spada et al., 2008) and can account for individual variance in distress beyond a general "metacognition" factor (Fergus and Bardeen, 2017).

In line with predictions of the metacognitive model, metacognitive beliefs are demonstrated to be reliably associated with state measures of anxiety and depression (see Sun et al., 2017 for a review). In addition, significant positive correlations have been reported between metacognitive beliefs and trait-anxiety (Cartwright-Hatton and Wells, 1997; Wells and Cartwright-Hatton, 2004). One study has shown that metacognitive beliefs positively predicted trait-anxiety when controlling for the presence of a diagnosed mental disorder (Nordahl and Wells, 2017). Among domains of metacognitive beliefs, negative metacognitive beliefs have consistently shown the strongest association with trait-anxiety. However, to our knowledge no study has tested the structural relations between each domain of metacognitive belief and the two domains of trait-anxiety or explored these relations over time.

The aim of the current study was therefore to explore the association between the different domains of metacognitive beliefs and domains of trait-anxiety using both a cross-sectional and longitudinal data-set. To evaluate the structural relationship of these variables and test the overall fit of models, we employed structural equation modelling. Derived from the S-REF model (Wells and Matthews, 1994), our hypotheses were as follows; (1) metacognitive beliefs will be positively correlated with the STAI-T depression and anxiety factors; (2) metacognitive beliefs will explain substantial variance in both STAI-T factors; (3) metacognitive beliefs will account for variance in STAI-T factors over time; and (4) negative metacognitive beliefs will be the strongest independent predictor of both the STAI-T factors in the cross-sectional- and in the longitudinal data.

# MATERIALS AND METHODS

# Participants and Procedure

The present study was based on an online self-report survey of psychological distress with two measuring points. The survey was conducted in Norway and was approved by the Regional

Committee for Medical and Health Research Ethics (REC; reference: REK-Midt, 2016/705). Participants were invited to participate through advertisement on social media (Facebook), and were offered participation in a lottery to win an I-pad if they completed the survey at both time points. Several Norwegian voluntary organisations for mental health assisted in distributing information about the survey. Thus, participants were gathered at convenience, but had to be 18 years old or above, and had to able to read Norwegian. The survey was conducted using a programme called "Select Survey," provided by the first author's faculty at the Norwegian University of Science and Technology. Upon entering the survey portal, participants were presented with an information sheet that was approved by REC and were informed that proceeding to the main survey would be regarded as a signed informed consent. Nine hundred and eighty two individuals completed the metacognitions questionnaire 30 (MCQ-30; Wells and Cartwright-Hatton, 2004) and the State-Trait Anxiety Inventory; Trait version (STAI-T; Spielberger et al., 1983) at time 1 (T1), and four hundred and twenty five also completed the same measures at time 2 (T2), 8 weeks after the first round of questionnaires. The sample characteristics for the cross-sectional- and the longitudinal sample are presented in **Table 1**.

# Measures

## The Metacognitions Questionnaire 30

The MCQ-30 (Wells and Cartwright-Hatton, 2004) is a 30-item self-report scale measuring beliefs about thinking (i.e., metacognitive beliefs). Each item are scored on a four-point scale ranging from 1 (do not agree) to 4 (agree very much), and each subscale has a range from 6 to 24 points. High scores reflect more reported problems with the construct in question. A five-factor structure exists: (1) positive beliefs about worry (e.g., "I need to worry in order to stay organised"); (2) negative beliefs about the uncontrollability and corresponding danger of worry

TABLE 1 | Sample characteristics in the cross-sectional- and the longitudinal data sets.


(e.g., "my worrying thoughts persists, no matter how I try to stop them"); (3) cognitive confidence (e.g., "I do not trust my memory"); (4) beliefs about need to control thoughts (e.g., "I will be punished for thinking certain thoughts"); and (5) cognitive self-consciousness (e.g., "I am constantly aware of my thinking"). The measure has shown good internal consistency with Cronbach's alpha ranging from 0.72 to 0.93 (Wells and Cartwright-Hatton, 2004) and has been validated in Norwegian samples (e.g., Grøtte et al., 2016). In the current study, the internal consistency was good (positive beliefs: α = 0.85, negative beliefs: α = 0.85, cognitive confidence: α = 0.88, need for control: α = 0.81, cognitive self-consciousness: α = 0.79).

### The State-Trait Anxiety Inventory – Trait Scale

The State-Trait Anxiety Inventory (trait version: form Y2) (STAI-T: Spielberger et al., 1983) is a 20 item self-report questionnaire of general distress proneness, and has been validated in Norwegian samples (e.g., Haseth et al., 1990). Each item is rated on a four-point Likert scale. Total scores range from 20 to 80 points, with higher scores reflecting stronger traits of general distress proneness. The STAI-T has good psychometric properties, with Cronbach's alpha in the range of 0.86 to 0.95, and test-retest correlations ranging from 0.73 to 0.86 (Spielberger et al., 1983). Further psychometric evaluation of the STAI-T has shown that it consists of two factors: (1) depression (e.g., "I feel like a failure"); and (2) anxiety (e.g., "I feel nervous and restless"). The depression factor consist of 13 items (item number; 1, 3–7, 10, 12–16, 19), while the anxiety factor consist of 7 items (item number; 2, 8–9, 11, 17–18, 20) (Bieling et al., 1998; Bados et al., 2010; Balsamo et al., 2013). The depression score ranges from 13 to 52 points, while the anxiety score ranges from 7 to 28 points. In the current study, the internal consistency was excellent (α = 0.96) for the total scale, and for the subscales; depression, α = 0.95; anxiety, α = 0.90.

# Statistical Analyses

Confirmatory factor analysis (CFA) was used to evaluate the factor structure of the proposed five-factor model of the MCQ-30 and the two-factor structure of the STAI-T. No secondary loadings were modelled, but the factors were allowed to inter-correlate. Bivariate correlations were used to explore the association between the MCQ-30- and the STAI-T subscales. Structural equation modelling was employed to evaluate the fit of an overall model were the MCQ-30 factors were used as predictors of the STAI-T factors in cross-sectional datasets. Three commonly recommended fit statistics were used to evaluate the models (Hu and Bentler, 1999; Kline, 2011; Brown, 2015); the comparative fit index (CFI), the standardised root mean square residual (SRMR) and root mean square error of approximation (RMSEA). The CFI should be above 0.90 to represent an adequate fit, the SRMR should be less than 0.08, and the RMSEA should be below or close to 0.06 and the upper limit of the 90% RMSEA confidence interval should not exceed 0.10. Finally, multiple hierarchical linear regression analyses were used to explore the prospective relationships between the MCQ-30 subscales and the STAI-T subscales.

# Factorial Structure of the MCQ-30 and the STAI-T

Initially we tested the 5 factor model of the MCQ-30 and the 2 factor model of the STAI-T using confirmatory factor analysis. In the T1 data, the MCQ-30 five factor measurement model showed the following fit indices: χ 2 (395) = 1622.05, p < 0.01, CFI = 0.90, SRMR = 0.07, RMSEA = 0.06 (90% CI = 0.05, −0.06), and in the T2 data, the fit indices were: χ 2 (395) = 1245.85, p < 0.01, CFI = 0.89, SRMR = 0.07, RMSEA = 0.07 (90% CI = 0.07, −0.08). The STAI-T two factor measurement model showed the following fit indices in the T1 data: χ 2 (169) = 961.63, p < 0.01, CFI = 0.93, SRMR = 0.04, RMSEA = 0.07 (90% CI = 0.07, −0.07), and χ 2 (169) = 714.12, p < 0.01, CFI = 0.92, SRMR = 0.05, RMSEA = 0.09 (90% CI = 0.08, −0.09) in the T2 data. Globally, these fit indices indicate an acceptable fit of the MCQ-30 five factor model and the STAI-T two factor model in this sample at T1 and at T2. Thus, we considered it acceptable to proceed with the planned analysis involving testing of relationships between multi-factorial constructs.

# Descriptive Statistics and Correlations Between Factors

As a first step, before testing predictive models, we ran correlational analyses to examine the basic pattern of relationships between domains of metacognitive beliefs and domains of traitanxiety (i.e., depression and anxiety) in the data from T1. Descriptive statistics and bivariate correlations between measures are presented in **Table 2**. All of the correlations were positive and significant at the 0.01 level. STAI-T depression and anxiety were strongly correlated with each other, and showed the strongest correlation with negative metacognitive beliefs among the MCQ-30 subscales.

# Cross-Sectional Relationships Between MCQ-30 Factors and Depression- and Anxiety Proneness

To explore if MCQ-30 factors would statistically predict depression and anxiety proneness we used structural equation modelling (e.g., Kline, 2011). The two trait-anxiety factors, depression and anxiety, were used as latent dependent variables indirectly measured by their respective items (reported in the methods section). The five MCQ-30 factors were defined as predictor variables measured by their respective six items per factor.

The hypothesised structural equation model is presented in **Figure 1** and showed the following fit indices: χ 2 (1154) = 3604.10, p < 0.01, CFI = 0.91, SRMR = 0.06, RMSEA = 0.05 (90% CI = 0.05, −0.05), indicating an adequate model fit to the data. Moreover, 64% of the variance in STAI-T depression and 83% of the variance in STAI-T anxiety was explained by metacognitions in this cross-sectional model. Positive beliefs about worry and beliefs about the need to control thoughts did not account for a significant amount of variance in depression TABLE 2 | Descriptive statistics and bivariate correlations among metacognitiveand trait-anxiety variables at time 1 (N = 982).


<sup>∗</sup>p < 0.01, MCQ-30pos = positive beliefs about worry, MCQ-30neg = negative beliefs about the uncontrollability and corresponding danger of worry, MCQ-30cc = cognitive confidence, MCQ-30nc = beliefs about the need to control thoughts, MCQ-30csc = cognitive self-consciousness, STAI-Tdep = trait-anxiety depression subscale, STAI-Tanx = trait-anxiety anxiety subscale.

and anxiety. However, negative beliefs about the uncontrollability and corresponding danger of worry was found to predict a substantial proportion of the variance in both depression and anxiety and was the main predictor of both trait-anxiety constructs. Cognitive confidence was a significant predictor of depression, but not anxiety, and cognitive self-consciousness was a significant predictor of anxiety but not depression.

To determine the consistency of this cross-sectional model over time we re-ran it on the time 2 data. This model showed the following fit indices: χ 2 (1154) = 2723.25, p < 0.01, CFI = 0.90, SRMR = 0.07, RMSEA = 0.06 (90% CI = 0.05, −0.06), indicating an adequate model fit to the data. Moreover, 63% of the variance in STAI-T depression and 82% of the variance in STAI-T anxiety were explained by metacognitions in this model. Negative beliefs about the uncontrollability and corresponding danger of worry predicted both anxiety and depression and was the main predictor of both constructs. Cognitive confidence was also a significant predictor of both depression and anxiety. Cognitive self-consciousness was a significant predictor of anxiety, but not depression. The other MCQ-30 factors were not significant predictors of depression or anxiety in this model. Overall, the model from the T1 data was largely replicated in the T2 data, suggesting that the cross-sectional structural associations between constructs are consistent over time.

# Prospective Relationships Between MCQ-30 Factors and Trait Anxiety

To explore a potential causal association of metacognitions in trait-anxiety, we intended to run SEM with a two-wave cross lagged panel design, a method that has the potential to shed light on temporal precedence. However, the planned statistical approach could not be employed due to very high stability in both domains of trait-anxiety in the longitudinal data (r = 0.93, p < 0.001 for depression, and r = 0.87, p < 0.001 for anxiety), which potentially would lead to spurious cross-over effects (Kline, 2011). Thus, as an alternative we used hierarchical multiple regression analyses. First we ran two models where the traitanxiety domains at T2 were used as dependent variables, and where gender/age, baseline symptom levels (T1 trait-anxiety; depression and anxiety) and T1 metacognitive belief domains

where used as predictors. Gender/age, T1 depression/anxiety and T1 negative metacognitive beliefs were force-entered into the models, while forward entry was used for the remaining T1 metacognitive belief domains to explore if any of these domains entered the model when negative metacognitive beliefs were accounted for. **Table 3** display results from these analyses.

In the final equations, negative- and positive metacognitive beliefs were significant predictors of both STAI-T depression and STAI-T anxiety measured 2 months later. The amount of variance accounted for by metacognitions was very small, a factor that is likely to result from the small amount of residual variance

TABLE 3 | Statistics for the regression equations with time 2 STAI-T depression/anxiety as the dependent and metacognitive belief domains as predictors after controlling for gender/age and time 1 STAI-T depression/anxiety (n = 425).


<sup>∗</sup>p < 0.05 and ∗∗p < 0.01 MCQ-30pos = positive beliefs about worry, MCQ-30neg = negative beliefs about the uncontrollability and corresponding danger of worry.

after controlling for time 1 trait-anxiety which changed little over time.

To further explore these findings, and shed some light on the directionality of associations between metacognitions and trait anxiety we ran two more hierarchical linear regressions where T2 MCQ-30 negative metacognitive beliefs, and T2 MCQ-30 positive metacognitive beliefs were used as dependent variables. In these models, we entered gender/age in the first step, T1 MCQ-30 negative-/positive metacognitive beliefs in the second step, and T1 STAI-T depression and STAI-T anxiety in the final step to explore whether T1 trait-anxiety domains could account for T2 metacognitions when T1 metacognitions were controlled. The results from these regressions suggested that STAI-T depression at T1 was not a significant predictor of T2 metacognitions. Moreover, T1 STAI-T anxiety was not a significant predictor of T2 positive metacognitive beliefs, but it did significantly predict T2 negative metacognitive beliefs.

# DISCUSSION

This study aimed to examine domains of metacognitive beliefs as predictors of trait-anxiety, a marker of psychological vulnerability to depression and anxiety.

In the cross-sectional analyses we found that metacognitive beliefs were positively and significantly correlated with both traitanxiety dimensions. Structural equation modelling of predictors of trait-anxiety domains showed an acceptable fit to the data with 64% of the variance in propensity to depression and 83% of the variance in propensity to anxiety explained by metacognitive beliefs. Here negative metacognitive beliefs were the most substantial contributor to both anxiety and depression with small additional contributions to anxiety of cognitive self-consciousness and to depression of cognitive confidence. The model was replicated in the cross sectional data at time 2 where an additional contribution of cognitive confidence to anxiety also emerged, but the overall model retained a good fit and showed stability of structural relations across time.

Longitudinal analysis informs the possible temporal relations between metacognition and psychological vulnerability. Here, we observed in the hierarchical regression that negativeand positive metacognitive beliefs prospectively predicted both domains of trait-anxiety of which negative metacognitive beliefs explained most of the individual variance. In the reverse model we found that STAI anxiety prospectively predicted negative metacognitive beliefs suggesting a bidirectional causal relationship between these constructs. However, for positive beliefs the pattern was uni-directional with positive beliefs at time 1 predicting both domains of trait-anxiety at time 2 but not the converse. Nonetheless, these results must be considered to be preliminary as other unmeasured factors may account for the relationships observed. Our results indicate a possible causal role for metacognitions in trait-anxiety, but the directionality in these factors requires more rigorous analysis.

The results from our study bring further support for the metacognitive model of psychological disorder, and question

the concept of trait-anxiety as a core (indivisible) vulnerability factor. In the metacognitive perspective (Wells and Matthews, 1994; Wells, 2009), negative affectivity and related constructs such as trait-anxiety and neuroticism may be better understood as markers of maladaptive metacognitions and thinking styles [i.e., the cognitive attentional syndrome (CAS); Wells, 2009]. In the S-REF model, traits are mainly associated with metacognitive beliefs and self-knowledge, and states with the immediate extent and character of metacognitive strategies, namely the CAS (Wells and Matthews, 1994). Metacognitive beliefs (traits) and metacognitive strategies (states) are likely to interact such that maladaptive aspects of personality are enhanced by higher levels of CAS activation. The present data suggest bi-directionality of anxiety and specific negative metacognitions over time, with unidirectionality associated more with positive metacognitions. Trait anxiety may be a topological marker for both the activation of the CAS (e.g., worry/rumination) and of metacognitive beliefs that promote and maintain such processes.

Moreover, our findings confirm a central tenet of the metacognitive (S-REF) model; that both common (i.e., negative beliefs about uncontrollability and danger) and more specific domains of metacognitive beliefs can underlie different presentations of distress or vulnerability. Furthermore, different domains of metacognitions may serve as causal factors constituting vulnerability (i.e., negative- and positive metacognitive beliefs) and as maintenance factors (i.e., negative metacognitive beliefs, cognitive confidence and cognitive selfconsciousness). The pattern of metacognitive predictors is interesting because negative beliefs about uncontrollability and danger emerged as a possible cause and consequence of trait-anxiety, which might be consistent with it having both a generative and maintenance role in susceptibility to distress.

The findings from the present study indicate that psychological vulnerability can be conceptualised within the S-REF model as predicted, a finding that has several clinical implications. Psychological vulnerability in the form of metacognitive knowledge can effectively be modified with Metacognitive therapy (MCT; Wells, 2009). A recent systematic review and meta-analysis shows that MCT is a highly effective treatment for anxiety and depression, and also that it effectively modifies maladaptive metacognitions (Normann and Morina, 2018). Several studies on MCT for individuals with generalised anxiety disorder have shown that severity of traitanxiety decreases following treatment (Wells and King, 2006;

# REFERENCES


Wells et al., 2010; van der Heiden et al., 2012; van der Heiden et al., 2013; Nordahl et al., 2018). Moreover, the S-REF model may inform further research on preventative mental health interventions. For example, it has been suggested that metacognitive therapy applications such as the Attention Training Technique (ATT; Wells, 1990, 2000) could enhance self-regulatory abilities in children by increasing flexible control over attention and thus modify maladaptive meta-level processes and knowledge (Murray et al., 2016, 2018).

This study has several limitations that should be acknowledged. First, the study relied on self-report measures, and a substantial proportion of the participants did not complete measures at time 2. Participants were mostly females. In addition, the sample was gathered at convenience online using social media, which may have biassed the sample characteristics (Wright, 2005). We must be cautious in generalising from these findings. Moreover, we had no control over current health status, meaning that some of the participants may have had psychiatric disorders and be experiencing levels of distress. Because of high stability in domains of trait-anxiety over 8 weeks, one should be cautious when drawing conclusions about the direction of causality based on this data. It remains to be determined if metacognitive belief domains also emerge as significant predictors of other measures of vulnerability such as neuroticism.

# CONCLUSION

In conclusion, the current study suggests that metacognitive beliefs may be an underlying mechanism of vulnerability attributed to trait-anxiety, and that there are both common and more specific domains of metacognitive beliefs associated with the propensity to depression and anxiety. This implies that "vulnerability" may be conceptualised within the metacognitive model and modified with metacognitive therapy (Wells, 2009) with a view to targeting specific dimensions of metacognitive knowledge and thus enhancing psychological resilience.

# AUTHOR CONTRIBUTIONS

HN and AW planned the study and wrote a first draught of the manuscript and all authors contributed substantially to the finalised version. HN, OH, and RH carried out the survey. HN, OH, HMN, and AW conducted the data analyses.



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Nordahl, Hjemdal, Hagen, Nordahl and Wells. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# What Comes First Metacognition or Negative Emotion? A Test of Temporal Precedence

Lora Capobianco1,2 \*, Calvin Heal<sup>3</sup> , Measha Bright<sup>1</sup> and Adrian Wells1,2

<sup>1</sup> Faculty of Biology, Medicine and Health, School of Health Sciences, Division of Psychology and Mental Health, The University of Manchester, Manchester, United Kingdom, <sup>2</sup> Greater Manchester Mental Health NHS Foundation Trust, Manchester, United Kingdom, <sup>3</sup> Faculty of Biology, Medicine and Health, School of Health Sciences, Division of Population Health, The University of Manchester, Manchester, United Kingdom

The Self-Regulatory Executive Function model predicts that emotional symptoms and metacognition can causally affect each other. Crucially, for the model metacognition must cause emotion disorder symptoms. Therefore, in time-series data involving repeated measurements, metacognitions should predict subsequent changes in emotion. 265 participants completed a questionnaire battery three times over a 2 month period. Structural equation modeling (SEM) using cross-lagged panel analysis tested the inter-relationships between metacognitive beliefs, anxiety and depression symptoms over time. The cross-lagged structural model was a significantly better fit than the autoregressive model. Metacognitive beliefs were found to predict subsequent symptoms of anxiety while symptoms of anxiety predicted later metacognition over different time courses. The metacognition factor representing uncontrollability and danger of thoughts appeared to be prominent in the effects observed. Metacognitions and depression were also positively related over time to a lesser degree, but in the crosslagged model these temporal relationships were non-significant. This is likely due to low levels of depression within the sample and low variability over time. The findings for anxiety are consistent with the S-REF model and with experimental and prospective studies supporting metacognitive beliefs as a causal mechanism in psychological distress symptoms.

#### Edited by: Karin G. Coifman,

Kent State University, United States

#### Reviewed by:

Roger Hagen, Norwegian University of Science and Technology, Norway Gabriele Caselli, Sigmund Freud University Vienna, Austria

\*Correspondence:

Lora Capobianco lora.capobianco@manchester.ac.uk

#### Specialty section:

This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology

Received: 17 May 2019 Accepted: 23 October 2019 Published: 19 November 2019

#### Citation:

Capobianco L, Heal C, Bright M and Wells A (2019) What Comes First Metacognition or Negative Emotion? A Test of Temporal Precedence. Front. Psychol. 10:2507. doi: 10.3389/fpsyg.2019.02507 Keywords: metacognitive beliefs, distress, cross-lagged analysis, structural equation modeling, S-REF

# INTRODUCTION

A crucial question in formulating the role of metacognitive factors in emotional symptoms concerns whether or not these factors have a causal or contributory role or merely represent an effect of such dysfunction. The Self-regulatory executive function model (S-REF; Wells and Matthews, 1994, 1996) proposes that specific metacognitions increase emotional dysfunction by, for example, interacting with environmental factors and giving rise to a pattern of extended negative thinking in response to stress. Thus, metacognition should precede symptoms in causal time-series data. Never the less, the model also allows for reciprocal causation, in which emotion can also impact on metacognition. For example, some anxiety or mood symptoms may impair cognitive capacity or be interpreted as a sign of loss of mental functioning thereby strengthening

**34**

metacognitions of lack of control. A pattern of temporal relationships not consistent with the model would occur if negative emotional symptoms only gave rise to later dysfunctional metacognitions, a result that would diminish the causal status of metacognition and present a challenge to the model.

The S-REF model proposes that psychological distress (e.g., anxiety or depression symptoms) is associated with the activation of a style of thinking called the cognitive attentional syndrome (CAS). The CAS is a state of perseverative negative thinking comprised of worry, rumination, focusing on threat, and other maladaptive coping strategies that inadvertently intensify and prolong emotion responses. The CAS is hypothesized to result from metacognitions which exist in the form of knowledge, experiences, and strategies. Such components direct attention, determine thinking style and coping in response to stress cognitions and challenges (Wells, 2009). Metacognitive knowledge is relatively stable and refers to the beliefs that individuals hold about their thinking and can be categorized into positive and negative content. Positive metacognitive beliefs concern the usefulness of cognitive activities that constitute the CAS, e.g., "If I worry, I will be prepared," while negative metacognitive beliefs concern the uncontrollability, dangerousness and importance of thoughts, e.g., "I cannot control my thinking." Such metacognitions, especially negative beliefs are thought to impact on emotion regulation by biasing control efforts leading to perseveration of negative thinking with the effect of increasing or extending negative emotions.

A large number of studies have now demonstrated that the metacognitions predicted by the model are associated with stress symptoms, anxiety or depression (e.g., Wells and Papageorgiou, 1995; Roussis and Wells, 2008; Bennett and Wells, 2010; Yılmaz et al., 2011; Hjemdal et al., 2013; O'Carroll and Fisher, 2013; Halvorsen et al., 2015; Bailey and Wells, 2016; Fergus and Bardeen, 2016; Capobianco et al., 2018a,b). For example, Takarangi et al. (2017) conducted a longitudinal study evaluating whether metacognitive beliefs and metamemory beliefs were associated with the development and maintenance of post-traumatic stress disorder. They found that metacognitive beliefs predicted severity of PTSD symptoms after exposure to a trauma, and the maintenance of PTSD symptoms over time (time 1 to time 2).

Results from experimental manipulations of metacognitive beliefs support a causal role in negative emotion symptoms. Myers and Wells (2013) experimentally manipulated thought-event fusion beliefs (a specific type of metacognitive belief) using a fake-EEG paradigm in individuals with high and low obsessions. They found that inducing such beliefs led to OCD-like symptomology, with this effect being strongest in those with pre-existing high levels of obsessions. Capobianco et al. (2018b) also conducted an experimental manipulation of metacognitive beliefs using a similar fake-EEG paradigm. They evaluated if manipulating the belief of thought importance impacted on physiological and subjective responses to induced stress. Individuals in the experimental condition showed higher levels of negative affect and lower levels of positive affect in response to stress and maintained low positive affect at recovery. In addition to metacognitive beliefs, metacognitive strategies have also been shown to prospectively predict traumatic stress symptoms (Holeva et al., 2001; Roussis and Wells, 2008), anxiety or depression (Yılmaz et al., 2011).

Following from the S-REF model and the results demonstrating an effect of metacognitions we tested the hypothesis that metacognitive beliefs would positively predict later psychological distress measured as anxiety or depression symptoms. We did so using Structural equation modeling (SEM) as this framework allows for the use of the time ordered nature of panel data to address questions of causal orderings (Berrington et al., 2006).

# MATERIALS AND METHODS

# Participants

For purposes of this study data from two samples were combined, in order to provide a sample size sufficient for SEM. Sample size and power calculations for SEM can be challenging (Wolf et al., 2013). Guidelines for SEM sample size varies; it has been suggested that a minimum sample size of 100–200 participants is required (Boomsma, 1982, 1985), but other suggestions include 5 or 10 participants per estimated parameter (Bentler and Chou, 1987), or 10 participants per variable (Nunnally, 1967). Therefore, based on the above recommendations as well as a recent evaluations using Monte Carlo simulations of sample size estimates based on model fit, suggesting a sample size of 250–300 participants (Westland, 2010; Wolf et al., 2013), we opted to combine samples from two sources to provide a sufficient sample size to conduct SEM. Two-hundred and sixty-six participants completed a questionnaire battery. In sample 1, participants (n = 150) were recruited from the University of Manchester. In sample 2, participants (n = 115) were recruited from both the University of Manchester and an online crowdsourcing website. Both samples used the same inclusion criteria; participants had to be at least 18 years of age and proficient in English. Participants ages ranged from 18 to 74 (M = 25.99, SD = 10.64). The sample was primarily female (213 women, 52 men). All participants from both studies completed the study using an online questionnaire software (SelectSurvey.Net). Both studies that provided data were approved by the University of Manchester Research Ethics Committee, reference 15286 (study 1) and reference 2017-2286- 3683 (study 2).

# Measures

Hospital Anxiety and Depression Scale (HADS; Zigmond and Snaith, 1983). The HADS is a 14-item measure with two subscales; anxiety and depression. A total score can also be calculated by summating all items. Items are rated using a 4 point likert scale, where higher scores indicate greater anxiety and depression. Subscales demonstrate good internal consistency, with alpha reliabilities of 0.80 for anxiety and 0.81 for depression (Bjelland et al., 2002) and 0.86 for the total scale (Crawford et al., 2001). The scale demonstrates good reliability and validity (Herrmann, 1997; Bjelland et al., 2002).

Meta-cognitions Questionnaire 30 (MCQ-30; Wells and Cartwright-Hatton, 2004). The MCQ-30 assesses metacognitive beliefs implicated by the S-REF model as linked to psychological vulnerability. The scale has five subscales: positive metacognitive beliefs about worry (e.g., "Worrying helps me to solve problems"), negative metacognitive beliefs about uncontrollability and danger (e.g.,"When I start worrying, I cannot stop"), cognitive confidence (e.g., "I have a poor memory"), cognitive self-consciousness (e.g., "I pay close attention to the way my mind works"), and need for control (e.g., "It is bad to think certain thoughts"). Responses are scored on a scale ranging from 1 (do not agree) to 4 (agree very much). The scale demonstrates good convergent validity, internal consistency, and acceptable test– retest reliability (Wells and Cartwright-Hatton, 2004; Spada et al., 2008; Yılmaz et al., 2008).

# Procedure

After expressing an interest in the study participants received a link to a web site (SelectSurvey.Net) containing the participant information sheet and consent form. Following consent they were able to access the questionnaires. The questionnaire battery was distributed three times within a 2 month period. A 2 month interval was selected as this has clinical relevance; within 1 month stress symptoms normally begin to decrease, however, if they persist longer it could be indicative of a chronic or delayed stress response (deRoon-Cassini et al., 2010), therefore this interval allowed us to investigate the short and long term effects of stress within a meaningful clinical time-frame. Questionnaires were administered at day 0, day 30, and day 60.

# Statistical Analysis Plan

Statistical analyses were conducted in two steps: (1) first we examined invariance of factors over time to ensure we could include the MCQ factors in cross-lagged panel analysis, (2) we then estimated a cross-lagged panel model.

Structural equation modeling was conducted using AMOS for SPSS v 0.23 (Arbuckle, 2014) which uses the maximum likelihood (ML) method to evaluate model fit to the corresponding observed variance-covariance matrices. Model fit was evaluated using a range of fit indices including: the comparative fit index (CFI), Root Mean Square Error of Approximation (RMSEA), Standardized Root Mean Square Residual (SRMR), and Tucker Lewis Index (TLI). The following thresholds were used to assess a good model fit: CFI ≥ 0.90, RMSEA ≤ 0.06, and TLI ≥ 0.95 (Hu and Bentler, 1995), and SRMR ≤ 0.10 (Kline, 2005).

# Latent Variable Identification

In SEM, the relationships between latent variables and between latent and observed measures can be evaluated (Bollen and Noble, 2011). As latent variables cannot be directly observed, they are modeled by specifying the observed, directly measurable variables that express the underlying construct. Latent and observed variables were specified a priori. Metacognitive beliefs were constructed as a latent variable to allow us to evaluate the contribution of individual subscales over time while anxiety and depression were modeled as observed variables in a single model that offered the potential of controlling overlaps between anxiety and depression symptoms at each time point and any temporal relationships between these symptoms.

# Hospital Anxiety and Depression

The HADS was modeled using the corresponding HADS subscales (anxiety and depression), as suggested by the original psychometric analysis of the scale (Zigmond and Snaith, 1983). Each subscale was treated as an observed variable rather than a latent variable as we were interested in separate overall measures of anxiety and depression rather than the contribution of the individual items to a latent general factor. The Cronbach's alpha of the anxiety subscale was 0.81 and for the depression subscale it was 0.75.

# Metacognitive Beliefs

The metacognition latent variable was modeled using the items corresponding to the five subscales of the MCQ-30, which is consistent with Wells and Cartwright-Hatton (2004). The scales demonstrated good internal consistency. Cronbach's alpha for the subscales for the current study were as follows: positive metacognitive beliefs = 0.87, negative metacognitive beliefs regarding uncontrollability and danger = 0.88, cognitive confidence = 0.89, need for control = 0.80, cognitive selfconsciousness = 0.85.

# Cross-Lagged Model Hypotheses

We ran 3-wave structural equation models using ML estimation within AMOS V23 to investigate the longitudinal relationships between metacognitive beliefs and emotional states i.e., anxiety and depression. We began with a basic auto-regressive model, in which the latent variables have a directional effect only on themselves (**Figure 1**). The autoregressive model is the simplest model and acts as a reference against which to compare more complex models. We used a standard formulation of this model, in which MCQ-30 (latent variable) and anxiety and depression (observed variables) are inter-correlated within time points. In addition, the errors on individual variables that model the latents (e.g., MCQ-30 subscales) are assumed to be correlated across time-points. Inclusion of correlated errors followed the suggestion by Fornell (1983), in that inclusion is theoretically driven. We then tested the robustness of this model to violations of the assumptions before proceeding further.

Having established an appropriate autoregressive model, we next tested our hypotheses about the relationship of metacognitive beliefs to anxiety and depression through cross lagged panel analyses. Cross-lagged panel models control for contemporaneous and autocorrelations while identifying timelagged reciprocal effects of constructs assessed repeatedly. We also accounted for the cross-lagged paths between anxiety and depression, which allowed us to evaluate and control any prospective relationships between anxiety and depression in testing if metacognition can prospectively predict anxiety or depression.

FIGURE 1 | Autoregressive Model of Metacognitive Beliefs, Anxiety and Depression: standardized estimates. Solid line, significant path; dotted line, non-significant path; Pos, Positive Metacognitive Beliefs, Neg, Negative Metacognitive Beliefs Regarding Uncontrollability and Danger; CC, Cognitive Confidence; NC, Need for Control; CSC, Cognitive Self Consciousness.



M, Mean; SD, Standard Deviation; HADS, Hospital Anxiety and Depression Scale; MCQ-30, Meta-cognitions Questionnaire 30; PMC, Positive Metacognitive Beliefs; NMC, Negative Metacognitive Beliefs.

# RESULTS

# Data Descriptives

Two-hundred and sixty-Five participants completed the study at all three time points. As less than 10% of the data was missing, mean values were imputed for missing data. Means and standard deviations of the questionnaires across time points are reported in **Table 1**.

Pearson's correlations (**Table 2**) were computed to evaluate the pattern of relationships between measures. All metacognitive beliefs were moderately to strongly positively correlated with anxiety and depression symptoms over time.

# Measurement Invariance

Measurement invariance was evaluated using the four invariance steps (configural, metric, scalar, and residual) as described by Putnick and Bornstein (2016), which coincides with those previously outlined by Widaman and Reise (1997), and Vandenberg and Lance (2000). The structure of the metacognition variable was evaluated across three time points,

#### TABLE 2 |PearsonCorrelations Between Variables.


HADS = Hospital Anxiety and Depression Scale; MCQ-30 = Meta-cognitions Questionnaire 30; PMC = Positive Metacognitive Beliefs; NMC = Negative Metacognitive Beliefs; NEG = Negative Metacognitive Beliefs regarding the Uncontrollability and Danger of Worry; CC = Cognitive Consciousness; NC = Need for Control; CSC = Cognitive Self- Consciousness; T1 = Time 1; T2 = Time 2; T3 = Time 3; ∗ = p<0.05, bold = p<0.01.

where we compared (a) an unconstrained model where all factor loadings and intercepts were allowed to vary freely, (b) a metric invariance model, where factor loadings were constrained equal, (c) a structural invariance model where the factor variances and covariance's were also constrained equal, and (d) a residual invariance model where the residuals of the observed variables were also constrained equal. Measurement invariance was met for the first three steps but not for the final residual invariance model. As the item residuals are not used in the interpretation of mean differences between the latent variables, this step was not strictly necessary to show measurement invariance in this case, but was included for completeness (Vandenberg and Lance, 2000). For this reason further investigations into which residual (s) differed between the two groups were not conducted. The results of the measurement invariance analysis support the interpretation of the subsequent cross-lagged analysis as probably not unduly influenced by instability in measurement models.

# Model Testing

To evaluate if metacognitive beliefs might be a causal mechanism of anxiety and depression over time, cross-lagged panel models were used.

The initial autoregressive model (**Figure 1**; χ <sup>2</sup> = 397.09, df = 165, p < 0.001) demonstrated adequate fit to the data, as the CFI and SRMR values were within the cut-offs for good fit, however, the RMSEA and TLI values were slightly above the cutoffs for good fit; CFI of 0.95, RMSEA of 0.07, SRMR of 0.08, and TLI of 0.93.

We then evaluated the full cross-lagged model (**Figure 2**; χ <sup>2</sup> = 362.36 df = 153, p < 0.001), with cross-lagged paths from metacognition T1 to HADS anxiety T2 and HADS depression T2, from metacognition T2 to HADS anxiety T3 and HADS depression T3, from HADS anxiety T1 to metacognition T2, from HADS anxiety T2 to metacognition T3, from HADS depression T1 to metacognition T2, and from HADS depression T2 to metacognition T3. We also accounted for the causal associations between anxiety and depression across time, as such cross-lagged paths between HADS depression and anxiety across time-points were included. Correlations within time points between anxiety, depression, and metacognitive beliefs were also accounted for. These cross-sectional associations are not depicted in **Figure 2** in order to increase legibility of the figure, however, they were included in the analysis. The cross-lagged paths significantly improved the goodness of fit (1χ <sup>2</sup> = 34.73, df = 12, p < 0.001; RMSEA = 0.07; CFI = 0.95, SRMR = 0.06; TLI = 0.94).

Metacognition was a positive and significant predictor of subsequent metacognitive beliefs (T1-T2: β = 0.81, p < 0.001, T2-T3: β = 0.76, p < 0.001). Similarly, anxiety was a positive and significant predictor of subsequent anxiety (T1-T2:β = 0.74, p < 0.001, T2-T3:,β = 0.73, p < 0.001) and depression predicted later depression (T1-T2:β = 0.68, p < 0.001, T2-T3:β = 0.71, p < 0.001. These results highlight the stability in metacognitive beliefs, anxiety, and depression over the testing intervals.

To evaluate if metacognitive beliefs predicted subsequent anxiety and depression and if the converse relationships applied, we examined cross-lagged regression parameters as follows:

The path from Anxiety T1 to Metacognition T2 was not significant,β = 0.04, p = 0.53, however, the path from Anxiety T2 to Metacognition T3 was significant, with a small beta:β = 0.19, p = 0.006. The finding that anxiety at T2 is predictive of subsequent metacognition is not surprising and is consistent with the metacognitive model as anxious thoughts and emotion can theoretically give rise to metacognition as previously described. However, of greater importance for the theoretical model is the path from metacognition to anxiety, here we found that metacognition at time 1 was predictive of subsequent anxiety with a small beta:β = 0.19, p = 0.008. Although, the path from metacognition at T2 to anxiety at time 3 was not significant,β = 0.0.06, p = 0.34. The beta coefficients for the paths from metacognition at T1 to anxiety at T2, and from anxiety T2 to metacognition are similar in magnitude, which raises the possibility of reciprocal causation. The result is consistent with the hypothesis that metacognitions can precede and predict negative emotion expressed as anxiety symptoms.

Depression T1 did not predict metacognition at T2 (β = −0.01, p = 0.82), nor did depression T2 predict metacognition T3 (β = 0.07, p = 0.23). Similarly, metacognition did not predict subsequent depression, metacognition T1 to depression T2 (β = 0.04, p = 0.61), and metacognition T2 to depression T3 (β = −0.02, p = 0.71). This result is not consistent with the metacognitive model applied to depression symptoms in the current sample. But it is unsurprising given that the sample had low levels of depression symptoms over time with little variation.

We also evaluated the temporal relations between anxiety and depression given that anxiety and depression commonly co-occur and may cause each other. Depression T1 did not predict anxiety at T2 (β = 0.04, p = 0.43), nor did depression T2 predict anxiety at T3 (β = −0.08, p = 0.11). While anxiety T1 did not predict depression at T2 (β = 0.03, p = 0.61), anxiety T2 did predict depression at T3 (β = 0.13, p = 0.03).

# DISCUSSION

The current study evaluated if metacognitive beliefs prospectively predicted psychological distress symptoms measured as anxiety and/or depression. We found evidence of temporal precedence and reciprocal causation over different time lags in anxiety and the data suggested that metacognitions might be the more reliable predictor of anxiety symptoms than the converse. However, the observed effect of anxiety on metacognitions, suggests some reciprocity in these temporal relationships which is consistent with S-REF theory. However, these relationships are only indicative, we cannot rule out the possible influence of other variables that may be acting on both metacognition and symptoms. A more robust test of causal relations would require direct manipulations of metacognition and emotion in evaluating their respective causal effects.

The results for depression were different and did not appear to support any causal relationship between metacognition and mood symptoms. The results are inconsistent with other studies that show prospective relationships (Fergus and Bardeen, 2016; Ryum et al., 2017; Takarangi et al., 2017) and the

prospective bivariate associations found in the current study. However, such studies and our bivariate analyses have not controlled for autoregressive and contemporaneous effects and the relationships observed may have been inflated by these factors. It is likely that the failure to find a cross-lagged relationship in the current study was impacted by the low-level of depression and lack of variability in these symptoms across time in the sample. Alternatively, it may be that the MCQ is less specific for assessing metacognitions associated with depression than with anxiety. None the less, previous studies have found that metacognitive beliefs are positively correlated with symptoms of depression and rumination both cross-sectionally (Papageorgiou and Wells, 2003; Halvorsen et al., 2015; Yılmaz et al., 2015; Huntley and Fisher, 2016) and longitudinally (Yılmaz et al., 2011).

The results add to a corpus of research supporting the idea that specific metacognitions may have a causal effect on emotion symptoms. For example, Capobianco et al. (2018b) demonstrated that induction of a metacognitive belief concerning the importance of thoughts impacted reactions to and recovery from stress exposure. For anxiety at least, the bivariate correlations in the present data set suggest that metacognitive beliefs concerning uncontrollability and danger have the strongest correlations with symptoms cross-sectionally and longitudinally, followed by metacognitions concerning "need for control" and cognitive self-consciousness. The relative strength of relationships is supported by the loadings of subscales on the latent metacognition factor where uncontrollability and need for control are the strongest contributors in the model. These findings are consistent with theory and meta-analyses, mainly of cross-sectional data, demonstrating a contribution from these metacognition domains in particular.

The limitations of the current study should, however, be considered when interpreting the findings. First, the study was primarily conducted in undergraduate students which limits the generalizability of findings, and the preponderance of women in the sample does not facilitate any examination or control of sex differences. The study did not evaluate the impact of environmental or additional factors that may influence the relationship between metacognition and symptoms and so it remains a preliminary and rudimentary test. The timescale of relationships between metacognition and emotion must also be considered. We included a time period that appeared to have clinical relevance based on the trajectory of stress responses linked more to anxiety

(i.e., acute stress and PTSD) but the timescale may not be appropriate for other emotional responses to develop and remit (e.g., depression symptoms). The HADS scores were mainly below clinical cut-offs, limiting any testing of effects that might be more relevant to clinical populations. As such, further research is required to evaluate the replicability of models and pattern of results within clinical samples.

Research evaluating the temporal relationships of metacognitions and symptoms of distress are required in both clinical and non-clinical populations to determine the dynamic temporal relationships between these variables. It is recommended that such studies examine a range of time-frames with a greater number of measurement panels. They should also consider the possibility that some metacognitions may precede emotion effects whilst others may maintain symptoms leading to an increase in recovery time, especially following stress-exposure.

In conclusion, despite the limitations of the current study, the results are consistent with a pattern of temporal relationships between metacognition and anxiety that is consistent with the S-REF model. The results for depression symptoms are inconclusive and most probably affected by floor effects in the data, but an implication of the anxiety result is that anxiety might be prevented by interventions that modify specific dysfunctional metacognitive beliefs.

# REFERENCES


# DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

# ETHICS STATEMENT

The studies involving human participants were reviewed and approved by University of Manchester Research Ethics Committee, reference 15286 (study 1) and reference 2017-2286- 3683 (study 2). The patients/participants provided their written informed consent to participate in this study.

# AUTHOR CONTRIBUTIONS

All authors have contributed to the manuscript revision, read and approved the submitted version. LC was responsible for manuscript writing, data collection, entry, and analysis. CH contributed to data analysis and manuscript writing. MB was responsible for data collection and data entry. AW was responsible for study oversight and manuscript writing. AW and LC contributed to the conception and design of the study.



**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer GC declared a past collaboration with one of the authors AW to the handling Editor.

Copyright © 2019 Capobianco, Heal, Bright and Wells. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Metacognition, Hardiness, and Grit as Resilience Factors in Unmanned Aerial Systems (UAS) Operations: A Simulation Study

Gerald Matthews<sup>1</sup> \*, April Rose Panganiban<sup>2</sup> , Adrian Wells3,4, Ryan W. Wohleber<sup>1</sup> and Lauren E. Reinerman-Jones<sup>1</sup>

1 Institute for Simulation and Training, University of Central Florida, Orlando, FL, United States, <sup>2</sup> Air Force Research Laboratory, Dayton, OH, United States, <sup>3</sup> Division of Psychology and Mental Health, School of Health Sciences, The University of Manchester, Manchester, United Kingdom, <sup>4</sup> Greater Manchester Mental Health NHS Foundation Trust, Prestwich, United Kingdom

Operators of Unmanned Aerial Systems (UAS) face a variety of stress factors resulting from both the cognitive demands of the work and its broader social context. Dysfunctional metacognitions including those concerning worry may increase stress vulnerability, whereas personality traits including hardiness and grit may confer resilience. The present study utilized a simulation of UAS operation requiring control of multiple vehicles. Two stressors were manipulated independently in a within-subjects design: cognitive demands and negative evaluative feedback. Stress response was assessed using both subjective measures and a suite of psychophysiological sensors, including the electroencephalogram (EEG), electrocardiogram (ECG), and hemodynamic sensors. Both stress manipulations elevated subjective distress and elicited greater highfrequency activity in the EEG. However, predictors of stress response varied across the two stressors. The Anxious Thoughts Inventory (AnTI: Wells, 1994) was generally associated with higher state worry in both control and stressor conditions. It also predicted stress reactivity indexed by EEG and worry responses in the negative feedback condition. Measures of hardiness and grit were associated with somewhat different patterns of stress response. In addition, within the negative feedback condition, the AnTI meta-worry scale moderated relationships between state worry and objective performance and psychophysiological outcome measures. Under high state worry, AnTI meta-worry was associated with lower frontal oxygen saturation, but higher spectral power in high-frequency EEG bands. High meta-worry may block adaptive compensatory effort otherwise associated with worry. Findings support both the metacognitive theory of anxiety and negative emotions (Wells and Matthews, 2015), and the Trait-Stressor-Outcome (TSO: Matthews et al., 2017a) framework for resilience.

#### Edited by:

Changiz Mohiyeddini, Northeastern University, United States

#### Reviewed by:

Roger Hagen, Norwegian University of Science and Technology, Norway Antonino Carcione, III Centro Psicoterapia Cognitiva, Italy

> \*Correspondence: Gerald Matthews gmatthews@ist.ucf.edu

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 18 September 2018 Accepted: 07 March 2019 Published: 26 March 2019

#### Citation:

Matthews G, Panganiban AR, Wells A, Wohleber RW and Reinerman-Jones LE (2019) Metacognition, Hardiness, and Grit as Resilience Factors in Unmanned Aerial Systems (UAS) Operations: A Simulation Study. Front. Psychol. 10:640. doi: 10.3389/fpsyg.2019.00640 Keywords: metacognition, worry, grit, resilience, stress, psychophysiology, Unmanned Aerial Systems, workload

# INTRODUCTION

Individual differences in resilience and stress vulnerability have profound personal consequences for life outcomes such as career success, personal relationship quality, and mental health. Recent work has demonstrated the complexity of resilience, which depends on multiple personality traits whose influence on stress outcomes varies across different demanding contexts

**43**

(Matthews et al., 2017a). The present study explores the contribution to resilience and stress vulnerability of worry traits including meta-worry, i.e., metacognitive worry about worry itself (Wells, 1994, 2005). A simulation of Unmanned Aerial System (UAS) control provided a testbed for manipulating stress and assessing multiple components of stress response. The overall aim was to examine how worry and additional traits for resilience predicted stress response within the frameworks provided by the metacognitive theory of maladaptive emotions (Wells and Matthews, 1996, 2015) and multifactorial resilience theory (Matthews et al., 2017a).

# Worry, Metacognition, and Stress

Personality traits for emotional vulnerability and resilience can be broadly divided into maladaptive traits that amplify harmful impacts of stressors and adaptive traits that support effective coping. Beyond broad traits such as neuroticism, theoretical considerations suggest a focus on dispositional worry and metacognition. Specifically, the Self-Regulatory Executive Function (S-REF) theory (Wells and Matthews, 1996, 2015) defines a Cognitive Attentional Syndrome (CAS) associated with perseverative worry, rumination and threat-monitoring that interferes with task-directed attention and causes psychological dysfunction. The CAS is typically triggered by an external threat or an intrusive thought, but persistence of the syndrome results from metacognitions that maintain attention on negative selfreferent thoughts. For example, the person may believe that intrusive thoughts are important, that worrying about them will resolve personal concerns, or that thoughts are uncontrollable (Wells, 2000). The impact of metacognitions in the S-REF model is multifaceted, potentially impacting motivation to regulate cognition and utilize effort, choice of response strategy, and the threat assigned to cognitive processes themselves.

Dispositionally worry-prone individuals are vulnerable to the CAS and states of worry in performance settings (Matthews and Funke, 2006). Dispositional worry and allied constructs such as rumination are also risk factors for both subclinical stress reactions to life events and emotional disorders (Kircanski et al., 2017; Ryum et al., 2017). However, dispositional worry is itself a complex construct, that may variously include specific personal concerns such as health and social status (Wells, 1994), maladaptive styles of stress processing such as excessive threat appraisal and avoidance coping (Borkovec et al., 2004; Matthews and Funke, 2006), and metacognitive factors, such as beliefs about worry as specified by S-REF theory (Wells and Matthews, 2015).

Evidence from both experimental and correlational studies demonstrates the role of metacognitions in acute stress in nonclinical samples. Palmier-Claus et al. (2011) used disturbing images as a stressor. They found that the negative affect response was stronger in individuals with negative metacognitive beliefs referring to the importance of controlling one's thoughts, and the uncontrollability of thoughts. Capobianco et al. (2018b) induced metacognitions directly with a fake EEG manipulation that lead participants to believe their negative thoughts would trigger a burst of aversive noise. The manipulation amplified and prolonged the negative emotional response to a subsequent stressor (the Trier Social Stress Test). A further experimental study utilizing the Trier test (Capobianco et al., 2018a) showed that a group in whom worry was induced experimentally showed elevated negative affect immediately following the stressor exposure.

In correlational studies, dysfunctional metacognitions have been associated with test anxiety and maladaptive coping (Matthews et al., 1999), perceived stress symptoms (Roussis and Wells, 2006; Spada et al., 2008), state anxiety (Spada et al., 2010), and anxiety and depression when life events are controlled (Yılmaz et al., 2011). Because these various stress responses are likely to distract from attention to tasks, it is expected that the negative metacognitions that drive them will be maladaptive in the performance context. Whilst there is a large body of research supporting central predictions of the S-REF model of vulnerability (Wells, 2013; Wells and Matthews, 2015), little is known about factors associated with the CAS and metacognitions that enhance resilience.

# A Multifactorial Perspective on Resilience: The TSO Framework

The current study focuses on stress response during performance of a multi-component cognitive task. A basic challenge in identifying the role of metacognition in this context is the complexity of individual differences in stress response. Resilience traits additional to metacognitive factors may also influence response. Furthermore, the nature of the stressor may moderate the relationship between traits for resilience and stress outcomes. Findings may depend too on the stress outcome measure examined. For example, psychophysiological measures can pick up stress responses of which the person is not consciously aware (Verkuil et al., 2010). Matthews et al. (2017a) proposed a Trait-Stressor-Outcome (TSO) framework for specifying dispositional individual differences in stress response across different contexts. It emphasizes that the traits that predict stress reactivity vary from stressor to stressor, and influence different stress outcomes. From the TSO perspective, we may ask which stressors elicit differential responding in individuals differing in metacognition, which outcome measures demonstrate differential response, and how the role of metacognition compares to other relevant resilience traits.

The present study investigated the stress of operating multiple UASs, aerial vehicles controlled remotely for purposes including reconnaissance and surveillance. Current military and civilian operations typically involve a two- or three-person team controlling the vehicle; in the future a single operator will control multiple vehicles with assistance from automation (Calhoun and Draper, 2015; Wohleber et al., 2018). Stressors include the cognitive challenges of managing complex interfaces, variable workload, social evaluation, and long workshifts (Tvaryanas et al., 2006; Paullin et al., 2011). Some of these stressors are more likely to elicit the CAS than others. Social-evaluative stress commonly elicits both worry (Zeidner, 1998) and physiological stress response (Dickerson and Kemeny, 2004). In the military training context, stress response is accentuated when trainees feel that their performance is being judged by peers and instructors and they receive critical feedback (Carroll et al., 2014). By

contrast, high cognitive workloads are also stressful but they may direct attention outward to manage high volumes of external task stimuli, limiting the potential onset of the CAS.

Performance stress is expressed in various ways, through subjective experience, changes in neural functioning, and objective performance impairment. Subjective states experienced in performance environments may be assessed using the Dundee Stress State Questionnaire (DSSQ: Matthews et al., 2002, 2013). It identifies 11 primary affective, motivational and cognitive dimensions that define higher-order factors of task engagement, distress, and worry. The worry state combines self-focused attention, low performance selfesteem, and intrusive thoughts about the task and personal concerns. Matthews (2016) reviewed studies showing that different task stressors elicit different patterns of response across the dimensions, that reflect the different appraisals and coping strategies that shape each state dimension. Acute stress response is often identified with sympathetic arousal but studies of stress elicited by high-workload tasks reveal a more complex picture. Matthews et al. (2015) recorded electrocardiac, electroencephalographic, and hemodynamic responses to multiple tasks, and found that responses from different physiological systems dissociated, implying that multiple brain systems may underpin response to task stressors. Multivariate assessment is important because different responses may have differing functional significance. For example, in a simulation of unmanned ground vehicle operation, Matthews et al. (2017b) found that subjective and physiological measures contributed independently to performance prediction; high distress, low heart rate variability, and high frequency EEG were all associated with performance impairment.

The TSO framework assumes that multiple traits may moderate stress response, depending on the context. Traits for resilience refer to focus on positive qualities supporting coping, whereas stress vulnerability traits define qualities such as worrying that are detrimental to coping. Broad trait models typically characterize positive and negativity emotionality dimensions as largely independent (Watson, 2000), but it is unclear whether resilience and vulnerability traits can be neatly partitioned into two separate categories; for example, dysfunctional metacognition may undermine the task-directed motivations that support resilience. According to TSO, traits adaptive for stressful performance settings should be those that maintain attentional focus, task-directed effort and problemfocused coping. Multiple traits are potentially relevant, but here we focus on hardiness (Bartone et al., 1989; Escolas et al., 2013), and grit (Duckworth and Quinn, 2009). Such traits have cognitive, motivational, and emotional aspects, but their relationship to metacognition and the CAS of S-REF theory (e.g., worry) is unknown.

The construct of hardiness as a general trait for resilience emerged from studies of personality traits that might buffer the health impacts of life stressors (Kobasa, 1982). Scales for hardiness (e.g., Bartone et al., 1989) have been widely utilized in studies of stress in organizational, military and other contexts, generally confirming that the trait enhances resilience and performance under stress (Maddi et al., 2012). A metaanalysis (Eschleman et al., 2010) confirmed that hardiness is substantially correlated with various measures of higher wellbeing and lower stress, including lower scores on measures of depression and traumatic stress. Hardiness was also associated with adaptive cognitive stress processes such as preferences for problem-focused and approach coping over avoidance and emotion-focused coping. Hardiness also correlates with more constructive appraisals (Cash and Gardner, 2011). The role of metacognitive style in hardiness has been overlooked. However, because the adaptive pattern of coping and appraisal associated with the trait tends to mitigate against development and maintenance of the CAS (Wells and Matthews, 2015), it is hypothesized that hardiness will be negatively associated with maladaptive metacognition.

Definitions of grit focus on long-term persistence and maintenance of motivation during adversity (Duckworth and Quinn, 2009), but this trait may also influence acute stress response to task performance challenges. Grit correlates positively with wellbeing and mental health, and negatively with stress and symptoms of depression (Goodman et al., 2017; Sharkey et al., 2017; Kannangara et al., 2018), although the literature is not fully consistent (Kannangara et al., 2018). Grit also correlates with cognitive and self-regulative processes that may confer resilience including positive control beliefs (Goodman et al., 2017), self-efficacy (Muenks et al., 2018), and self-control (Duckworth and Gross, 2014). It is also associated with lower levels of brooding and reflective rumination (White et al., 2017). In addition, studies of grit in the academic context show relationships with processes supporting selfregulated learning including adaptive metacognitive strategies for planning, monitoring and regulating the learning process (Wolters and Hussain, 2015). Thus, high-grit individuals should be more effective at self-regulation when required to perform a stressful cognitive task. From a theoretical standpoint, grit is associated with positive attitudes despite setbacks and failure (Lucas et al., 2015), and with low levels of ruminative processes (White et al., 2017). These characteristics should act against prolonged CAS activation in stressful task environments.

# The Present Study

There has been rather little research on the relationship between dispositional worry, metacognition, resilience, and stress responses in complex, demanding performance environments. This lack of evidence represents a limitation of both CAS and TSO models. In the current multi-UAS control task, the participant must guide vehicles to target locations and photograph them while monitoring for vehicle health and avoiding areas of danger. Panganiban and Matthews (2014) developed and validated two stress manipulations, one that increased cognitive demand and one that delivered negative feedback about performance. We considered that negative feedback was more likely than high cognitive demand to activate the CAS, because it involved direct personal criticism.

In the present study, a within-subjects design was used. All participants performed under both stressors, as well as in two control conditions, one prior to each stressor (four conditions

in total). We aimed to test whether traits for stress vulnerability (i.e., those predisposing activation of the CAS) and resilience predicted physiological and subjective responses, utilizing a suite of sensors previously applied across a range of demanding task environments (Matthews et al., 2015; Reinerman-Jones et al., 2016). We administered the Anxious Thoughts Inventory (AnTI: Wells, 1994), which assesses traits related both to specific worry concerns (social and health) and to meta-worry, along with scales for two adaptive resilience constructs, hardiness (Bartone et al., 1989) and grit (Duckworth and Quinn, 2009).

Stress responses in demanding performance environments change dynamically throughout the test session (Matthews and Campbell, 2010). People differ in anticipatory stress and worry before exposure to stressors (Brosschot et al., 2006); a study of medical students (O'Carroll and Fisher, 2013) found that trait anxiety and metacognitive factors predicted anxiety immediately prior to a clinical examination. Thus, evaluating stress reactivity requires control for individual differences in stress at baseline. In this study, we analyzed both baseline and reactivity data. Associations between resilience traits and subjective states at baseline identifies factors associated with stress state in the absence of substantial overt demands. We also evaluated individual differences in reactivity, by testing for associations between traits and stress response with baseline levels of stress controlled. Subjective and physiological responses were analyzed to test for specificity of response; i.e., some individuals might show strong responses to negative evaluation but not cognitive overload, and vice versa. Having identified individual differences in stress reactivity to negative feedback, we then ran a further analysis to test whether dispositional meta-worry moderated associations between worry and objective outcomes as predicted by the CAS model. The specific research issues addressed were as follows:

# Stress Profiles of Cognitive Demand and Feedback Stressors

We expected that both stress manipulations would elevate subjective distress (Panganiban and Matthews, 2014) and psychophysiological stress indices including highfrequency EEG activity (Reinerman-Jones et al., 2016). However, we also anticipated qualitative differences in response associated with each stressor, including higher workload with cognitive demand and higher state worry with negative feedback.

## Associations Between Traits and Stress States: Baseline and Control Conditions

Metacognitive factors correlate with perceived stress in the absence of an overt stressor (Spada et al., 2008), and traits for worry and metacognition are associated with anticipatory anxiety (O'Carroll and Fisher, 2013). Thus, we hypothesized that the AnTI traits would be associated with elevated DSSQ distress and worry, as well as psychophysiological stress measures. We also anticipated negative associations between AnTI traits and hardiness and grit, as well as correlations between these traits and higher task engagement, lower distress, and lower worry.

# Worry and Resilience Traits and Reactivity to Stressors

We tested whether traits would predict stress response over and above any associations evident in the control conditions. To do this, we computed measures of stress reactivity specific to each stressor. We expected that the AnTI would predict subjective and physiological responses to negative feedback more strongly than responses to cognitive demand, because feedback is more likely to activate the CAS due to its higher self-relevance. Accounts of hardiness and grit do not clearly link these qualities to specific stressors so their associations with reactivity were investigated on an exploratory basis.

# Metacognition and the Functional Significance of Worry

Worry states are broadly if modestly detrimental to performance (Zeidner, 1998; Matthews and Funke, 2006), but recent work has also identified potential functional advantages of worry including motivating problem-solving and coping efforts (Sweeny and Dooley, 2017). We can infer from the S-REF theory (Wells and Matthews, 2015) that relationships between worry and adaptive outcomes may be moderated by metacognitive style. Specifically, individuals high in meta-worry are likely to react to the awareness of worry by re-directing attention and effort from task performance to processing and regulating the worry state, whereas those low in meta-worry are more likely to use worry as a spur to increase task-directed effort. This hypothesis was tested against objective measures of performance and psychophysiological response in the negative feedback stressor condition, in which CAS activation was most likely.

# MATERIALS AND METHODS

# Participants

Participants were 68 undergraduate students (31 women, 37 men, Mage: 19.3 years) at the University of Central Florida. They received course credit for participation. Participants were excluded if they reported current or recent treatment for any emotional disorder, eating disorder, schizophrenia or other psychosis, stress or any related emotional condition. Those currently taking psychoactive medications were also excluded.

# Subjective Measures

# Anxious Thoughts Inventory (AnTI: Wells, 1994)

This questionnaire includes 22 items answered on 4-point response scales. It includes subscales for social worry (e.g., "I worry about my appearance"), health worry (e.g., "I have thoughts about becoming seriously ill"), and meta-worry (e.g., "I have difficulty clearing my mind of repetitive thoughts"). Subscale alpha coefficients quoted by Wells (1994) ranged from 0.75 to 0.84.

# Hardiness Scale (Bartone et al., 1989)

This measure of resilience has 30 items, answered on 4-point response scales. The subscales are commitment (e.g., "Most days, life is really interesting and exciting to me," challenge (e.g., "I

like it when things are uncertain or unpredictable"), and control (e.g., "When I make plans, I'm certain I can make them work"). Bartone et al. (1989) reported an alpha of 0.83 for total hardiness, and subscale alphas ranging from 0.62 to 0.82.

## Short Grit Scale (Duckworth and Quinn, 2009)

This questionnaire includes 12 items, answered on 5-point response scales, which assess capacity to sustain effort and interest in demanding activities (e.g., "Setbacks don't discourage me"). Scale alphas in four samples ranged from 0.73 to 0.83.

## Short Dundee Stress State Questionnaire (DSSQ: Matthews et al., 2013)

The short, 21-item version of the DSSQ assesses subjective state responses related to task engagement (e.g., "I was determined to succeed"), distress (e.g., "I felt tense"), and worry (e.g., "I reflected about myself "). Items are answered on 4-point scales. Scale alphas range from 0.78 to 0.83 (Matthews et al., 2013).

# NASA Task Load Index (NASA-TLX: Hart and Staveland, 1988)

This workload measure requires the respondent to use 0–100 scales to rate 6 sources of task load (mental demand, physical demand, temporal demand, effort, frustration, performance). Overall workload is calculated as an average of ratings, with performance reverse scored. The scale authors reported a testretest reliability of 0.83.

# Psychophysiological Measures

A suite of sensors used in previous studies recorded multiple psychophysiological responses. Brief descriptions are given here: see previous reports for further detail (Barber et al., 2011; Matthews et al., 2015). Multiple responses were recorded simultaneously during an initial baseline period and throughout task performance.

# Electroencephalogram (EEG)

The ABM B-Alert X10 system assessed nine channels of EEG. Following filtering and artifact removal, spectral power was averaged across three frontal sites for theta (4–8 Hz), alpha (9– 13 Hz), beta (14–30 Hz), and gamma (30–100 Hz) bandwidths. EEG data were analyzed as percent change from baseline.

# Electrocardiogram (ECG)

The ABM System B-Alert X10 system also recorded ECG. Mean Inter-Beat Interval (IBI) and Heart Rate Variability (HRV) were recorded. IBI was analyzed as percent change from baseline for each task condition. HRV was calculated as the SD of all beats (measured in ms) during each condition.

# Functional Near-Infrared Spectroscopy (fNIR)

Hemodynamic changes in the left and right hemispheres of the prefrontal cortex were measured using Somanetics' INVOS Cerebral/Somatic Oximeter. The fNIR method analyzes the spectral absorption of NIR light by brain tissue. Regional oxygen saturation (rSO<sup>2</sup> ) during each condition was calculated as the percent change from baseline.

# Transcranial Doppler Sonography (TCD)

Cerebral blood flow velocity (CBFV) in the left and right hemisphere middle cerebral arteries was measured using Spencer Technologies' ST3 Digital Transcranial Doppler system. The system transceiver emits ultrasound pulses that are reflected back to the sensor from the moving blood cells; velocity is calculated from analysis of the Doppler shift in frequency. CBFV was calculated as the percent change from baseline.

# Apparatus

We used the Java-based "Research Environment for Supervisory Control of Heterogeneous Unmanned Vehicles" (RESCHU) multi-UAV simulator developed by the Human and Automation Lab at the Massachusetts Institute for Technology (Boussemart and Cummings, 2008). Full details are provided by Panganiban (2013). In brief, RESCHU simulates complex dynamic supervisory control. A single operator controls multiple UASs performing surveillance missions, using the mouse to control the vehicles via a point-and-click interface. The display includes multiple windows as shown in **Figure 1**. The aim was to assign UASs to searchable targets represented by red and gray diamond symbols on a map display. Participants visually identified key objects on arrival of the UAS at the target site. Each UAS was identified by a number. To perform the task, the participant first allocated a UAS to a given target, allocating odd-numbered UASs to red targets and even-numbered UASs gray targets. The participant then used the mouse to define waypoints along a path to the target, in order to avoid hazardous regions, represented by yellow circles. If a UAS entered such a region it took damage. When the UAS arrived at the target, the participant was informed in the message window. The participant accessed a "payload window" that displayed a camera view of the ground below. The message window specified a specific object to locate, such as "yellow car" or "a building with a blue roof." The participant used the mouse to control the camera view and to zoom in and out as necessary to locate the object. The task is made more difficult by the expiration of targets and the disappearance and reappearance of hazards. Target areas and hazard areas have countdown timers and each moves to a new position on the map once its timer reaches zero. The task is considered to require multiple cognitive capabilities including planning, visual scanning, visual memory, allocation of attention, and multi-tasking which together support integrated executive functioning in a complex and dynamic task environment (Ratwani et al., 2010).

Stress manipulations were similar to those used by Panganiban and Matthews (2014). For lower-stress control trials, participants controlled two UASs. Fourteen targets and nine hazards were present on the screen. Targets expired after 60 s and hazards relocated every 5 s. In the negative feedback stressor condition, the same task configuration was used, but scripted feedback referring to participants' performance was provided in the mission window every 30 s. Approximately two-thirds of the feedback statements were negative (e.g., "You are not meeting expectations"); the remainder were neutral ("You are performing adequately"). Messages were presented in a pseudo-random sequence unrelated to actual performance. This manipulation

FIGURE 1 | RESCHU simulator. The payload window for search tasks is located on the top left. The message window is below the payload window, and below that is a vehicle information display. The map display shows targets (red and gray), hazards (yellow), and UASs (blue).

was expected to activate the CAS in vulnerable individuals. In the cognitive demand stressor condition, cognitive demands were increased by increasing the number of UASs, the numbers of targets and hazards, and decreasing the time for which each target was available. In this condition, participants controlled six UASs, and with 18 targets and 14 hazards consistently present on the screen. Targets expired after 45 s, hazards after 5 s. Measures of performance effectiveness were (1) the command ratio, the number of targets engaged divided by the number of targets assigned, and (2) search accuracy, the number of objects located divided by the number of targets engaged. We also assessed (3) waypoints added, the total number of waypoints set in routing vehicles to targets.

# Procedure

Following an informed consent interview, participants completed questionnaires including the AnTI, Hardiness and Grit scales, and pre-task DSSQ. The physiological sensors were then attached and data recording quality was verified. Participants watched a blank screen for 5 min during which baseline physiological measures were secured. Participants then received training on the task. They viewed a Powerpoint slideshow which explained the nature of the task and then practiced on the lower cognitive demand version of the task. Performance was monitored by the experimenter to ensure participant competence was sufficient to move onto the main part of the task. Participants then performed a sequence of four trials in one of two orders; either control – negative feedback – control – high demand or control – high demand – control – negative feedback. Thus, each stressor was preceded by its own control condition. Order was counterbalanced across participants. Stressor trials were 10 min in duration; control trials were 5 min. After each trial, the participant completed the NASA-TLX and a posttask DSSQ. Finally, physiological sensors were removed and participants were debriefed.

# RESULTS

The study provided an extensive data set. Thus, analyses were targeted to address the four research issues previously identified, and they are presented as follows. First, we verified that the two stressors were effective in eliciting stress responses, and we ran ANOVAs to test whether they elicited different

patterns of stress response. Second, we computed correlations between the various traits and stress states in relatively undemanding conditions, i.e., at baseline and in control conditions. This analysis tested whether the AnTI correlates with stress in the absence of an overt stressor. Third, we computed correlations between traits and the stress reactivity measures for the cognitive demand and negative feedback conditions, testing whether the AnTI specifically predicts stress response to feedback, as hypothesized. Fourth, we focused in on the role of meta-worry as a moderator of responses to negative feedback. We used a regression approach to test for interactions between AnTI meta-worry and subjective worry state in predicting objective performance and physiological outcomes, testing for whether meta-worry controls whether or not worry states are maladaptive.

# Stress Profiles of Cognitive Demand and Negative Feedback Stressors

Dependent stress response measures were the three DSSQ scales, NASA-TLX workload, and the psychophysiological measures from EEG, ECG, fNIR and TCD. A 2 × 2 (stress level × stress type) repeated measures ANOVA was run for each one. A significant main effect of stress level, with no interaction, implies that both stressors influenced the response measure. A significant interaction indicates a differential effect of stressors on the measure. The significant effects in this analysis are summarized in **Table 1** (full ANOVA tables are available from the authors). There were no significant effects on DSSQ worry, ECG IBI, or fNIR.

**Figure 2** illustrates stressor effects on subjective variables. Both stressors increased distress and workload, but both effects were stronger for the cognitive demand manipulation, as evidenced by the significant interactions between factors. For task engagement, only the interaction reached significance. Cognitive demand increased engagement slightly, whereas negative feedback reduced engagement more substantially.

**Figure 3** shows principal stressor effects on the physiological variables. For most, only the main effect of stress level was significant. Both manipulations tended to increase heart rate variability and high frequency EEG spectral power (beta and gamma). Small-magnitude increases in theta and alpha under stress (not graphed) were also obtained. The only stressor-specific effect was for CBFV; blood flow velocity was lowest in the negative feedback condition.

A similar analysis of performance measures showed significant stressor effects on all three performance measures. The command ratio was lower in the high demand condition (M = 0.57, SD = 0.10) compared to the negative feedback condition (M = 0.77, SD = 0.12), the control condition for high demand (M = 0.77, SD = 0.11), and the control condition for negative feedback (M = 0.70, SD = 0.08). Search accuracy (proportion correct) was lower in both the high demand condition (M = 0.84, SD = 0.10) and in the negative feedback condition (M = 0.85, SD = 0.07) relative to the respective control conditions (M = 0.85, SD = 0.09; M = 0.87, SD = 0.08). The number of waypoints set was higher in the high demand condition (M = 4.54, SD = 3.49) than in the negative feedback condition (M = 3.19, SD = 2.50), or in the two respective control conditions (M = 2.91, SD = 2.47; M = 3.06, SD = 2.90). This last effect primarily reflects the need to set more waypoints when there are larger number of vehicles to direct.

# Associations Between Traits and Stress States: Baseline and Control Conditions

**Table 2** shows intercorrelations of the traits and subjective state measures at pre-task baseline. All AnTI scales were associated with higher DSSQ worry, and also with lower

TABLE 1 | ANOVA summary statistics for stress response measures that show significant stressor effects.


<sup>∗</sup>p < 0.05, ∗∗p < 0.01. <sup>1</sup>Analysis also included hemisphere factor.

task engagement, showing relationships with anticipatory stress. The AnTI, especially its social worry and metaworry scales, was significantly negatively correlated with both hardiness and grit scales, with the exception of the challenge subscale. Hardiness and grit were correlated with more positive subjective states, but, by contrast with the AnTI, they were associated with (lower) distress, rather than with worry.

Subjective state variables were averaged across the two control conditions, to estimate state when task demands were undemanding. The correlations across the two control conditions for the DSSQ scales were 0.53 (task engagement), 0.66 (distress), and 0.73 (worry), showing individual differences were fairly consistent across the two conditions. The AnTI scales (except health worry) remained significantly positively correlated with state worry, but associations with task engagement were nonsignificant. DSSQ correlates of grit and hardiness were similar to those at baseline, with some differences in detail; for example, in the control conditions, both traits were significantly negatively correlated with state worry. Correlations between the trait scales and psychophysiological measures in the control conditions were also calculated but significant associations were few, and did not suggest any clear relationship between the traits and stress responses (data are available from the authors on request).

# Worry and Resilience Traits and Reactivity to Stressors

We calculated residualized indices of reactivity by regressing each subjective and physiological stress response measure for the two stressor conditions against the same measure in the matched control condition. For example, state worry for the negative feedback condition was regressed against state worry in the preceding control condition, and the standardized residual was calculated. The residual expresses the extent to which the measure is higher or lower than its value in the control condition predicts. Cross-stressor correlations in residuals were all nonsignificant, e.g., the three DSSQ residual correlations ranged from 0.08 to 0.18.

**Table 3** shows correlations between the trait measures and residuals for the subjective state variables, for negative feedback and cognitive demand stressors. The AnTI showed a highly specific set of associations with worry reactivity. Total AnTI score, and two out of three subscales, were significantly correlated with state worry response. The additional resilience traits were

TABLE 2 | Correlations between resilience traits and DSSQ state measures at baseline and in control conditions.


<sup>∗</sup>p < 0.05, ∗∗p < 0.01.


Social 0.048 0.101 0.261<sup>∗</sup> 0.212 0.203 0.141 Health −0.030 0.103 0.285<sup>∗</sup> 0.172 −0.008 0.075 Meta-worry 0.113 0.105 0.196 0.022 0.156 0.046

Commitment 0.122 −0.306<sup>∗</sup> −0.212 0.170 −0.143 −0.050 Control 0.114 −0.246<sup>∗</sup> −0.198 0.064 0.014 −0.155 Challenge 0.149 −0.282<sup>∗</sup> −0.306<sup>∗</sup> 0.045 −0.004 −0.097

Hardiness Total 0.052 −0.394∗∗ −0.344∗∗ −0.221 −0.247<sup>∗</sup> 0.088

Grit Total 0.009 −0.155 0.033 0.280<sup>∗</sup> −0.345∗∗ 0.157

TABLE 3 | Correlations between resilience trait measures and stress reactivity: Subjective response (residualized).

<sup>∗</sup>p < 0.05, ∗∗p < 0.01.

more broadly correlated with reactivity. Total hardiness was associated with an attenuated distress response to both stressors, and with reduced worry in the negative feedback condition. All three hardiness subscales predicted lower distress response to negative feedback. Grit was exclusively associated with reactivity to the cognitive demand stressor, specifically with higher task engagement and lower distress.

Comparable correlations for residuals for selected psychophysiological measures are provided in **Table 4**. In this analysis, most of the correlations were non-significant, and the trait scales were significantly correlated only with EEG measures. Multiple significant correlates of theta and gamma response were found. The AnTI scales were associated with weaker theta and stronger gamma response. The hardiness commitment scale along with grit predicted stronger theta response; commitment also predicted lower gamma.

# Metacognition and the Functional Effects of Worry

It was hypothesized that individuals high in AnTI meta-worry would be more likely to show maladaptive responses with increasing state worry, relative to those low in meta-worry. Given the theoretical rationale for meta-worry being more likely to influence stress response to negative feedback than to cognitive demand, along with the preceding analyses, this hypothesis was tested only in the negative feedback condition, using a regression approach. Each performance and psychophysiological variable was treated as the dependent variable in turn.

The dependent variable was predicted from linear terms for AnTI meta-worry and DSSQ state worry in the negative feedback condition, and the centered product term representing the interaction. In the analyses of performance, there were no significant linear or interactive effects for the command ratio or search accuracy measures. However, for waypoints added, the interaction was significant (β = −0.293, p < 0.05), though not the linear terms. The regression lines for individuals 1 SD above and below the mean are plotted in **Figure 4** (top). As worry increases, individuals high in meta-worry assign progressively fewer waypoints, suggesting reducing task effort. Low meta-worry persons show the opposite trend.

For the physiological variables, the meta-worry × state worry interaction was significant for the left hemisphere fNIR rSO<sup>2</sup> response (β = −0.296, p < 0.05), right fNIR rSO<sup>2</sup> response (β = −0.279, p < 0.05), EEG beta (β = −0.308, p < 0.05), and EEG gamma (β = −0.338, p < 0.01). Linear terms were nonsignificant in all cases. The interactions for fNIR resemble those



<sup>∗</sup>p < 0.05, ∗∗p < 0.01.

for waypoints added (**Figure 4**, center). Increasing worry appears to decrease frontal oxygen saturation in those high in metaworry, with the opposite effect in low meta-worry individuals. Plots of the regression lines for high frequency EEG (beta and gamma) show that power tended to decrease with increasing worry in low meta-worry persons, with those high in metaworry showing the opposite trend. These regressions include a linear trend toward decreasing power as state worry increases (significant at 0.05<p < 0.10 in both equations).

# DISCUSSION

Traits for resilience predicted subjective and physiological responses to negative feedback and cognitive demand stressors in a multi-UAS control simulation. As expected, worry traits, including meta-worry, were generally associated with higher levels of situational stress, whereas hardiness and grit appeared protective. The data also revealed more subtle relationships between traits and stress outcomes. As predicted, the AnTI was predictive of stressor reactivity primarily in the negative feedback condition, consistent with cognitive-attentional theory (Wells and Matthews, 2015). The moderator effect of meta-worry on relationships between subjective state worry and objective stress responses was also consistent with theory; worry appears to be especially maladaptive for those high in meta-worry. Hardiness and grit were negatively correlated with the AnTI worry scales: maladaptive metacognitive style may impair development of a resilient personality. **Table 5** summarizes the evidence supporting each of the major hypotheses of the study. The remainder of this discussion addresses the four relevant research questions, as well as limitations and practical applications of the study.


TABLE 5 | Summary of major research questions and outcomes confirming hypotheses.

# Stress Profiles of Cognitive Demand and Feedback Stressors

Both stressors elicited higher state distress, as expected, but the effect was larger for cognitive demand. High workload plays a major role in provoking the subjective distress response in task performance contexts, as the person appraises the task as uncontrollable and utilizes multiple forms of coping to manage overload (Matthews and Campbell, 2009; Matthews et al., 2013). Contrary to expectation, negative feedback did not elicit higher state worry. As discussed subsequently, the different stressors may have influenced the qualitative nature rather than the intensity of worry. The stressors were differentiated by task engagement, which declined under negative feedback, suggesting that it may have been demotivating. By contrast, task engagement was sustained in the high demand condition, consistent with evidence from other complex tasks that are sufficiently challenging to be motivating (Matthews, 2016).

Responses to stressors were less differentiated at the physiological level, with both eliciting increased power in highfrequency EEG bands. Both stressors also elevated HRV, a somewhat unexpected finding given that increased workload typically reduces this index. Phasic HRV increases may reflect emotion-regulation and successful engagement of cognitive inhibitory processes (Kemp et al., 2017). In the performance context, participants' efforts to focus on a demanding though challenging task may have encouraged inhibitory strategies. The stressors were differentiated by the CBFV response, which was lower in the negative feedback condition. Declining CBFV is typically a marker for loss of sustained attention and vigilance (Warm et al., 2012); it corresponds to the loss of task engagement also seen in this stressor condition.

Overall, the findings suggest that both manipulations induced substantial subjective stress, but not the classical sympathetic arousal response, given that there was no stressor effect on mean heart rate. Instead, the marked increase in high-frequency EEG power suggests a more "cognitive" expression of stress that may reflect performance concerns and, as suggested by the HRV responses (Kemp et al., 2017), efforts at cognitive stress-regulation. From the military perspective, stress of this kind may become increasingly significant as Warfighters shift from active combat roles to those that are remote from physical danger such as controlling unmanned vehicles and cyber operations. The greater differentiation of stressor impacts in the subjective data supports previous findings that physiological and subjective indices reflect distinct elements of the stress response, both of which add to evaluation of operator functioning (Matthews et al., 2017b).

# Associations Between Traits and Stress States: Baseline and Control Conditions

Previous studies found that trait worry predicts a range of stress outcomes (e.g., Roussis and Wells, 2006; Spada et al., 2008, 2010). Analyses of state data from the baseline and control conditions, confirmed that the AnTI predicted higher worry, even in the absence of an overt stressor. Individuals high in trait worry and meta-worry may be prone to anticipate that the task will pose a threat (O'Carroll and Fisher, 2013), and to focus their attention on threat concerns even in undemanding task conditions (Spada et al., 2008). The AnTI also predicted lower baseline task engagement. Grit and hardiness scales were generally more predictive than the AnTI of distress, and of task engagement in control conditions. Consistent with the TSO framework (Matthews et al., 2017a), multiple trait measures are required to define the individual's stress vulnerability.

Total scores on the hardiness and grit scales were both negatively associated with AnTI meta-worry. We cannot make casual inferences from cross-sectional data, but these associations are at least compatible with a role for dysfunctional metacognitions in undermining resilience. Hardiness and grit both support persistence in the face of adversity through active coping with obstacles to personal goals (Bartone et al., 1989; Duckworth and Quinn, 2009; Kelly et al., 2014). Effective coping may be more difficult if attention is directed toward self-referent worry and rumination (Hong, 2007). Indeed, a longitudinal study found that metacognitive style predicted subsequent anxiety, in a non-clinical sample (Ryum et al., 2017).

# Worry and Resilience Traits and Reactivity to Stressors

The study tested whether traits were associated with reactivity to the two stressors, over and above any general tendency toward higher levels of stress. Reactivity to stressors was assessed using residualized measures capturing the unique response to the stressor concerned. Consistent with the TSO framework

(Matthews et al., 2017a), cross-stressor correlations for reactivity measures were close to zero, and associations between traits and reactivity varied with the trait and with the outcome measure. In fact, there was a double dissociation between worry traits and grit, with the AnTI predicting only reactivity to negative feedback, and grit predicting only reactivity to high demand. The AnTI was selectively related to worry response, consistent with the S-REF model (Wells and Matthews, 2015). Overall state worry levels in task conditions were quite low, due to substantial cognitive workload directing attention outward to task stimuli (Matthews et al., 2013). Nevertheless, individuals high in the various facets of trait worry assessed by the AnTI appeared to be sensitive to state worry. The nature of worries activated in the two stressor conditions may have differed, with negative feedback eliciting self-referent concerns about personal competence, and the high demand condition activating concerns more directly related to task goals.

Hardiness correlated with reactivity to both stressors, but it was generally more predictive of response to negative feedback than to cognitive demand. Total hardiness was associated with attenuated distress and worry responses to the feedback manipulation, the study stressor more likely to promote selfevaluation. Hardiness is associated with styles of appraisal and coping that are adaptive in a performance setting (Eschleman et al., 2010; Cash and Gardner, 2011; Escolas et al., 2013) and are likely to suppress the CAS (Wells and Matthews, 2015). The challenge component of hardiness was the best predictor of reduced worry response, the primary symptom of CAS suppression. Thus, the capacity to embrace uncertainty over personal competence and see it as a positive experience may be the element of hardiness that counteracts the tendency for negative feedback to elicit the CAS, and points toward a need to further investigate metacognitive aspects of the trait.

Grit predicted higher task engagement and lower distress under high demand; the motivational qualities associated with grit may be especially important under these circumstances. Task engagement is associated both with intrinsic motivation and striving for performance excellence (Matthews et al., 2001). The role of grit in promoting task engagement is consistent with Lucas et al.'s (2015) finding that high grit participants persist with a difficult task even when they are failing. Grit also correlated with positive emotion and expectancies under these circumstances. Here, the more adaptive subjective state response to high workload experienced by those high in grit may be a consequence of self-regulative processes such as maintaining a sense of self-efficacy (Muenks et al., 2018) and adaptive management of task demands (Wolters and Hussain, 2015) that are especially well-suited for dealing with cognitive overload.

The traits were more weakly associated with physiological measures of stressor reactivity than with the subjective ones, but the negative feedback EEG data were notable for the consistent set of associations between higher AnTI scores and lower theta and higher gamma response. Theta and gamma may be functionally inter-related, based on evidence for crossphase coupling (Belluscio et al., 2012). Both frequency bands are influenced by emotion-regulation (Tolegenova et al., 2014), as well as by demanding cognitive processing (Ishii et al., 2014). A magnetoencephalography (MEG) study showed overlapping theta and gamma synchronization responses to emotional stimuli in multiple brain areas including amygdala and frontal cortex (Luo et al., 2014). Tentatively – and with due regard for the challenges of using EEG to infer brain processes – the data may signal individual differences in cognitive regulation of emotion. Higher gamma in higher-worry individuals is attributed to negative emotional arousal and anxiety (Headley and Paré, 2013), and disproportionate worrying (Oathes et al., 2008), whereas lower frontal theta indicates lower task-directed effort (Gevins and Smith, 2003), poorer working memory maintenance (Hsieh and Ranganath, 2014), and unsuccessful emotion-regulation (Ertl et al., 2013). Conversely, the high theta/low gamma pattern of the low AnTI scorer may reflect successful emotion-regulation that supports task-directed attention and mitigation of anxiety and worry. Frontal gamma desynchronization may also be associated with a mechanism for interrupting task-irrelevant cognitive activity (Ishii et al., 2014).

# Metacognition and the Functional Significance of Worry

Results thus far discussed suggest AnTI trait worry showed a distinctive pattern of associations with stress outcomes including generally higher state worry along with a more specific subjective and EEG response to negative feedback that may indicate poor emotion-regulation. However, these findings do not indicate a specific adaptive role for metacognition, i.e., meta-worry. The final set of analyses aimed to investigate the role of meta-worry in maladaptive stress outcomes by testing whether it moderated objective correlates of state worry.

A moderator effect of meta-worry was found for the number of waypoints used, but not for the two overall performance measures. Behaviorally, in high meta-worry persons, state worry appeared to reduce task-directed effort, i.e., setting simpler paths to avoid hazards. By contrast, those low in meta-worry seemed to try harder as they become more worried. For these individuals, the worry state may be adaptive in motivating adaptive and coping task effort, blocking development of the CAS (Wells and Matthews, 2015). However, in individuals with high meta-worry, which is a marker for negative beliefs about the uncontrollability and danger of the worry process (Wells, 2005), full CAS activation occurs as the individual diverts resources to mental self-regulation. Findings parallel Eysenck and Calvo's (1992) proposal that anxious individuals preserve processing effectiveness through compensatory effort. Berggren and Derakshan's (2013) review of the evidence for the hypothesis found mixed results. One explanation for inconsistency in findings is that the compensatory effort hypothesis is only valid for individuals low on dysfunctional metacognitions. Tentatively, compensatory effort might be impaired in high meta-worry because knowledge concerning control of attention is compromised and greater imminent threat is posed by cognition itself. A similar finding is evident in pathological worry, in which individuals with generalized anxiety disorder (GAD) report that worry is advantageous for coping and motivation, but it appears to become disruptive to functioning

and be a characteristic feature of GAD when meta-worry develops (Wells and Carter, 2001).

The physiological findings are consistent with this explanation. fNIR measures are indicative of task workload (Ayaz et al., 2012). On this metric, high meta-worry individuals show declining workload as state worry increases, implying reduced on-task effort. The concurrent increases in highfrequency EEG, including gamma, may be associated with the activation of the CAS (Wells, 2009), and self-focused attention as the person attempts to process the significance of their own worries.

More generally, the findings suggest a re-evaluation of the functional significance of worry in performance environments. Typically, worry is seen as a detrimental influence, as in classic studies of cognitive interference and test anxiety (Sarason et al., 1995). However, meta-analyses of the association between worry and measures of academic performance suggest that the effect size for the correlation is a modest −0.2 or so (e.g., Richardson et al., 2012). Studies of various attentional tasks have suggested that worry is typically a weaker correlate of poor performance than low task engagement and/or high stress (Matthews, 2016). The present findings support more nuanced accounts of worry (e.g., Sweeny and Dooley, 2017) that identify possible motivational benefits to the state. Similarly, studies of stress and skilled performance suggest some individuals are able to utilize stress symptoms as a motivator, for example, in sports (Matthews et al., in press).

The current study identifies metacognition as a critical determinant of the consequences of worry. A somewhat comparable moderator effect was obtained by Nordahl and Wells (2017), in a sample of socially anxious individuals. Metacognitive belief was the only one of several cognitive variables that uniquely predicted whether the person was working or not. Dysfunctional metacognitions may limit the person's capacity to function despite social anxiety. A similar study with a more diverse sample showed that metacognitive beliefs about the need for mental control predicted whether the person was working or on disability benefits, over and above trait anxiety and mental disorder (Nordahl and Wells, 2018). Dysfunctional metacognitions may limit the person's capacity to function despite social anxiety and other emotional conditions.

The present findings support the central proposition of S-REF theory that meta-cognitive dysfunction is a major driver of worry states (Wells and Matthews, 2015). Maladaptive beliefs about worries are an element of stable self-knowledge associated with personality that increases the likelihood of CAS activation. Metacognitions refer to both beliefs about processes, such as the importance of attending to intrusive thoughts, and to specific beliefs about thought contents (Wells, 1995), In challenging performance contexts, worry may indeed be elevated by processbased metacognitions, consistent with test anxiety research (Matthews et al., 1999). However, the distinctiveness of worry and metacognition as constructs is confirmed by the finding that the objective correlates of state worry vary with metacognitive style. The role of thought content in the performance context merits further investigation: for example, high meta-worry individuals may interpret thoughts of failure as actual failure.

# Limitations

The current study used a student sample asked to perform a complex task simulation following a relatively short training and practice period. Generalization of findings to samples of expert UAS operators is thus questionable. Greater skill and experience may attenuate stress response (Matthews et al., in press), but there is also more at stake in the real environment, which might elevate stress. Furthermore, operators face chronic stressors such as long work-shifts (Tvaryanas et al., 2006) that may moderate acute response. Lack of experience with the task may also have limited the validity of the performance measures. We observed substantial performance variability across participants; longer test sessions that allowed participants to develop a stable performance strategy would have been desirable. From a clinical perspective, relationships between personality and stress variables found in non-clinical samples will not necessarily generalize to clinical populations, given that relatively mild stress states may not represent severe clinical anxiety conditions well.

There are also issues related to stress assessment. To keep the data analysis tractable, we calculated responses averaged across each task condition, but there may have been considerable variation in stress within each condition. Further research might test the role of metacognitive style in response to discrete, high-stress events. The experiment was also not designed to investigate dynamic stress processes, such as changes in coping strategy within experimental conditions. The study exemplifies a multivariate assessment approach that specifies a profile of subjective and objective stress response across multiple measures (Matthews, 2016; Matthews and Reinerman-Jones, 2017). The differing sensitivities of the various measures justify the multivariate approach, but its application also multiplies the number of analyses and the risk of chance findings. The current study aimed to guard against this danger by using theory to guide data analysis, but replication of findings would be desirable. Conversely, more advanced analytic techniques could refine measures, such as spectral frequency analysis of the ECG to better separate sympathetic and parasympathetic response components (e.g., Kemp et al., 2017). On the predictor side, the AnTI metaworry scale assesses only a single aspect of metacognitive style, and there are further dimensions of metacognition that may moderate stress response (e.g., Wells and Cartwright-Hatton, 2004; Wells, 2005).

# CONCLUSION

The current study confirms that traits for worry, hardiness and grit predict stress response in a complex multi-UAS control environment. Findings support the central tenet of the TSO framework (Matthews et al., 2017a) that resilience is a multifaceted construct. The predictive validity of resilience and stress-vulnerability traits varied across stressors and across stress outcome measures. Within this broad framework, the role of the AnTI worry traits in predicting outcomes was consistent with the S-REF model (Wells and Matthews, 2015). Worry traits were more relevant to negative feedback than to cognitive demand, and they appeared primarily to influence state worry and EEG bands

that may reflect attempts at emotion-regulation. The S-REF model also predicted that the functional significance of worry states would vary with metacognition (meta-worry). Findings within the negative feedback stressor condition suggested that the maladaptive CAS may accompany worry states only when the person is disposed to dysfunctional metacognitions.

From an applied standpoint, the data support multifactorial assessments of populations required to perform complex or otherwise stressful tasks, including military populations. The various stressors prevalent in the UAS environment (Tvaryanas et al., 2006) may elicit qualitatively different stress responses, requiring different strategies for mitigation. Teaming situations in particular may involve negative evaluation from team-mates, especially when inexperienced teams are required to tackle difficult tasks that strain team cohesion. Current personnel selection emphasizes broad measures of negative affectivity such as neuroticism in the Five Factor Model (Huang et al., 2014), but more narrowly specified traits, including those for metacognitive dispositions, may improve predictive validity for performance under stress, especially if the trait can be matched to the stressor appropriately.

Profiling strengths and vulnerabilities may also allow training to be tailored to the individual to optimize resilience. For example, Wells (2000) Attention Training Technique (ATT) is a component of metacognitive therapy that is also effective for mitigating anxiety in non-clinical samples (Fergus and Wheless, 2018). ATT might help operators high in meta-worry manage evaluative stress. By contrast, interventions designed to enhance task motivation or strategy might be better suited to help operators lacking grit deal with high workloads.

# REFERENCES


# ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the University of Central Florida Internal Review Board; with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the University of Central Florida Internal Review Board.

# AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

# FUNDING

This research has been sponsored by Army Research Laboratory (HAVIC RAOS#74 – PROJECT# 64-01-6293). The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either express or implied, of ARL or the U.S. Government.

# ACKNOWLEDGMENTS

Some of the study findings were presented at the 59th Annual Meeting of the Human Factors and Ergonomics Society, Los Angeles, October 26–30, 2015.

physiological activation, and health. J. Psychosom. Res. 60, 113–124. doi: 10.1016/j.jpsychores.2005.06.074




**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Matthews, Panganiban, Wells, Wohleber and Reinerman-Jones. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Metacognitions Questionnaire and Its Derivatives in Children and Adolescents: A Systematic Review of Psychometric Properties

Samuel G. Myers <sup>1</sup> \*, Stian Solem<sup>2</sup> and Adrian Wells 3,4

<sup>1</sup> Division of Psychology, Bar Ilan University, Ramat-Gan, Israel, <sup>2</sup> Department of Psychology, Norwegian University of Science and Technology, Trondheim, Norway, <sup>3</sup> Division of Clinical and Health Psychology, The University of Manchester, Manchester, United Kingdom, <sup>4</sup> Greater Manchester Mental Health NHS Foundation Trust, Prestwich, United Kingdom

Background: The Metacognitions Questionnaire (MCQ) and its derivatives have been instrumental in research examining the Self-Regulatory Executive Function Model in adults. Studies testing whether findings are applicable to children and adolescents have been increasing and several different measures adapting the MCQ for younger populations have been developed. The current study aimed to systematically review the psychometric properties of MCQ measures or derivatives used in young people (aged 18 or less), to help assess current findings in this population and to guide future research in this growing area of investigation.

#### Edited by:

Gianluca Castelnuovo, Catholic University of the Sacred Heart, Italy

#### Reviewed by:

Silvia Casale, University of Florence, Italy José Muñiz, Universidad de Oviedo, Spain

\*Correspondence: Samuel G. Myers s.g.myers2013@gmail.com

#### Specialty section:

This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology

Received: 06 February 2019 Accepted: 30 July 2019 Published: 04 September 2019

#### Citation:

Myers SG, Solem S and Wells A (2019) The Metacognitions Questionnaire and Its Derivatives in Children and Adolescents: A Systematic Review of Psychometric Properties. Front. Psychol. 10:1871. doi: 10.3389/fpsyg.2019.01871 Method: Systematic searches were carried out on PubMed and PsycINFO of studies published up to June 2018. Additional studies were identified through Google Scholar and article references. Validity, reliability, range and responsiveness of measures were examined as well as analyses of age and gender differences on scores.

Results: Forty-five articles were identified. The total sample consisted of 7,803 children and adolescents (6,922 non-clinical, 881 clinical) aged 7–18. Studies used one of seven versions of the questionnaire, five adapted from the MCQ for younger populations: (1) The Metacognitions Questionnaire-Adolescent version; (2) The Metacognitions Questionnaire-Child version; (3) The Metacognitions Questionnaire-Child Version-Revised; (4) The Metacognitions Questionnaire-Child-30; and (5) The Metacognitions Questionnaire-65 Positive Beliefs Scale Revised; and two adult versions used without adaptation: (1) The Metacognitions Questionnaire-30 and (2) The Cognitive Self Consciousness Scale-Expanded. The validity and reliability of the Metacognitions Questionnaire-Adolescent version had the most extensive support. Other questionnaires had either mixed psychometrics or promising initial findings but more limited data.

Conclusions: It is recommended that studies using adolescents (age 12–18) consider using the Metacognitions Questionnaire-Adolescent version. Based on initial data, it is suggested studies using younger populations should consider the Metacognitions Questionnaire-Child-30 but further psychometric research into this and other measures is needed.

Keywords: metacognitions questionnaire, children, adolescents, review, psychometrics

# INTRODUCTION

# Rationale

The Metacognitions Questionnaire (MCQ-65; Cartwright-Hatton and Wells, 1997) is a 65 item measure that assesses metacognitive belief domains implicated in the Self-Regulatory Executive Function Model of psychological disorder (S-REF; Wells and Matthews, 1994, 1996; Wells, 2000). Metacognition refers to the beliefs, processes and strategies used when cognition is interpreted, monitored or controlled (Flavell, 1979). In the S-REF model, dysfunctional metacognitions lead to perseverative styles of thinking, biased attention, and ineffective self-regulation strategies (the Cognitive Attentional Syndrome; CAS, Wells, 2000) which is considered central to psychological disorder. The MCQ-65 has five subscales assessing the following metacognitions: (1) Positive beliefs about worry (PB), e.g., "Worrying helps me to avoid problems in the future," (2) Negative beliefs about the uncontrollability and danger of worry (NB), e.g., "My worrying is dangerous for me," (3) Beliefs about the need for control of thoughts (NFC), e.g., "It is bad to think certain thoughts," (4) Beliefs concerning cognitive competence (CC), e.g., "I have a poor memory," and (5) Cognitive self-consciousness (CSC), e.g., "I think a lot about my thoughts."

To facilitate ease of use, Wells and Cartwright-Hatton (2004) reduced the items of the MCQ and developed the Metacognitions Questionnaire-30 (MCQ-30), a 30 item version of the MCQ with the same factor structure as the original questionnaire, which has become the "gold standard" measure in adult research.

A large number of studies have used the MCQ-65 and MCQ-30 in adult populations. Findings for the MCQ-65 suggest acceptable psychometric properties of the scale (see Wells, 2009 for a review). However, most research has examined the shorter version-the MCQ-30. The five factors of the MCQ-30 have been replicated in several language versions in non-clinical populations (e.g., Spada et al., 2008; Yilmaz et al., 2008; Cho et al., 2012) as well as in populations with psychological disorders (Martín et al., 2014; Grøtte et al., 2016) and physical health difficulties (Cook et al., 2014; Fisher et al., 2016). Although most studies have examined single-order models consisting of the five subscales only, Fergus and Bardeen (2017) examined a bi-factor model consisting of the five subscales, and the total score as a general metacognitive factor, with results supporting this model.

Theoretically consistent positive relationships between MCQ subscales and a range of psychological disorders and symptoms have been shown cross-sectionally (e.g., obsessive-compulsive symptoms, Myers and Wells, 2005; problem drinking, Spada and Wells, 2005; trauma symptoms, Roussis and Wells, 2006; worry, e.g., Wells and Cartwright-Hatton, 2004; psychotic symptoms e.g., Bright et al., 2018) and prospectively (e.g., Sica et al., 2007; Yilmaz et al., 2011). These studies support the trans-diagnostic significance of metacognitive beliefs as proposed by the S-REF model and the convergent validity of the MCQ-30. The negative beliefs about uncontrollability and danger subscale has shown the strongest relationships with symptoms across studies (see e.g., Wells and Cartwright-Hatton, 2004; Spada et al., 2008; Bailey and Wells, 2013) supporting the central nature of this belief in metacognitive theory (see Wells, 2009). Both the MCQ-65 and the MCQ-30 have been shown to differentiate clinical and non-clinical participants, with a meta-analysis looking at both these measures together finding significantly higher scores in a range of clinical groups on all MCQ subscales, with the negative beliefs, and need for control subscales, showing the largest effects (Sun et al., 2017).

Metacognitive Therapy (MCT; Wells, 2000, 2009) is based on the S-REF model and focuses on modifying metacognitive beliefs and strategies. Results from a recent meta-analysis of MCT suggest it is a highly effective therapy for a range of psychological difficulties (Normann and Morina, 2018). Significant changes in the MCQ-30 have been demonstrated following metacognitive treatment (e.g., Wells et al., 2010, 2012). According to S-REF theory, decreases in symptoms following treatments should be mediated by changes in metacognition even when the treatment does not directly target these metacognitions. In support of this, several studies have demonstrated significant changes in MCQ scores following a range of effective non-metacognitive interventions (e.g., Solem et al., 2009; Fernie et al., 2016).

The MCQ has been instrumental in metacognitive research in the adult population. There has been far less research into metacognitive theory and therapy in child or adolescent populations. However, the development of the Metacognitions Questionnaire-Adolescent version (MCQ-A; Cartwright-Hatton et al., 2004) encouraged an increase in metacognitive research in adolescents. The MCQ-A is similar to the MCQ-30 but the wording of some items was modified slightly to make it easier for younger readers to understand. Additionally, the development of versions of the MCQ adapted for children, namely the Metacognitions Questionnaire for Children (MCQ-C30; Gerlach et al., 2008) and the Metacognitions Questionnaire-Child version (MCQ-C; Bacow et al., 2009) has supported metacognitive research in pre-adolescents. These questionnaires were adapted from the MCQ-A by simplifying words and phrases further to make them understandable to a younger age group. Recently the Metacognitions Questionnaire-Child Revised (MCQ-CR; White and Hudson, 2016) has been developed with the aim of making the questionnaire understandable to younger children (from age 7 to 8). Studies using young populations have also used the positive belief scale of the MCQ-65 adapted to be understandable to children (Meiser-Stedman et al., 2007) as well as adult versions of both the MCQ-30 and a measure derived from the cognitive self-consciousness subscale of the MCQ-30, the Cognitive Self Consciousness scale-Expanded (CSC-E; Janeck et al., 2003).

Results using these questionnaires in children and adolescents have been promising, particularly in showing relationships between the MCQ and a range of symptoms (e.g., Cartwright-Hatton et al., 2004; Debbané et al., 2009). A meta-analysis examining the relationships between metacognitive constructs, mainly assessed by MCQ-based measures, and anxiety, found low-medium to high effect sizes for the five factors and total MCQ score (Lønfeldt et al., 2017c). These results appear to support the application of S-REF theory to explaining anxiety and other psychological symptoms in younger populations. However, in assessing this literature it is important to consider the validity of the MCQ measures used in this population. There are several reasons why psychometric findings in adults cannot automatically be assumed to apply to younger populations and adaptations of the scale need to be assessed in children and adolescence populations:


It is also important to assess the psychometrics of these questionnaires used with younger populations because the multiple versions of the instrument present a challenge for future researchers in deciding which version to use for which age group of children and/or adolescents. A review and assessment of the psychometrics of these scales would provide information to help inform choices.

# Objective

The aim of the current study was to carry out a systematic review of the psychometric properties of the MCQ and derivatives in children and adolescent populations. It aimed to examine the validity, reliability and responsiveness of the measures. Additionally, it aimed to explore any age or gender differences in scores. Details of the psychometric dimensions assessed in this study are outlined below.

As the central aim of the current study was to evaluate psychometric parameters of the MCQs rather than test theory and because it was possible that the different versions of the scale may have substantive psychometric differences, we did not aim to carry out a meta-analysis of across-measure relationships between metacognitions and symptoms.

# Validity

Four sources of evidence for validity were examined (see Urbina, 2004; American Educational Research Association, American Psychological Association, & National Council on Measurement in Education, 2014): (1) evidence based on content (2) evidence based on factor structure (3) evidence based on relations with associated measures and (4) evidence based on relations with a criterion.

The current study aimed to assess two aspects of validity evidence related to content: (a) the extent to which the MCQ children or adolescent questionnaires cover the same dimensions of metacognition that the adult MCQ aims to measure, (b) the level of understandability of items to their target population.

Evidence for validity based on factor structure was assessed by examining (a) factor analyses of the measures, (b) whether there was measurement invariance based on gender or age. As the factors of the MCQ are based on theoretically central and distinct forms of metacognition in the S-REF model, and because metacognitions assessed by the MCQ may develop early (see Myers and Wells, 2015), it was hypothesized that the MCQ in children/adolescents would have a similar factorial structure to that found in adults, therefore representing the same set of latent constructs. For these reasons we also hypothesized that the MCQ was likely to be invariant across age, at least in studies that did not include very young children. Based on findings of invariance of the factor structure for men and women in two adult studies of the MCQ (Ramos-Cejudo et al., 2013; Fergus and Bardeen, 2017) we hypothesized that the MCQ in children/adolescents would also be invariant across gender.

Consistent with S-REF theory, the MCQ-30 has been shown to positively and significantly correlate with a range of symptoms in adults. The current study aimed to assess evidence of validity of the questionnaires in children and adolescents by examining the size and significance of correlations between the MCQ total score and subscales and validated symptom measures. Based on previous findings in adults, described earlier, we hypothesized that of the subscales, negative beliefs about uncontrollability and danger (NB) would have the strongest and most consistent relationships across symptom dimensions, with the other subscales also showing relationships but of a more specific nature and of lower magnitude.

One form of validity evidence based on relations with a criterion, is the ability to detect group differences (see Cronbach and Meehl, 1955). It was hypothesized that MCQ scores would be significantly higher in clinical than non-clinical groups. Results in adults (see meta-analyses by Sun et al., 2017) suggest these should exist for most subscales and across disorders but that the most consistent and strongest differences should be for NB and Need for Control (NFC) with moderate effects for Cognitive Confidence (CC) and Cognitive Self-Consciousness (CSC) and less strong and reliable effects for Positive Beliefs (PB).

# Reliability

Two forms of reliability were tested: (a) the internal consistency of the subscales and of the total score, (b) test-retest reliability, as a test of the stability of the measure over time.

# Distribution of Scores

We also examined whether the total score and subscales of the MCQ measures presented a range of scores, as a restricted range would impact on both validity and reliability of the measures.

### Responsiveness

Responsiveness refers to the ability of a measure to detect changes in the construct being measured. We aimed to assess whether the MCQ measures in children and adolescents changed following successful treatment. It was hypothesized that there would be some change on MCQ scores following any form of treatment which led to symptom changes but that decreases in MCQ scores would be particularly apparent following Metacognitive Therapy which directly targets metacognitions.

# Age Differences

Our study aimed to explore the presence of any age differences in the metacognitions measured by the MCQ within this population. General metacognitive skills and knowledge first develop in childhood (e.g., Schneider, 2008). Implicit metacognitions may already be present in 2 month old infants and some children as young as three can report on their metacognitions to some extent (Marulis et al., 2016), with a significant quantitative and qualitative increase in metacognitive skills between age 5 and 7 (Bryce and Whitebread, 2012). Metacognitive knowledge and some metacognitive skills continue to develop through adolescence (Schneider, 2008). Thus, the detection of metacognitions measured by the MCQ may vary depending on age.

#### Differences Between Sexes

The study also aimed to examine if there were any differences between sexes in MCQ scores. Studies in adults have produced somewhat inconsistent results with some studies finding no differences (Wells and Cartwright-Hatton, 2004; Grøtte et al., 2016) and others finding differences in some individual subscales (Spada et al., 2008; O'Carroll and Fisher, 2013). Fergus and Bardeen (2017) suggest that this inconsistency may be explained by the fact that any differences between sexes in MCQ scores that exist may be small. It was therefore hypothesized that there would be no consistent differences between sexes in scores on the MCQ in children and adolescents.

# METHODS

# Eligibility Criteria

Eligibility criteria for inclusion were:


Articles were excluded if they had an English abstract but the main text was not in English or they analyzed results for participants aged 18 (or younger) together with older participants.

# Search Strategy

Searches were carried out on PubMed and PsycINFO, using Boolean logic and the following keywords:

• Child OR adolescent OR adolescence

AND

• Metacognitions Questionnaire OR Metacognitions Questionnaire:

Additional searches were carried out on Google Scholar. References in identified articles were also examined for relevant articles.

# Data Extraction

The following information was extracted from all articles where present:


Where studies used the same or overlapping samples as previous studies, the results were only extracted when these were separate analyses to those reported previously.

Factor analysis results of both exploratory and confirmatory factor analysis were examined. All absolute and comparative fit indices reported were extracted apart from Chi-square because of its sensitivity to sample size (Bentler and Bonett, 1980). Studies reported one or more of the following fit indices: Absolute fit indices: Goodness of Fit Index (GFI); Adjusted Goodness of Fit Index (AGFI); Root Mean Square Error of Approximation (RMSEA); Root Mean Square Residual (RMSR). Comparative Fit indices: Normed Fit Indices (NFI); Relative Fit Index (RFI); Comparative Fit Index (CFI); Parsominous Fit Index (PFI). The following criteria were used to assess these fit indices. For the RMSEA 0.08 and less was considered adequate and 0.05 and less was considered good (see MacCallum et al., 1996). For the RMSR less than 0.08 is considered good (Hu and Bentler, 1999). For all other indexes 0.90 was considered adequate and 0.95 good (see Bentler and Bonett, 1980; Hu and Bentler, 1999; Kline, 2005).

When assessing Cronbach alpha scores and test-retest interclass correlations we used the guidelines given by the bib27 Review Model 2013: Cronbach alphas r < 0.70 = Inadequate 0.70 ≤ r < 0.80 = Adequate, 0.80 ≤ r < 0.90 = Good, r ≥ 0.90 = Excellent; Test-retest r < 0.60 = Inadequate, 0.60 ≤ r < 0.70 = Adequate, 0.70 ≤ r < 0.80 = Good, r ≥ 0.80 = Excellent.

For tests of validity based on associated measures, we included any correlations reported between symptom measures which were based on child-report and had been validated in at least one prior study, and the MCQ measures. We did not include the few correlations reported between an MCQ measure and a measure of child symptoms as reported by parents as evidence suggests significant disparity between child and parent reports of symptoms (De Los Reyes and Kazdin, 2005; Canavera et al., 2009). As concurrent validity rather than specificity was being tested, we did not include correlations or regressions that controlled for other symptoms e.g., correlations between the MCQ and anxiety controlling for worry.

For tests of validity based on relations with a criterion, where differences between clinical and non-clinical groups were reported as significant we calculated effect sizes based on the means, standard deviations and number of participants, using RevMan Software.

The assessment of effect sizes was based on Cohen (1988), for correlations (r), 0.1 = small, 0.3 = medium and 0.5 = large; for differences between means, Cohen's d, 0.2 = small 0.5 = medium and 0.8 = large.

When assessing the effects of treatments or interventions on MCQ scores we also examined the effectiveness of the intervention on primary outcome measures, as decreases in metacognition would only be expected following a successful intervention.

Results of psychometrics are presented based on the suggested order of evaluating measurement properties suggested by the COSMIN methodology (Prinsen et al., 2018). Validity evidence based on content was assessed first as initially it is important to assess whether a measure is comprehensive and comprehensible (Prinsen et al., 2018). Then the internal structure was examined by assessing validity based on factor structure and internal consistency. Then other reliability and validity evidence were

assessed followed by responsiveness. Age and gender analyses of differences in scores were exploratory and were examined last.

# Quality Assessment of Studies

The methodological quality of studies was assessed on the following criteria, based on a modified version of the Newcastle-Ottowa scale for cross-sectional studies (Herzog et al., 2013): (1) Research question and design; (2) Sampling method; (3) Sample size; (4) Data collection; (5) Method of dealing with missing data; (6) Analysis (**Appendix** with scoring system). The maximum possible score if all criteria were met was 8. Two of the authors independently marked the studies and any differences in scores were discussed and resolved.

# RESULTS

# Search Results and Study Characteristics

A PRISMA diagram (Moher et al., 2009) of the search results and study selection process is presented in **Figure 1**.

As shown, 105 articles were produced by the literature searches. Of these, eleven were duplicates. Ninety-four articles were screened, 33 were rejected at the screening stage as examination of title and/or abstract showed they either clearly included participants over the age of 18 or were not peerreviewed articles. Of the remaining 61, 16 were excluded as examination of the full text showed they either: (1) included participants over 18, (2) did not use an MCQ measure, (3) did not have data on the MCQ or (4) were not in English. Forty-five articles met the inclusion criteria and were included in the review. These articles consisted of 34 separate groups of participants. Descriptions of the methodologies, a score for the quality of the studies as well as a summary of psychometric and other findings for the 45 articles are shown in **Table 1**.

In total there were at least 7,803 separate participants in the studies. Of these 6,922 were non-clinical and 881 were clinical. Ages ranged from 7 to 18.

# Metacognitions Questionnaires Used

One of seven versions of the Metacognitions Questionnaire or a subscale or subscales of the MCQ were used:


The MCQ-A is a 30 item measure, based on the MCQ-30 with the wording simplified slightly with the aim of making it more understandable to adolescents. Like the MCQ-30 each item is scored on a scale of 1 (do not agree) to 4 (agree very much). Therefore, the possible range of scores for the total scale is 30– 120, and for each subscale 6–30. It was used in 12 articles in this review, consisting of 11 separate population samples. Age range across studies was 7–18, although nine out of these 11 samples used adolescents of 12–18 years, the age group the questionnaire was originally devised for. English, French, Dutch, and Farsi versions of the MCQ-A were used in the studies.

The MCQ-C is a 24 item measure, based on the MCQ-A but with the wording further simplified with the goal of making it understandable to children as young as 7. An important difference between the MCQ-C and other versions of the MCQ is that the developers omitted the six items making up the cognitive confidence subscale. They justified omitting it based on a study that suggests that this scale in adults may be made up of different factors (Hermans et al., 2008) and they argued that it should be omitted until this was clarified. The removal of this subscale means the possible range of scores for the total scale of the MCQ-C is 24–90. It was used in 19 articles in the review, made up of 17 different samples, with an age range across the studies of 7– 17. English, Turkish, Italian and Serbian versions of the MCQ-C were used.

The MCQ-CR is a 30 item measure. It was developed after Smith and Hudson (2013) tested the understandability of the MCQ-C in fourteen 7–8 year olds and found that a significant proportion of children did not understand six items. The MCQ-CR consists of 12 items from the MCQ-C without adaptation, as well as 12 more items taken from the MCQ-C and simplified further to be understandable to 7 and 8 year olds. The MCQ-CR reverted to the five-factor model of the MCQ and also included the six items of the cognitive confidence subscale, modified to make them understandable to children aged 7–8. The MCQ-CR adds an option for each item of indicating that the participant does not understand the item. The MCQ-CR was used by one study in the review (age range 7–12) and an English version was used.

The MCQ-C30 was based on the MCQ-A but with the wording simplified further to be understandable to children. Unlike the MCQ-C it retained the five-factor structure of the MCQ. It was used by eight studies in the review, consisting of five separate samples, age ranged from 7 to 17. The MCQ-C30 was originally developed in German, studies in the review used Danish or English versions of the questionnaire.

The MCQ-30 is the version developed in adults and is described earlier. It was used in two studies in the review without adapting it for younger participants, these studies had separate samples, ages in the two studies together ranged from 12 to 18. Both studies used English versions of the questionnaire.

The MCQ-PBR is a 19 item measure that consists of the positive beliefs about worry scale from the MCQ-65 with 10 items adapted to make them understandable to children. It was used by two studies with overlapping samples, age range 10–16. Both studies used English versions.

The CSC-E consist of 14 items and is an expanded version of the cognitive-consciousness scale of the MCQ-65. It was developed using an adult population (Janeck et al., 2003) but the

Myers et al.

#### TABLE 1 | Study methodology, quality score, psychometrics and main findings relevant to the review.


(Continued)

MCQ Psychometric Review in Youth


(Continued)

MCQ Psychometric Review in Youth


(Continued)


Myers et al.

(Continued)


(Continued)

MCQ Psychometric Review in Youth


MCQ Psychometric Review in Youth

ADIS-IV-C/P, Anxiety Disorders Interview Schedule-Child/Parents Version; CAWS-Worry, Child and Adolescent Worry Scale; CDI, Children's Depression Inventory-Short Form; CY-BOCS, Children's Yale-Brown Obsessive-Compulsive Scale; LOI-CV, Leyton Obsessional Inventory–Child Version; MASC, Multidimensional Anxiety Scale for Children; MCQ, Metacognitions Questionnaire, -A Adolescent version, -C Child Version, -C30 Child-30; MCQ Subscales; PB, Positive beliefs about worry; NB, Negative beliefs about worry; CC, Cognitive Confidence; NFC, Need for control; CSC, Cognitive self-consciousness; CR, Child Revised; MOCI, Maudsley Obsessive-Compulsive Inventory; Penn State Worry Questionnaire for Children (PSWQ-C); PSS-SR, Post Traumatic Stress Disorder Symptom Scale Self-Report; RCADS, Revised Children's Anxiety and Depression Scale; RCMAS, Revised Children's Manifest Anxiety Scale; SCARED, Screen for Child Anxiety Related Disorders -r: Revised; SCAS, Spence Children's Anxiety Scale; SPQ, Schizotypal Personality Questionnaire (SPQ); SDQ, Strength and Difficulties Questionnaire; SPAI, Social Phobia and Anxiety Inventory; STAI, State Trait Anxiety Inventory, -C Child version, -T Trait version; RIES-C, Revised Impact of Event Scale-Child Version.

English version of the CSC-E was used without adaptation by one study in the review with adolescents, ages 15–17.

# Symptoms Measured

Results extracted for tests of concurrent validity examined relationships between the MCQ measures and worry, anxiety, obsessive-compulsive symptoms, depression, post-traumatic symptoms, general emotional difficulties, psychotic symptoms and dissociation. Symptom measures for individual studies are given in **Table 1**.

# Assessment of Study Quality

Scores on the quality assessment scale (maximum possible 8) ranged from 2 to 7 with a mean of 5.13. All studies had clear research questions and appropriate design. Most studies used validated symptom measures and used appropriate analyses which were described appropriately. Studies varied as to the amount of possible bias in their sampling method with the strongest studies attempting to make their samples representative, by for example using schools in locations evenly spread across a country. Studies were marked down on sampling method if samples were clearly not representative or were at risk of not being representative e.g., using individual schools without discussing how representative these schools were. Few studies carried out power calculations. The adequacy of sample size was assessed by power calculations we made based on parallel adult studies, and studies varied as to whether they had adequate power according to these criteria. Missing data was only reported and addressed in a minority of studies.

# Metacognitions Questionnaire-Adolescent Version (MCQ-A)

# Validity Evidence Based on Content

**Comprehensiveness**

The MCQ-A includes all items of the MCQ-30 with wording slightly simplified.

# **Understandability**

Cartwright-Hatton et al. (2004) report that the MCQ-A has a Flesch-Kincaid Reading Grade Level of 3.6. This means the questionnaire should be understandable to most children aged 9 and up. Beyond this no other data is available regarding its understandability to younger populations.

# Validity Evidence Based on Factor Structure

Three studies examined the factor structure of the MCQ-A. In their validation study, Cartwright-Hatton et al. (2004) carried out an exploratory factor analysis on a non-clinical sample (n = 158). They reported that a five-factor solution was chosen based on the Scree test and including only factors with Eigen values above one. The factors and item loadings corresponded closely with the adult MCQ-30, although goodness of fit indices were not reported. The Confirmatory Factor Analysis (CFA) of the MCQ-A carried out by Ellis and Hudson (2011) using a mixed clinical and nonclinical sample (total n = 114) found an adequate or good fit on four out of five fit indices: GFI = 0.96, AGFI = 0.95, NFI = 0.94, RFI = 0.94, PNFI = 0.86. Wolters et al. (2012) using a nonclinical sample (n = 317) found the five-factor structure had an adequate or good fit on all fit indices GFI = 0.95, AGFI = 0.94, NFI = 0.91, RFI = 0.90. Additionally, Wolters et al. found that a second-order model consisting of a higher-order factor (total score) and five lower-order factors (subscales) had an acceptable or good fit on most criteria. GFI = 0.94, AGFI = 0.93, NFI = 0.90, with the RFI of 0.89 just outside the criteria for acceptable fit. No studies examined measurement invariance across gender or age.

## Internal Consistency

Internal consistency was examined in seven studies without overlapping participants (Cartwright-Hatton et al., 2004; Matthews et al., 2007; Ellis and Hudson, 2011; Wilson et al., 2011; Farrell et al., 2012; Wolters et al., 2012; Mazloom et al., 2016). Five of these studies examined the Cronbach alphas of the total score and subscales and a further two only examined total scores. Cronbach alphas were adequate to excellent for the total score and all subscales apart from NFC (range 0.70– 0.92). Results for the NFC were mixed, in three samples they were below the 0.7 threshold of adequacy (range 0.57–0.66) but Cronbach alphas were adequate in three other samples (range 0.70–0.77).

# Test-Retest Reliability

Test-retest analysis was examined for the total score and subscales in three samples, in two papers (Cartwright-Hatton et al., 2004: 2 weeks test-retest; Wolters et al., 2012-Non-clinical sample, 7–21 weeks test-retest; OCD sample, 6–12 weeks testretest). Results of intraclass correlations were mostly good to excellent (range 0.72–0.95) apart from poor reliability for the NB subscale (0.24) and the total score (0.34) in the former study and NFC (0.35) in the non-clinical sub-sample of the latter study.

# Ranges

Three studies reported ranges for the MCQ-A (Cartwright-Hatton et al., 2004 [total score range only]; Matthews et al., 2007 and Wolters et al., 2012 [for clinicals and non-clinical participants separately]). Across-study ranges for the total score in non-clinical participants were 30–116. For subscales (across two studies) non-clinical ranges were PB 6–24, NB 6–24, CC 6– 22, NFC 6–20, CSC 6–24. In the one study (Wolters et al., 2012) that reported ranges for clinical participants, for the total score, the range was 36–104, for subscales: PB 6–24, NB 6–22, CC 6–22, NFC 6–22, and CSC 7–24. Results suggested the measure picked up a broad range of MCQ scores.

## Validity Evidence Based on Relations With Associated Measures

Nine studies with non-overlapping samples examined correlations between the MCQ-A and a range of psychological symptom measures. Results are shown in **Table 2**.

As shown the Total Score and the NB subscale significantly related to a range of symptoms in all analyses, with effect sizes ranging from medium to high. The other subscales related significantly to symptoms in most but not all analyses. Effect sizes for PB, CC, and CSC ranged from low to medium and for NFC from low to high.

TABLE 2 | Across-study correlations between MCQ-A and symptom measures.


#### \*p < 0.05.

Number of samples with correlations between symptoms and MCQ-A Total Score (TS) and subscales—Obsessive-Compulsive symptoms: TS, five samples (Cartwright-Hatton et al., 2004; Matthews et al., 2007; Crye et al., 2010; Farrell et al., 2012; Wilson and Hall, 2012) subscales, three samples (Cartwright-Hatton et al., 2004; Matthews et al., 2007; Wilson and Hall, 2012); Anxiety: TS, three samples (Cartwright-Hatton et al., 2004; Wolters et al., 2012 - non-clinical group; Wolters et al., 2012 - OCD group) subscales, four samples (Cartwright-Hatton et al., 2004; Wolters et al., 2012 - non-clinical group; Wolters et al., 2012 - OCD group; Wilson and Hall, 2012); Depression: TS and subscales, three samples (Cartwright-Hatton et al., 2004; Wolters et al., 2012 - non-clinical group; Wolters et al., 2012 - OCD group); Worry: one sample (Wilson et al., 2011); Post-traumatic symptoms: one sample (Mazloom et al., 2016); Psychotic symptoms: one sample (Debbané et al., 2009 -controlling for age and IQ).

#### Validity Evidence Based on Relations With a Criterion

Three studies (Cartwright-Hatton et al., 2004; Ellis and Hudson, 2011; Wolters et al., 2012) examined differences between the MCQ-A total score and subscales in clinical and non-clinical groups. The clinical groups consisted of people with "an emotional disorder" not specified (Cartwright-Hatton et al., 2004), anxiety disorders or anxiety disorders with comorbid depression (Ellis and Hudson, 2011) and Obsessive-Compulsive Disorder (Wolters et al., 2012). In all three studies the total score (range d = 1.06 to 1.49) and the NB (range d = 1.54 to 2.41) and NFC (range d = 0.61 to 1.00) subscales were significantly higher in clinical groups than control groups-effect sizes high for total score and NB, medium to high for NFC. PB was higher in two out of three studies (range d = 0.19 to 0.67) while CSC (range d = 0.31 to 0.78) and CC (range d = 0.20 to 1.03) were both higher in one out of three studies.

### Responsiveness

One study (Sanger and Dorjee, 2016) reported on changes in MCQ-A scores following an intervention, which consisted of a course of mindfulness training in non-clinical adolescents. The intervention was successful in leading to significantly increased response inhibition as shown by increased N2 negativity response to an attention task measured by an EEG. However, it did not lead to predicted changes on P300 mean amplitude (measures of attention efficiency). Significant pre to post differences were found on the total score of the MCQ-A (d = 0.64, medium effect size) as well as on NFC (d = 1.15, large effect size) compared to a control group, but not on the other subscales.

### Age

Four articles tested whether there were within-study relationships between age of participants and the MCQ-A total score and subscales: Matthews et al. (2007) in a non-clinical sample, age range 13–16, found that the MCQ-A total score, as well as the NB, NFC, and CSC subscales significantly negatively correlated with age, although correlations were low (range −16 to −19). Wilson et al. (2011) using a non-clinical sample, age range 11–16, and Ellis and Hudson (2011) using a mixed non-clinical and clinical sample, age range 12–17, found no significant correlations between age and the total score or subscales. Wolters et al. (2012), age range 12–18, also found no relationship between the total score or subscales and age in their clinical sample. In their non-clinical sample, there was a small positive relationship between the MCQ-A total scale and age, r = 0.12.

### Differences Between Sexes

Three studies examined differences between sexes in MCQ-A scores. Matthews et al. (2007) and Wilson et al. (2011) using the MCQ-A total score and subscales, Crye et al. (2010) using the MCQ-A total score only. No significant differences were found on any scores.

# Metacognitions Questionnaire-Child Version (MCQ-C)

# Validity Evidence Based on Content

#### **Comprehensiveness**

The MCQ-C does not include the six items designed to assess cognitive confidence in the MCQ-30. Bacow et al. (2009) justified omitting it based on the fact that this scale in adults may be made up of different factors (Hermans et al., 2008) and they argued that it should be omitted until this was clarified. Thus, one factor assessed in the MCQ-30 is not assessed in the MCQ-C.

### **Understandability**

Bacow et al. (2009) report that the MCQ-C has a Flesch-Kincaid Reading Grade Level of two-meaning it should generally be understandable to children ages 7–8. However, Smith and Hudson (2013) tested the understandability of the questionnaire in a sample of fourteen 7–8 year olds and found that a significant proportion of these children did not understand six items on the MCQ-C. Additionally, White and Hudson (2016) reported that six further items were assessed as being above Grade 2 level according to Fry's (1977) criteria.

# Validity Evidence Based on Factor Structure

Three studies (Bacow et al., 2009; Irak, 2012; Stevanovic et al., 2016) examined the factor structure of the MCQ-C. Both Bacow et al. in a combined clinical and non-clinical sample (n = 98) and Irak using a large non-clinical sample (n = 470), carried out a confirmatory factor analysis of the four-factor structure of the MCQ-C. Results of fit indices in Bacow et al.'s were mixed with an RMSEA of 0.077 suggesting an adequate fit but a CFI of 0.85 suggesting a poor fit. Irak's fit indices were good for RMSEA = 0.05 and RMR = 0.08, adequate for GFI = 0.90 and just marginally below adequate for CFI = 0.89 and AGFI = 0.88. Stenanovic et al. split their sample (n = 467) into two, with both of these samples having mixed clinical and non-clinical participants. They first carried out an exploratory factor analysis of the MCQ-C on one part of the sample (n = 233). This resulted in a three rather than four factor structure, made up of 16 items in total described as: (1) Cognitive monitoring, (2) Specific positive worry beliefs, and (3) General positive worry beliefs. In a subsequent CFA using the second part of their sample (n = 234) testing this three-factor structure, an adequate fit was obtained when three items of one of the scales were removed. No studies reported examining measurement invariance based on gender or age.

# Internal Consistency

Eleven articles (Bacow et al., 2009; Irak, 2012; Smith and Hudson, 2013; Benedetto et al., 2014; Holmes et al., 2014; Carr and Szabó, 2015; Donovan et al., 2017; Francis et al., 2017, 2018; Hearn et al., 2017a), representing 10 separate samples reported Cronbach alphas for the MCQ-C (two studies Francis et al., 2017 and Francis et al., 2018 reported Cronbach alphas for different parts of the scale in the same sample). Internal reliability of the MCQ-C total score was adequate to good in the three studies that reported it (range 0.73 to 0.87). Scores on subscales varied depending on the study PB (nine studies) range 0.46 to 0.86, NB (eight studies) range 0.60 to 0.78, NFC (three studies) 0.25 to 0.64, CSC (three studies) 0.61i to 0.75.

# Test-Retest Reliability

Test-retest reliability, over a 3 week period, reported only by Irak was good to excellent for all subscales and the total score (range 0.76 to 0.82).

# Ranges

Two studies with the same sample gave ranges for the MCQ-C (Francis et al., 2017, total score; Francis et al., 2018, PB and NB subscales). Ranges were broad: Total score ranged from 26 to 79, PB 6 to 22 and NB 6 to 24.

# Validity Evidence Based on Relations With Associated Measures

Eleven studies with non-overlapping samples examined correlations between the MCQ-C and a range of psychological symptom measures. Results are shown in **Table 3**.

As shown the Total Score and the NB subscale significantly related to a range of symptoms in all analyses. Effect sizes for the total score ranged from low-medium to high and for NB from medium to high. The other subscales related significantly to symptoms in most but not all analyses with effect sizes ranging from low to medium.

Validity Evidence Based on Relations With a Criterion Four studies examined differences between clinical and nonclinical populations: [Bacow et al., 2009; Smith and Hudson, 2013 using clinical groups with anxiety disorders; Donovan et al. (2016) using a group with Generalized Anxiety Disorder (GAD), and Hearn et al. (2017b) using a group with Social Anxiety Disorder, the GAD comparison in Hearn et al. was not included as the GAD sample overlapped with Donovan et al. (2016)]. Smith and Hudson found significantly higher scores in the clinical than the non-clinical group for the total score (d = 0.69), PB (d = 0.45) (medium effect sizes), and NB (d = 0.87; large effect size). However, NFC and CC did not distinguish the groups. Donovan et al. and Hearn et al. examined only NB and PB, both found that NB (ds of 1.72 and 1.15; both large effects) but not PB (ds of 0.52 and 0.25) was significantly higher in the clinical group than a non-clinical control. Bacow et al. (2009), with worry content controlled, found no significant differences between a clinical and non-clinical group on the total score or subscales beyond significantly higher scores on CSC in the non-clinical group. Of note in this study was that 60% of the non-clinical group had sub-clinical symptoms.

# Responsiveness

Holmes et al. (2014) in a trial treating GAD using Cognitive Behavior Therapy (CBT), in children aged 7–12, examined changes on only the NB and PB subscales of the MCQ-C. Examination of primary outcome measures showed the treatment was successful in reducing diagnostic GAD status and severity of disorder post-treatment compared to a control group, as well as leading a larger increase in overall functioning. They found a significant decrease in NB but not PB from pre-treatment to both post-treatment and 3 month follow up, effect size not reported. However, the decrease in NB was not significantly different from the decrease seen in a wait list control, assessed only at post-treatment.

Hearn et al. (2018) in a trial of CBT with patients with Social Anxiety Disorder (age of participants 8–17) also examined changes on only the NB and PB of the MCQ-C. They examined scores on measures at 12 week assessment when some but not all participants had completed treatment and at 6 month followup. At 12 week assessment there were significant reductions in diagnostic severity and social anxiety symptoms in the treatment group compared to the wait-list control. However, there were no significant difference on diagnostic status. They found no significant reductions at 12 week assessment on NB and PB. However, there were significant reductions from pre-treatment to 6 month follow-up in both NB and PB.

# Age

Three articles examined relationships between scores on the MCQ-C and age. Bacow et al. (2009), age range 7–17, examined age differences in their clinical group only, due to the small sample size of their non-clinical group. Of the four subscales TABLE 3 | Across-study correlations between MCQ-C and symptom measures.


#### \*p < 0.05.

Number of samples with correlations between symptoms and MCQ-C Total Score (TS) and subscales—Obsessive-Compulsive symptoms: TS and subscales, two samples (Irak, 2012; Boysan et al., 2016); Anxiety: TS, three samples (Irak, 2012; Kadak et al., 2013; Smith and Hudson, 2013) PB and NB four samples (Irak, 2012; Smith and Hudson, 2013; Benedetto et al., 2014; Hearn et al., 2017a) NFC and CSC, three samples (Irak, 2012; Smith and Hudson, 2013; Benedetto et al., 2014); Depression: TS, two samples (Bacow et al., 2009; Kadak et al., 2013) subscales, one sample (Bacow et al., 2009); Worry: TS, one sample (Bacow et al., 2009) PB, six samples (Hearn et al., 2017a; Bacow et al., 2009; Kertz and Woodruff-Borden, 2013; Carr and Szabó, 2015; Donovan et al., 2017; Francis et al., 2018, controlling for recruitment site) NB, five samples (Bacow et al., 2009; Kertz and Woodruff-Borden, 2013; Donovan et al., 2017; Hearn et al., 2017a; Francis et al., 2018, controlling for recruitment site) NFC and CSC, one sample (Bacow et al., 2009); Post-traumatic symptoms: one sample (Kadak et al., 2013); Dissociation: one sample (Kadak et al., 2013); Emotional Difficulties: one sample (Smith and Hudson, 2013).

and total score of the MCQ-C the only significant relationship was a positive relationship between CSC and age (only the unstandardized regression coefficient is reported (0.46). Irak (2012) split his sample into children (age 8–12) and adolescents (13–17). There was a significant difference between the groups on PB scores only, with the older group scoring higher. Carr and Szabó (2015) in a non-clinical sample, age range 7–12, examined only PB and found no relationship between this subscale and age.

#### Differences Between Sexes

Three studies examined differences in MCQ-C scores (Benedetto et al. using the MCQ-C subscales, Francis et al., 2018 using just the PB and NB subscales of the MCQ-C, and Irak, 2012, using the MCQ-C total score and subscales). Benedetto et al. and Francis et al. found no differences between scores of males and females. Irak found that females scored significantly higher than males on negative beliefs about worry, and the total score only.

#### Age X Gender Interaction

Two studies examined the interaction between age and gender; both studies used the MCQ-C total score and subscales. bib4 (2009; age range 7–17) found that for younger participants (1 SD below mean age) there were no gender differences on MCQ-C subscales or total score. However, in adolescents (1 SD above mean age) girls scored higher than boys on the MCQ-C total score only. bib44 (2012; age range 8–17) found no interaction effect between age and gender on the total score or subscales.

# Metacognitions Questionnaire-Child 30 (MCQ-C30)

#### Validity Evidence Based on Content

#### **Comprehensiveness**

The MCQ-C30 includes all items of the MCQ-30 with wording simplified to be understandable to children.

#### **Understandability**

No data on reading level or understandability was presented for this measure.

## Validity Evidence Based on Factor Structure

One study (Esbjørn et al., 2013) examined the factor structure of the MCQ-C30. This study carried out a CFA examining the fit of a two-level model with the higher-order factor consisting of the total score and the five subscales making up lower-order factors. They also included gender as a predictor of the total score. In their full non-clinical sample (n = 974) fit indices for the model were acceptable or good: CFI = 0.94, TLI = 0.93, RMSEA = 0.039. They subsequently carried out two CFAs on their sample split by age. This is a test of measurement invariance across age. Results for 13–17 year olds (n = 420) indicated an adequate fit to the two-level model: CFI = 0.90, TLI = 0.90, RMSEA = 0.06, while results for the 9–12 year olds (n = 554) were acceptable on one measure: RMSEA = 0.06, but marginally short on two others: CFI = 0.87, TLI = 0.86. Tests comparing the model fit of the two age groups showed no significant differences between them.

#### Internal Consistency

Five studies with non-overlapping samples (Esbjørn et al., 2013, 2015; Campbell et al., 2018 only results for clinical sample included for this study as non-clinical sample overlapped with another study Esbjørn et al., 2016, 2018) reported Cronbach alphas. Cronbach alphas for the total score ranged from just below the adequate cut-off to excellent: range 0.69–0.91. Subscale scores were somewhat mixed: PB 0.64–0.87, NB 0.65–0.78, CC 0.66–0.82, CSC 0.62–0.75<sup>1</sup> . NFC generally showed the weakest Cronbach alphas, with scores ranging from 0.59 to 0.68.

### Test-Retest Reliability

No studies using this measure reported test-retest results.

#### Ranges

Ranges for the MCQ-C30 were not presented in any study.

<sup>1</sup> In one study (Esbjørn et al., 2016) ranges of Cronbach alphas were given across most subscales including the CSC together rather than separately, this score is the lowest score mentioned in that study.

# Validity Evidence Based on Relations With Associated Measures

Three studies with non-overlapping samples examined correlations between the MCQ-C30 and psychological symptom measures. One of these studies (Campbell et al., 2018) had a very small sample (n = 23) which meant even some medium effect sizes were not significant in this study. Studies using both the Total score and subscales of the MCQ-C30 examined relationships with anxiety, depression and worry.

Results are shown in **Table 4**.

As shown the Total Score significantly related to symptoms in all analyses with effect sized ranging from medium to high. NB and NFC significantly related to anxiety and worry but not depression with all effect sizes ranging from medium to high. PB significantly related to symptoms in three out of four analyses and CC and CSC in two out of four with effect sizes for these subscales ranging from low to medium.

## Validity Evidence Based on Relations With a Criterion

Only one study examined differences between clinical and nonclinical groups with Esbjørn et al. (2015) finding that a group with Generalized Anxiety Disorder scored significantly higher on all subscales apart from CSC: PB (d = 0.70), NB (d = 1.58), CC (d = 0.69), and NFC (d = 1.07; range of effect sizes for significantly different scores medium to large) and that an Anxiety Disorder group scored significantly higher on NB (d = 1.15) and NFC (d = 0.87) than a non-clinical group (both large effect sizes).

# Responsiveness

Two studies examined changes in MCQ-C30 scores following treatment. Esbjørn et al. (2018) in a trial of MCT for GAD in participants aged 7–13 found 86.4% were free of GAD and 72.7% were free of all anxiety disorders post-treatment, at 6 month follow-up figures were 75 and 65.9%, respectively. The total score of the MCQ-C30 and all subscales apart from CC, were significantly reduced from pre to post-treatment, and all reductions remained significant at 6 months follow-up apart from PB. The effect-size, reported for the total score only, was large both from pre to post treatment, d = 0.84, and from pre-treatment to follow up, d = 1.08.

TABLE 4 | Across-study correlations between MCQ-C30 and symptom measures.


\*p < 0.05.

a (Campbell et al., 2018). b (Esbjørn et al., 2013). c (Esbjørn et al., 2016).

### Normann et al. (2016) in a trial of CBT for patients aged 7–12, with several anxiety disorders, examined changes on the total score of the MCQ-C30. The treatment successfully reduced anxiety symptoms from pre-treatment to post-treatment (medium effect) and pre-treatment to follow-up (large effect). The MCQ-C30 Total Score changed significantly from pretreatment to post-treatment d = 0.55 (a medium effect size), and from pre-treatment to follow-up d = 0.87 (large effect size). There was a significant decrease from post-treatment to follow-up.

## Age

One study (Lønfeldt et al., 2017b; age range 9–17) using the MCQ-C30 examined relationships between the total score and subscales and age. They found small but significant negative relationships between the MCQ-Total score (−0.08) as well as NB (−0.08) and NFC (−0.10) and age, for other subscales the relationship was not significant.

## Differences Between Sexes

Two studies examined differences on the MCQ-C30. Esbjørn et al. (2013) found a small but significant correlation between gender and the total score (subscales not examined), with girls scoring higher, but this difference was made non-significant when anxiety was controlled for Lønfeldt et al. (2017a) found that NB but not other scales or the total score were significantly higher in girls than boys.

# Metacognitions Questionnaire-Child Revised (MCQ-CR)

This questionnaire has only been examined in its validation study (White and Hudson, 2016) results are outlined below.

# Validity Evidence Based on Content

# **Comprehensiveness**

The MCQ-CR includes all items of the MCQ-30 with wording simplified to be understandable to children as young as 7–8.

# **Understandability**

The MCQ-CR includes the possibility of indicating "I don't understand" for each item Examination of responses suggested 75% of 7 year olds and 83% of 8 year olds understood all items on the MCQ-CR. However, there was a negative correlation (r = −0.23) between number of items filled in as "I don't understand" and age, indicating that understanding increased with age. A t-test comparing 7–8 year olds with 9–12 year olds found significantly greater lack of understanding in the younger group. For other analyses in the White and Hudson (2016) study items scored as "I don't understand" were treated as missing data. If only one item of a subscale was missing, data was replaced by the mean of that subscale, if more items were missing they were deleted pairwise or listwise depending on whether used in bivariate or multivariate analyses.

# Validity Evidence Based on Factor Structure

In a CFA testing the five-factor structure, the RMSEA result 0.06 was acceptable while IFI (0.89) and TLI (0.87) were just under the acceptable criteria.

# Internal Consistency

Cronbach scores for the subscales and total score were adequate to excellent-range 0.76–0.90.

## Test-Retest Reliability

White and Hudson did not explore test-retest and this is currently unknown.

# Ranges

Ranges for the MCQ-CR were broad: Total score 30–104, PB 6–22, NB 6–24, CC 6–23, NFC 6–24, CSC 6–23.

## Validity Evidence Based on Relations With Associated Measures

White and Hudson examined correlations between the MCQ-CR and both anxiety and worry. There were significant positive relationships between the total score and all subscales of the MCQ-CR and anxiety, with large effect sizes for Total score r = 0.56 and NB r = 0.56, medium effect sizes for CC r = 0.31, NFC r = 0.47, and CSC r = 0.46, and a small effect size for PB, r = 0.20. The Total score (r = 0.55) and NB (r = 0.65) also were significantly related to worry with large effect sizes, and NFC (r = 0.46) and CSC (r = 0.48) significantly related to worry with medium effect sizes. PB (r = 0.08) and CC (r = 0.13) were not significantly correlated with worry.

Criterion-based validity evidence, and responsiveness were not tested.

## Age

The White and Hudson (2016) study had an age range of 7–12. They found a significant negative correlation between age and CSC (r = −0.36) and NFC (r = −0.15), relationships with other subscales and the total score were not significant.

### Differences Between Sexes

Differences on the total score and subscales were examined and all were non-significant.

# Other MCQ Measures

There was less comprehensive psychometric data available for other MCQ measures used. The two studies that used the MCQ-30 (Gallagher and Cartwright-Hatton, 2008; Welsh et al., 2014) did not report level of understandability, factor-analysis data, internal consistency, test-retest data, range, or analysis of age and gender relationships. Concurrent-based evidence of validity of the MCQ-30 total score came from Gallagher and Cartwright Hatton's finding that it significantly correlated with anxiety, only unstandardized betas were reported. Criterion based evidence for validity of the total score and some subscales came from Welsh et al.'s finding that a group, aged 12–17, at high risk of psychosis, scored significantly higher on the total score (d = 1.16), NB (d = 1.49), CC (d = 0.93) and NFC (d = 0.92) than controls, all effect sizes were large.

The positive beliefs about worry subscale of the MCQ-65 (MCQ-PBR) adapted for children by Meiser-Stedman et al. (2007) had excellent internal consistency (0.90). Concurrent based validity was supported by the fact that it significantly correlated with a measure of trauma symptoms cross-sectionally (r = 0.34). It also significantly correlated prospectively with

trauma symptoms 6 months after the trauma (r = 0.38) but this relationship became non-significant when time 1 trauma symptoms were controlled for (Meiser-Stedman et al., 2009). Criterion based evidence of validity came from the fact that scores on the MCQ-PBR were significantly higher in a group with Acute Stress Disorder than a control group (d = 0.66; a medium effect size) (Meiser-Stedman et al., 2007).

The CSC-E used by Jacobi et al. (2006) had adequate internal consistency (0.77). Concurrent-based evidence for validity was shown by significant relationships between the CSC-E and measures of obsessive-compulsive symptoms, anxiety and depression, individual rs not given. Criterion-based validity evidence was not assessed. The MCQ-PBR and CSC-E are purportedly unidimensional but this was not tested in these studies nor was level of understandability, test-retest data, range, or age and gender differences discussed.

# DISCUSSION

# Overview

Forty-five studies that used MCQ measures or derivatives in children/adolescents were identified in the review reflecting the growth in this research area. Studies used one of seven versions of MCQ measures or derivatives. Of these, one was adapted from the MCQ for use in adolescents-the Metacognitions Questionnaire-Adolescent version (MCQ-A); four for use with children: Metacognitions Questionnaire-Child version (MCQ-C), Metacognitions Questionnaire Children-30 (MCQ-C30), Metacognitions Questionnaire-Child version Revised (MCQ-CR) and the Metacognitions Questionnaire-65 Positive Beliefs scale Revised (MCQ-PBR); and two measures developed for adults were used without adaptation: Metacognitions Questionnaire-30 (MCQ-30), and the Cognitive Self Consciousness-Expanded scale (CSC-E).

The MCQ-A (12 studies) and MCQ-C (19 studies) were the most commonly used and the largest amount of psychometric data is available for these measures. The MCQ-C30 was used in eight studies but these consisted of only five separate samples. Other MCQ measures were each used in two or less studies.

Most studies using the MCQ-A only recruited adolescent participants (aged 12 and older)-the age group that the questionnaire was designed-for, so psychometric data for the MCQ-A largely represents the measure's properties as used with adolescents. Of the four measures designed for use with children, studies examining the MCQ-C, MCQ-C30, and MCQ-PBR used participants with a range of ages spanning children and adolescents (range across measures 7–18), while the one study that examined the MCQ-CR used children aged 7–12. Studies that used adult measures-the MCQ-30 and CSC-Eused adolescent samples and so results reflect their use with this population.

# Factor Structure

The strongest evidence supporting factor structure and latent constructs they represent exists for the MCQ-A as its five-factor structure was supported in the three studies that examined it (Cartwright-Hatton et al., 2004; Ellis and Hudson, 2011; Wolters et al., 2012). However, some caution must be applied when interpreting these results as only one of these studies (Ellis and Hudson, 2011) included clinical populations in their sample and only one of these studies (Wolters et al., 2012) had a sample >300. While all three studies examined a single-order model consisting of the five subscales of the adult versions of the MCQ, only Wolters et al. also examined a second-order model with total score as the higher-order factor and the five-subscales as lower order factors. Results suggested an adequate fit reflecting a recent study in adults which found that the data supported the MCQ-30 as having a second-order or bifactorial model in a large adult population (Fergus and Bardeen, 2017). Initial factor-analysis results of the MCQ-A are promising and were in line with our hypothesis that the factor structure of MCQ in younger populations would be similar to the one found in adults. However, further studies of both single and secondorder models are warranted, particularly using large clinical populations. Studies examining measurement invariance of the MCQ-A factors across gender and age are needed as this has not yet been assessed.

Evidence for the four-factor structure of the MCQ-C, examined in three studies, was mixed. The removal of the cognitive confidence subscale, one of the factors present in earlier versions of the MCQ, from the MCQ-C means that participants were not exposed to the same items as those who completed the full 30 item version in other studies. It is possible that this led to somewhat different responses in the retained items and it is also possible that this could affect item clustering and latent variables emerging from factor-analyses. An additional problem with removing one subscale from the measure is that one important type of metacognition identified in the S-REF model is not assessed. It also prevents comparison of results on this subscale with results from other versions of the MCQ. A strength of other MCQ full-scale measures in contrast is that their structure and items match the MCQ-30, allowing comparison of analyses using all subscales and the total score from children to adolescents to the adult population. The reasoning given by Bacow et al. (2009) for removing the cognitive confidence factor was that results from a study suggested that cognitive confidence may comprise several different elements–confidence in memory, reality monitoring and attention (Hermans et al., 2008) and that they wished to remove this factor until further research clarifies this. However, in our view this does not justify dropping this subscale from the questionnaire.

The five-factor structure of the MCQ-C30, with the total score as a higher-order factor, was supported, particularly in 13–17 year olds, but was only tested in one study and this study used a nonclinical population. Further studies examining the MCQ-C30's factor structure and measurement invariance using clinical and non-clinical populations are needed.

The MCQ-CR was only used in one study (White and Hudson, 2016). The factor-analysis results examining a fivefactor structure was only partially supportive of its latent structure. The MCQ-CR introduced the possibility of responding "I don't understand" to each item and the impact of this on factoranalysis and other results needs to be considered. An advantage of having the possibility of giving this response is that it can help in assessing which items are not well-understood. However, a significant disadvantage is that it introduces a new response to each item, that is not part of the original measure, which might bias interpretation and the desired response to the items. For example, rather than completing items based on the first overall impression, the person is asked to analyze their own understanding or doubts about the meaning of items in this context which may introduce deliberation and bias responses. Additionally, it raises the question as to how to treat items scored as "I don't understand." In the White and Hudson study they were treated as missing data which, depending on the amount of missing items, was replaced by means or deleted. A problem with this is that certain items may have not been generally understood more than others and so the pattern of missing data may not have been random.

The factor structure of the other MCQ measures i.e., the MCQ-30, MCQ-PBR, and CSC-E were not examined in the studies included in the review and remain to be explored in children/adolescent populations.

# Internal Reliability

The internal validity of most subscales and the total score of the MCQ-A were supported by adequate to excellent Cronbach alphas across studies although evidence for the internal reliability of the Need for Control subscale was mixed and needs further exploration. Of note, in the validation study of the MCQ-30 in adults the NFC subscale had the poorest internal reliability (Wells and Cartwright-Hatton, 2004). The internal reliability of the MCQ-C total score was supported in the three studies that examined it. The internal reliability of individual subscales varied between studies, with PB and NFC in particular having weak internal consistency in certain studies but not others. There was a similar pattern with the MCQ-C30 with general support for the total score but variations on subscales, with NFC having the lowest range of Cronbach alphas. Internal validity of the MCQ-CR Total score and subscales, MCQ-PBR and CSC-E were supported but were only examined in one study each and further exploration is needed.

# Validity Evidence Based on Relations With Associated Measures

Concurrent-based evidence for validity was strong across MCQ measures used, with significant relationships demonstrated between the different measures and a range of psychological symptoms. As per our hypothesis, of the subscales, the strongest and most consistent results were for NB. NB correlated significantly with a range of symptoms in almost all analyses across MCQ measures and all correlations represented medium or large effect sizes using Cohen's criteria. Results for NB reflect findings using the MCQ-30 in adults where NB relates strongly to a range of symptoms (e.g., Wells and Cartwright-Hatton, 2004; Spada et al., 2008; Bailey and Wells, 2013). This is consistent with the central role of beliefs concerning the uncontrollability and danger of thoughts in prolonging and intensifying psychological difficulties in the S-REF model (Wells and Matthews, 1994). The total score also emerged as a consistent predictor of symptoms, in fact it significantly related to different symptoms in every analysis that used it across MCQ measures. The strong and consistent findings for the total score may indicate the importance of a general metacognitive factor across disorders, while some of the subscales apart from NB may have more variability as to their levels of importance depending on which type of psychological difficulty.

# Validity Evidence Based on Evidence of Relations With a Criterion

Criterion-based evidence of validity of the total score and NB and NFC subscales of the MCQ-A came from findings in three studies that these scores were consistently higher in clinical than non-clinical groups, all with large or medium effect sizes, results for other subscales were less consistent. NB and NFC also distinguished clinical and non-clinical groups in the one study that tested this using the MCQ-C30 and along with CC in a study using the MCQ-30 (Welsh et al., 2014). Results reflect our hypothesis that, of the subscales, NB and NFC would most consistently and strongly differentiate clinical and non-clinical groups. This parallels findings in adults, with a meta-analysis examining across-study differences on MCQ subscales between clinical and non-clinical groups finding that the negative beliefs and need for control subscales were highest in clinical groups when compared to non-clinical controls, with large effect sizes (Sun et al., 2017). In studies that used the MCQ-C, the NB subscale was significantly higher in clinical than non-clinical groups in three out of four studies. However, the NFC subscale did not emerge as significantly higher in the two studies that examined this although in one of these studies most of the comparison non-clinical group had sub-clinical symptoms which in a fairly small sample was likely to have obscured results. The number of studies comparing clinical and non-clinical children/adolescents across MCQ measures is relatively small and further comparisons are needed particularly as in the metaanalysis of a large number of adult studies, all MCQ subscales emerged as significantly higher in clinical compared to nonclinical groups.

# Responsiveness

Only five studies in the review examined changes in MCQ scores following an intervention or treatment. Studies examining MCQ scores following CBT or Mindfulness interventions using the MCQ-A, MCQ-C, and MCQ-C30 found decreases on at least some subscales and/or total score giving initial support for some responsiveness for these measures. These results were in line with our hypothesis that there should be some change in MCQ scores following any form of treatment that was successful in reducing symptoms. The only study in the review (Esbjørn et al., 2018) that carried out a trial of Metacognitive Therapy (MCT), used the MCQ-C30 as one of their outcome measures. This is a particular test of responsiveness as MCT for GAD, examined in this study, attempts to modify a number of the belief domains measured by the MCQ. The findings of large effects for decreases on the total score, and significant decreases in most subscales at post-treatment is a promising finding for the use of the MCQ-C30 to measure changes in metacognitions following treatment in young populations. Results are consistent with our hypothesis that changes in MCQ scores would be particularly apparent following MCT. The responsiveness of the MCQ-C30 was also supported by a CBT trial which examined changes in the total score of the MCQ-C30 and found medium effects at post-treatment and large effects at follow-up. Further studies of responsiveness of the different MCQ measures, particularly following MCT, are needed.

# Age

Studies that examined the relationships between age and the MCQ-A total score and subscales (age range across studies 11– 18) found either no or small relationships. This is supportive of the idea that these metacognitions could be fully formed as early as 11 and remain stable across adolescence. However, studies did not break down the distribution of ages within their studies. Findings with other MCQ measures, that included younger participants, were somewhat mixed with individual subscales emerging in only some analyses as being related to age either positively or negatively, using the MCQ-C, MCQ-C30, and MCQ-CR. To fully test if there are any age differences in MCQ scores between children/adolescents of different ages, future studies should consider recruiting participants with an even distribution of age, or directly comparing scores of groups of younger and older children. The one study in the review that did the latter (Irak, 2012) found that 13–17 had higher scores on the positive belief subscale only compared to 8– 12 year olds, further studies are needed to see if this result is replicated. Current findings, together with the fact that ranges of MCQ measures when given were broad, suggest that dysfunctional metacognitions could develop at an early age. This is consistent with findings that suggest there may be childhood factors that lead to vulnerability to the development of these metacognitions, such as early experiences of emotional abuse (e.g., Myers and Wells, 2015; Østefjells et al., 2017) and parenting style (Gallagher and Cartwright-Hatton, 2008; Spada et al., 2012; Lønfeldt et al., 2017b).

# Differences Between Sexes

Most studies that examined differences between males and females on scores of MCQ measures (MCQ-A, MCQ-C, MCQ-C30, MCQ-CR) did not find significant differences which suggest they may, as hypothesized, not be present or may be small. Of note in one of the minority of studies that found differences (Esbjørn et al., 2013; a significantly higher score for girls on the total score) was that controlling for anxiety removed the effect, suggesting it may have been caused by elevated anxiety symptoms in girls. As higher prevalence rates for having an anxiety disorder in females compared to males have been found in children (Anderson et al., 1987); adolescents (Lewinsohn et al., 1998), and adults (Kessler et al., 1994) it may be important for future studies to control for anxiety in analyses of sex differences on MCQ scores in all these groups.

# Test-Retest Reliability

Test-retest reliability was examined in few studies using any of the MCQ measures but results with the MCQ-A and MCQ-C mainly support the stability of the measures over time. More research is needed into this across measures.

# Understandability

The understandability of the measures also needs further investigation. No study, to our knowledge, has examined the understandability of the MCQ-A to adolescents or preadolescents or whether the MCQ-A is more understandable to adolescents than the MCQ-30 and there is a need for these issues to be investigated. The understandability of the MCQ-C30, MCQ-PBR, and CSC-E to children or adolescents has also not been examined, while the one study that examined the understandability of the MCQ-C found that six items were not understandable to most of the small sample of 7–8 year olds tested. Although the MCQ-CR was found to be understandable to most 7–8 year olds in the one study that used it, understanding increased with age.

# CONCLUSIONS

The choice of which version of the MCQ to use in future studies in younger populations may well be influenced by the age group of the population being examined. The MCQ-A has largely good psychometric parameters in adolescents, the population it was designed for, and few studies have used it with younger populations. We suggest future studies using adolescents should certainly consider using the MCQ-A. Studies whose participants include pre-adolescent children and who want to measure the full range of constructs measured by the MCQ-30 should consider using the MCQ-C30 which has initial, although currently relatively limited, psychometric data supporting it. Two studies suggest that the MCQ-C30 is responsive to changes in metacognition following treatment and so the MCQ-C30 may be a particularly appropriate choice for treatment trials that include children.

The youngest age of children included in studies in the review was seven and psychometrics for children younger than this are unknown. The fact that studies only recruited children aged seven and above reflects the traditional view that this is the age where children can report on metacognitive knowledge (Flavell, 1979). However, a recent study (Marulis et al., 2016) suggests that when measured appropriately some younger children-age 3- 5 may be able to report on their metacognitions. Although not using an MCQ measure (Wilson and Hughes, 2011), found that some 6 year olds held both positive and negative metacognitions about worry. Future studies may consider examining children

# REFERENCES


younger than seven on MCQ measures although content and means of administration may well have to be adapted further to accommodate this group.

Strengths of almost all studies reviewed include clearly stated aims/hypotheses, the use of standardized symptom or diagnostic measures and appropriate analyses. Studies varied as to the appropriateness of selection criteria and the adequacy of sample size. Only a minority of studies discussed and corrected for missing data. Although the quality of studies was generally good, the methodological limitations, in particular variable sample sizes, should be born in mind when interpreting psychometric results. A number of studies included younger children and as results from two studies suggest some younger children may have difficulty in understanding some MCQ items, caution must be applied in interpreting some psychometric results of these studies. Although a number of studies included clinical populations most used non-clinical populations thus psychometrics for non-clinical groups are more extensive. No studies carried out analyses of psychometrics based on Item Response Theory (IRT) which has a number of advantages over analyses based on Classic Test Theory. Future studies would be strengthened by carrying out psychometric analyses based on IRT.

Bearing in mind these limitations, this review suggests that several MCQ measures have promising psychometrics in younger populations. The metacognitions assessed by the MCQ appear to be present in children/adolescents and can be assessed by self-report measures. The similarity of a number of results, particularly of concurrent and criterion based tests of validity, in comparison with results in adults, suggest consistent patterns of relationships between the metacognitions assessed by the MCQ and mental health symptoms. Research into metacognitive theory in children and adolescents is growing; research into metacognitive therapy in this population is in its infancy but initial results are promising (Simons et al., 2006; Esbjørn et al., 2018). Further testing and development of metacognitive measures in children and adolescents should help advance this promising area of research and practice.

# AUTHOR CONTRIBUTIONS

SM and AW were involved in study conceptualization. SM and SS were involved with systematic search, article collection and quality assessment, and data analysis. SM carried out data extraction and synthesis. All three authors contributed to the manuscript.

Psychiatry 44, 69–76. doi: 10.1001/archpsyc.1987.018001300 81010


associated with youth social anxiety disorder? J. Affect. Disord. 208, 33–40. doi: 10.1016/j.jad.2016.09.052


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Myers, Solem and Wells. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# APPENDIX

# METHODOLOGICAL QUALITY ASSESSMENT

#### Research question and design


0 Statistical tests not appropriate and/or not appropriately described

# Modeling the Relationships Between Metacognitive Beliefs, Attention Control and Symptoms in Children With and Without Anxiety Disorders: A Test of the S-REF Model

Marie Louise Reinholdt-Dunne<sup>1</sup> \*, Andreas Blicher<sup>1</sup> , Henrik Nordahl<sup>2</sup> , Nicoline Normann<sup>1</sup> , Barbara Hoff Esbjørn<sup>1</sup> and Adrian Wells<sup>3</sup>

<sup>1</sup> Department of Psychology, University of Copenhagen, Copenhagen, Denmark, <sup>2</sup> Department of Psychology, Norwegian University of Science and Technology, Trondheim, Norway, <sup>3</sup> School of Psychological Sciences, University of Manchester and Greater Manchester Mental Health NHS Trust, Manchester, United Kingdom

#### Edited by:

Gianluca Castelnuovo, Catholic University of the Sacred Heart, Italy

#### Reviewed by:

Michael Simons, RWTH Aachen University, Germany Łukasz Gawe¸ da, University Medical Center Hamburg-Eppendorf, Germany Sandra Sassaroli, Studi Cognitivi S.p.A, Italy

> \*Correspondence: Marie Louise Reinholdt-Dunne Marie.Reinholdt@psy.ku.dk

#### Specialty section:

This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology

> Received: 07 August 2018 Accepted: 07 May 2019 Published: 07 June 2019

#### Citation:

Reinholdt-Dunne ML, Blicher A, Nordahl H, Normann N, Esbjørn BH and Wells A (2019) Modeling the Relationships Between Metacognitive Beliefs, Attention Control and Symptoms in Children With and Without Anxiety Disorders: A Test of the S-REF Model. Front. Psychol. 10:1205. doi: 10.3389/fpsyg.2019.01205 In the metacognitive model, attentional control and metacognitive beliefs are key transdiagnostic mechanisms contributing to psychological disorder. The aim of the current study was to investigate the relative contribution of these mechanisms to symptoms of anxiety and depression in children with anxiety disorders and in nonclinical controls. In a cross-sectional design, 351 children (169 children diagnosed with a primary anxiety disorder and 182 community children) between 7 and 14 years of age completed self-report measures of symptoms, attention control and metacognitive beliefs. Clinically anxious children reported significantly higher levels of anxiety, lower levels of attention control and higher levels of maladaptive metacognitive beliefs than controls. Across groups, lower attention control and higher levels of maladaptive metacognitive beliefs were associated with stronger symptoms, and metacognitions were negatively associated with attention control. Domains of attention control and metacognitions explained unique variance in symptoms when these were entered in the same model within groups, and an interaction effect between metacognitions and attention control was found in the community group that explained additional variance in symptoms. In conclusion, the findings are consistent with predictions of the metacognitive model; metacognitive beliefs and individual differences in selfreport attention control both contributed to psychological dysfunction in children and metacognitive beliefs appeared to be the strongest factor.

Keywords: anxiety disorders, childhood anxiety, metacognition, attention control, prevention, psychological treatment

# INTRODUCTION

Anxiety disorders are the most common psychological problems in children and adolescents with prevalence estimates ranging from 3–20% (Costello et al., 2005; Cartwright-Hatton et al., 2006). They are associated with considerable developmental, psychosocial and psychopathological complications (Beesdo et al., 2009). For example, anxiety has a negative impact on school

**85**

functioning (Mychailyszyn et al., 2010) and is a major risk factor for developing comorbid disorders such as depression (Wittchen et al., 2003). Moreover, childhood anxiety disorders predict psychopathology in adolescence and adulthood (Bittner et al., 2007; Copeland et al., 2009), and are associated with substantial functional impairment in later life (Copeland et al., 2014). Hence, identifying factors underlying anxiety disorders and vulnerability to developing them may inform the development of effective prevention- and treatment interventions that could benefit individuals and society.

Cognitive theories of anxiety implicate biases in information processing in the development and maintenance of anxiety (e.g., Williams et al., 1988). Such biases can be observed in the content of interpretations of experience where the sense of danger and threat predominate (Beck et al., 1985). They are also evident at the level of attentional processes, where anxiety and depression are associated with biased attention for negative emotion-related stimuli (Bar-Haim et al., 2007; Cisler and Koster, 2010). A major challenge is to identify the factors that give rise to bias in psychological disorders. Early models viewed bias as the result of automatic or reflexive processes, but this has been questioned. For example, Wells and Matthews (1994) proposed specific multiple influences on bias including metacognitive beliefs and the individual's goals and strategies for self-regulation of which volitional attention control is a major component. Attention control has been conceptualized as the ability to control attention in inhibiting a dominant response in favor of a less accessible, subdominant response that may be more functional (Rothbart and Bates, 1998; Derryberry and Reed, 2002). Thus, attention control is viewed as a self-regulatory capacity, and it has been shown to moderate the association between attentional bias for threat and anxiety in adults (e.g., Derryberry and Reed, 2002; Bardeen and Orcutt, 2011) and in children (e.g., Lonigan and Vasey, 2009; Susa et al., 2012). Consequently, individual differences in attention control could contribute to resilience or vulnerability to emotional distress (e.g., Lonigan et al., 2004; Muris et al., 2004, 2007, 2008; Susa et al., 2012).

The role of such influences of attention and their link with emotional vulnerability has been captured in detail in the Self-Regulatory Executive Function (S-REF) model (Wells and Matthews, 1994, 1996), which is the basis of the metacognitive model of emotional disorder and treatment (Wells, 2009). In this model, psychological dysfunction is associated with a style of thinking called the cognitive attentional syndrome (CAS). The hallmark of the CAS is perseverative thinking consisting of worry/rumination, threat monitoring and maladaptive coping strategies. The activation and persistence of the CAS is dependent on underlying metacognitive knowledge (i.e., beliefs about cognition). Metacognitive knowledge refers to the information that individuals hold about their own cognition and internal states (e.g., "my worrying thoughts are uncontrollable") and about coping strategies (e.g., "worrying helps me to get things sorted out in my mind") and is implicated across mental disorders (Wells, 2009; Sun et al., 2017). Such metacognitions also contribute to psychological vulnerability when the presence of a mental disorder has been accounted for Nordahl and Wells (2017).

Within the S-REF model, attention control is considered a general resource that facilitates cognitive regulation and the ability to disengage from conceptual processing and perseverative self-focused attention (i.e., the CAS). Whilst this ability is separate from but related to the effects of metacognitive knowledge (Wells and Matthews, 1994) individual differences in attention control (i.e., executive functions) could affect the individual's ability to disengage from the CAS. Attentional control is likely to be comprised at least in part of knowledge or beliefs about attention and studies separating the effects of attention performance (skills) and beliefs about attention in psychological disorder are lacking.

In the metacognitive model in particular, beliefs about poor attention control are of interest as they are likely to be part of a broader dysfunctional metacognitive knowledge base hypothesized to underlie psychological disorder. Furthermore, different dimensions of metacognition may interact and increase the risk or severity of psychological disorder symptoms. In particular, high levels of perceived attention control could help to ameliorate the negative effects of beliefs about the dangerousness of thoughts on anxiety. In contrast, low levels of perceived attention control might enhance the negative effect of metacognitions about the uncontrollability and danger of worrying.

In adults, studies have found support for an association between greater maladaptive metacognitive beliefs and lower perceived attention control (Spada et al., 2010; O'Carroll and Fisher, 2013; Spada and Roarty, 2015; Fernie et al., 2016), and both perceived attention control and metacognitive beliefs have been found to explain unique variance in performance test anxiety (O'Carroll and Fisher, 2013), state anxiety in students before end-of-year examinations (Spada et al., 2010), and decisional procrastination (Fernie et al., 2016). Moreover, Fergus et al. (2012) found that the relationships between activation of the CAS and symptoms became increasingly stronger as self-reported attention control decreased, indicating that activation of the CAS is associated with especially deleterious effects for individuals with low attention control.

In sum, attention control (beliefs) and other metacognitive beliefs may be central to understanding psychological disorder and vulnerability. However, research on the relationship between attention control and metacognitive beliefs and their individual or combined contribution to symptoms is scarce, and to the authors' knowledge has not been investigated in children. The aim of the current study was therefore to investigate the relative contribution of attention control and metacognitive beliefs in children with anxiety disorders and in non-clinical controls. We set out to examine differences between community controls and clinical patients and to explore the unique and interactive effects of metacognitive beliefs and attention control within each group. Our hypotheses were as follows; (1) the clinical group will report greater severity of symptoms, lower attention control and higher levels of maladaptive metacognitive beliefs than the control group; (2) attention control will be negatively associated with symptoms; (3) metacognitive beliefs will be positively associated with symptoms; (4) metacognitive beliefs will be negatively associated with attention control;

(5) attention control and metacognitive beliefs will account for unique variance in symptoms; (6) there should be an interaction between metacognitive beliefs and attention control that contributes to symptoms. Because we cannot predict based on theory whether the interaction occurs in non-patients and/or patients we tested the model in the clinical and non-clinical groups separately.

# MATERIALS AND METHODS

## Participants and Procedure

A child community sample was recruited by sending invitation letters to 1601 families with children aged 8 to 12 years of age living within the catchment area of Center for Anxiety, Department of Psychology, University of Copenhagen. The sample was randomly selected by the Danish Central Office of Civil Registration, and the invitation letter specified that only typically developing children could participate. Families that wished to participate completed a questionnaire booklet online at home, prior to entering the clinic. When entering the clinic, mothers completed the parent version of the Anxiety Disorders Interview Schedule (ADIS; Silverman and Albano, 1996). The interview showed that all participating children were free of psychiatric disorders.

In the clinical sample, families referred their preadolescent children to the clinic, although they had often been recommended to contact the clinic by other professionals, e.g., psychiatrists and school psychologists. Consequently, preadolescents in aged between 7 and 14 that had a primary anxiety disorder, either generalized anxiety disorder, separation anxiety disorder, specific phobia, or social phobia were eligible as participants for the study if they also had an IQ above 70, and one parent native speaker of Danish. The children were assessed with the ADIS (Silverman and Albano, 1996). A combined diagnosis was derived from child and parent ratings, and showed that 110 (65.1%) of the children fulfilled the diagnostic criteria for generalized anxiety disorder, 33 (19.5%) for separation anxiety disorder, 14 (8.3%) for social phobia, and 12 (7.1) for specific phobia. The majority of the children (141; 83.4%) had comorbid anxiety disorders. Sixteen children (9.5%) also had comorbid mood disorder (Dysthymia or Major depressive disorder). A total of 351 children participated in this study, 182 community children (100 girls; 54.9%) between 7 and 12 years of age (M = 10.00, SD = 1.40) and 169 children diagnosed with a primary anxiety disorder (89 girls; 52.7%) between 7 and 14 years of age (M = 9.93, SD = 1.83) were included. Comparison of the community and clinical groups using Chi square and independent t-tests (on categorical and continuous variables, respectively) showed no significant group differences in gender or age distribution between the two groups.

## Ethics Statement

Ethical approval for the study was obtained from the Institutional Review Board at the Department of Psychology, University of Copenhagen. The study complies with ethical standards in the 1964 Helsinki declaration and its later amendments regarding assessment and treatment for children enrolled in psychological research studies. Written informed consent to participate was obtained from all parents of participating youth, and assent was obtained from the youth.

# Measures

The Revised Child Anxiety and Depression Scale – Child version (RCADS, Chorpita et al., 2000) is a 47-item self-report questionnaire measuring child anxiety and depression symptoms. RCADS consists of six subscales: Major depression, social phobia, panic disorder, separation anxiety, generalized anxiety, and obsessive-compulsive disorder. The major depression subscale consists of ten items, the social phobia and the panic disorder subscales consist of nine items, the separation anxiety subscale consists of seven items, and the generalized anxiety and the obsessive-compulsive disorder subscales consist of six items. A total score can be computed by summing the subscales. Validation of the Danish version of RCADS has shown satisfactory psychometric properties (Esbjørn et al., 2012). In the current study, internal consistency was excellent in both the community group (α = 0.94) and the clinical group (α = 0.93).

Attentional Control Scale for Children (ACS-C; Derryberry and Reed, 2002) is a 20-item self-report questionnaire measuring subjective attentional control. ACS-C consists of three subscales: Attention focusing, attention shifting, and flexible control of thought. The attention focusing subscale consists of nine items, the attention shifting subscale consists of six items, and the flexible control of thought subscale consists of five items. Items have to be scored on a 4-point scale with 1 = never, 2 = sometimes, 3 = often, and 4 = always. After recoding inversely formulated items, a total score can be computed by summing the subscales. The ACS-C has shown acceptable psychometric properties (e.g., Muris et al., 2004, 2007, 2008). In this study internal consistency was satisfactory in the clinical group (α = 0.74) and slightly below satisfactory level in the community group (α = 0.57).

Metacognitions Questionnaire for Children (MCQ-C30, Esbjørn et al., 2013) is a 30-item self-report questionnaire measuring metacognitive beliefs and processes in children and is a simplified version of the original adult scale (Wells and Cartwright-Hatton, 2004). MCQ-C<sup>30</sup> consists of five subscales: Positive beliefs about worry, negative beliefs about uncontrollability and danger of worry, cognitive confidence, need for control, and cognitive self-consciousness. All the subscales consist of six items. Items are scored on a 4-point scale with 1 = do not agree, 2 = agree slightly, 3 = agree moderately, and 4 = agree very much. A total score can be computed by summing the subscales. The Danish version of the questionnaire has shown satisfactory psychometric properties (Esbjørn et al., 2013). In the present study internal consistency was satisfactory in both the community group (α = 0.89) and the clinical group (α = 0.86).

# Overview of Data Analyses

Independent samples t-tests were used to compare the community and the clinical group on the RCADS, and on the subscales of the ACS-C and the MCQ-C30. Then we ran

bivariate correlational analyses to investigate the relationship between these variables.

To explore if there was any interaction effect between attention control and metacognitive beliefs on symptoms, we used structural equation modeling (Bentler, 1995). MCQ-C<sup>30</sup> total score, ACS-C total score, and the interaction between these two were used as observed variables, while symptoms (RCADS) was used as a latent variable consisting of all the RCADS subscales. Evaluation of the path coefficient from the interaction variable to the latent construct symptoms was of particular interest, as a significant path coefficient would indicate that moderation occurred.

Hierarchical linear regression analyses were run in each group to test the relative contribution of the attention control domains and metacognitive belief domains. Moreover, if the SEM analysis revealed a moderation effect, we planned to add this interaction variable to the regression models as a means to evaluate its relative contribution over domains of attention control and metacognitive beliefs. RCADS was used as the dependent variable throughout. Gender and age was controlled in the first step. In the second step, we entered the ACS-C subscales, and the MCQ-C<sup>30</sup> subscales were entered in step 3. If the SEM analyses revealed moderation, we planned to enter the interaction variable on the fourth step to test whether the interaction between metacognitive beliefs and attention control explained additional variance in the final equation when unique effects of attention control and metacognitive beliefs were controlled.

# RESULTS

# Group Comparisons

We found significant differences between the groups in symptom severity (RCADS total score), in all three domains of attention control, and in all domains of metacognitive beliefs except for judgments of cognitive confidence; the clinical group scored significantly higher on symptoms and metacognitive beliefs, and significantly lower on attention control compared to the community group. Descriptive statistics and group comparisons are presented in **Table 1**.

# Correlational Analyses

In both groups, there was a significant association between RCADS and ACS-C focusing and shifting, indicating that lower levels of attention control in these two domains are associated with higher levels of symptoms, while there was no association between RCADS and the ACS-C flexible subscale in any of the groups. RCADS was significantly associated with all domains of metacognitive beliefs in the community group, and with all but positive metacognitive beliefs in the clinical group, indicating that higher levels of symptoms are associated with higher maladaptive metacognitive beliefs. Moreover, lower levels of attention control were associated with higher levels of maladaptive metacognitive beliefs; the ACS-C focusing subscale was significantly and negatively associated with all domains of metacognitive beliefs in the community group, while it was significantly negatively associated with all metacognitive belief TABLE 1 | Group comparisons between the community- and the clinical group on age, symptom severity (RCADS), attentional control (ACS-C), and metacognitive beliefs (MCQ-C30); mean score, standard deviation and t-value.


RCADS, revised child anxiety and depression scale – child version; ACS-C, attention control scale for children; focus, attention focusing; shifting, attention shifting; flexible, flexible control of thought; MCQ-C30, metacognitions questionnaire for children; pos, positive metacognitive beliefs; neg, negative metacognitive beliefs; cc, cognitive confidence; nc, need for control; csc, cognitive self-consciousness. ∗∗p < 0.01.

domains except positive metacognitive beliefs in the clinical group. The ACS-C shifting subscales was significantly and negatively associated with negative metacognitive beliefs and judgments of cognitive confidence in both groups, and with need for control in the clinical group, but not with positive metacognitive beliefs or cognitive self-consciousness in any of the groups or need for control in the community group. The ACS-C subscale flexible control of thought was significantly and negatively correlated with negative metacognitive beliefs and cognitive confidence in the clinical group, but was not associated with metacognitions in the community group. The bivariate correlations are presented in **Table 2**.

# Structural Equation Modeling

Structural equation modeling (Bentler, 1995) was used to investigate if there was an interaction effect between attention control and metacognitive beliefs in predicting distress in each group. The total score from the ACS-C and MCQ-C<sup>30</sup> together with their interaction (ACS-C total score × MCQ-C<sup>30</sup> total score) were treated as observed variables, and symptoms were treated as a latent variable consisting of each of the RCADS subscales. The path coefficients were of particular interest, and if the path from the interaction variable to the dependent variable had no predictive value, it was deleted to evaluate a second model without the interaction. Evaluation of overall model fit was conducted according to Hu and Bentler (1999), where the Comparative Fit Index (CFI) and the Tucker-Lewis Index (TLI) should be close to or more than 0.95, the standardized root mean square residual (SRMR) should be less than 0.08, and the root mean square error of approximation (RMSEA) should be less than 0.06, to represent good model fit.


well when the interaction variable was included as the CFI and SRMR were within recommendations, the TLI was borderline of its recommended value, while the RMSEA was above recommended value; χ 2 (24) = 52.315, p = 0.001, CFI = 0.964, TLI = 0.947, RMSEA = 0.081, SRMR = 0.0367. All the standardized regression weights in the model were significant at the 0.001 level, which indicated that there was an additional interaction effect between attention control and metacognitive beliefs in the community sample. In this model, the squared multiple correlation for symptoms (RCADS) was 0.66, indicating that 66% of the variance in symptoms was accounted for by the predictors. We also evaluated the model fit in the community sample without the interaction term, and this model also fitted well, showing slightly better fit indices than the first model: χ 2 (19) = 41.233, p = 0.002, CFI = 0.968, TLI = 0.952, RMSEA = 0.080, SRMR = 0.0365. However, a chi square difference test showed that the model with the interaction variable was significantly better than the model without the interaction term: 1χ <sup>2</sup> = 11.082, 1df = 5, (p < 0.05). The model with the interaction variable in the community sample is presented in **Figure 1**.

In the community group, the data fitted the model reasonably

In the clinical sample, the path from the interaction variable (attention control x metacognitive beliefs) to symptoms was non-significant, indicating that there was no additional contribution of the interaction effect. The interaction variable was therefore deleted before evaluating the model fit. All standardized regression weights in this second model were significant at 0.01 level and the squared multiple correlation for symptoms (RCADS) was 0.52, indicating that 52% of the variance in symptoms was accounted for by the predictors. Still, the model did not provide an optimal fit to the data in the clinical group; χ 2 (19) = 48.644, p < 0.000, CFI = 0.942, TLI = 0.914, RMSEA = 0.096, SRMR = 0.0479. The model without the interaction variable in the clinical group is presented in **Figure 2**.

# Hierarchical Linear Regression Analyses

In the community group, gender and age was not a significant predictor of symptoms in any of the steps in the regression model. On the second step, all domains of attention control made unique contributions to symptoms and together they explained an additional 27% of the variance. In the second step, metacognitive beliefs explained 34% of the variance in symptoms over and above the attention control domains. Need for control was nonsignificant as a predictor, but all other domains of metacognitive beliefs were significant predictors, and negative metacognitive beliefs explained most variance. Adding metacognitive beliefs to the model led the shifting subscale of the ACS-C to be nonsignificant as a predictor, while the two other attention control subscales remained significant indicating that they accounted for unique variance in symptoms. Building on the SEM-analysis, the interaction effect was entered in the model in the fourth step and explained an additional 1 % of the variance. In this final step of the equation, attention control focusing and shifting, negative metacognitive beliefs, cognitive confidence, cognitive self-consciousness and positive metacognitive beliefs together

fpsyg-10-01205 June 7, 2019 Time: 17:5 # 5

with the interaction variable remained significant predictors and explained unique variance in symptoms.

In the clinical group, age was non-significant as a predictor of symptoms, while gender was significant as a predictor in the first step, showing that female gender was associated with higher levels of symptoms. In the second step, the ACS-C focusing subscale was significant as a predictor of symptoms, while the two other ACS-C subscales were not. In sum, attention control accounted for an additional 24 % of the variance in this step. Furthermore, gender remained a significant predictor of symptoms in the second step. In the third step, metacognitive beliefs explained an additional of 22 % of the variance in symptoms over and above age/gender and attention control. Of the MCQ-C<sup>30</sup> subscales, negative metacognitive beliefs and need for control were significant individual predictors. After adding metacognitive beliefs to the model, the ACS-C focusing subscales remained significant as a predictor, while gender became non-significant. The regressions are presented in **Table 3**.

To further explore the interaction effect in the community group, we examined two scatter plots with symptoms (RCADS total score) represented along the Y-axis. In the first scatter plot, attention control (ACS-C total score) were represented along the X-axis. The participants were separated in three group based on their total MCQ-C<sup>30</sup> score; group 1 consisted of the lowest scoring one-third of the sample; group 2 consisted of the one third of the individuals that had a moderate score; group 3 consisted of the one third of the individuals with the highest score. In the second scatter plot, metacognitive beliefs (MCQ-C<sup>30</sup> total score) were represented along the X-axis, and the sample was divided in low, moderate and high attention control groups following the same principle as outlined above. The scatter plots are presented in **Figures 3**, **4**.

Inspection of the plots shows that as dysfunctional metacognitions increase from moderate to high the negative relationship between attention control and symptoms becomes stronger. There is no effect at low levels of metacognitions. Conversely, at higher levels of attention control, the positive

TABLE 3 | Hierarchical regression analysis in the community- and clinical group separately, with RCADS total score as the dependent, gender/age and subscales from the ACS-C and MCQ-C<sup>30</sup> as predictors.


RCADS, revised child anxiety and depression scale – child version; ACS-C, attention control scale for children; focus, attention focusing; shifting, attention shifting; flexible, flexible control of thought; MCQ-C30, metacognitions questionnaire for children; pos, positive metacognitive beliefs; neg, negative metacognitive beliefs; cc, cognitive confidence; nc, need for control; csc, cognitive self-consciousness; β, standardized beta coefficients. <sup>∗</sup>p < 0.05, ∗∗p < 0.01.

relationship between metacognitions and symptoms becomes weaker but there is no effect at low levels of attention control.

# DISCUSSION

Metacognitive beliefs and attention control are two influences on cognitive regulation that have been implicated in the metacognitive model of psychological disorders. This model predicts differences between these factors in clinical and nonpatient individuals, for example that higher endorsements of maladaptive metacognitive beliefs and lower attentional control abilities should be found in clinical compared to non-clinical samples. It also predicts that these factors should be positively associated with symptoms of anxiety and depression in both groups, and that they may interact to moderate the strength of association each of these factors has with anxiety symptoms.

As predicted, the clinical child group showed elevated scores on dysfunctional metacognitive beliefs and lower scores on attention control compared to community controls. Negative metacognitive beliefs about worry differentiated the most between the groups among all predictors, while confidence in memory did not differentiate between the groups. Among the attention control dimensions, the flexible control of thought subscale differentiated the most between groups.

Within each group, we found the expected positive relationship between symptom severity and metacognitive beliefs, with the strongest association with negative metacognitive beliefs about worry. There was no association with positive beliefs in the clinical group, but all other relationships between

metacognitive belief domains and symptoms were significant in both groups and showed relationships of moderate or moderate to low strength. The expected negative relationship between symptoms and attention control was evident as attention focusing and shifting showed a moderate to low association with symptoms, while the relationship with the flexible control of thoughts subscale was non-significant in both groups.

With the exception of positive metacognitive beliefs in the clinical sample, attention focusing was negatively associated with all domains of metacognitive beliefs in both groups. The same relationship was observed between attention shifting, negative metacognitive beliefs and cognitive confidence in both groups, and also between attention shifting and need for control in the clinical group, indicating that maladaptive metacognitive knowledge is related to lower (perceived) ability to control attention.

On testing for interaction effects, the interaction between the total score on the MCQ-C<sup>30</sup> and the ACS-C, was found in the community- but not the clinical sample. This is an interesting finding because it suggests that a multiplicative effect of metacognitions and attention control on symptoms might be most relevant to sub-clinical anxiety and depression symptoms (at least in children). This raises an intriguing but speculative possibility, but one that is nonetheless congruent with the metacognitive model; that strongly held maladaptive metacognitive beliefs can neutralize the emotional benefit conferred by strong attention control beliefs, or conversely that strong attention control can remediate the negative effects of strongly held maladaptive metacognitive beliefs. But these findings point to a possible mechanistic or process-based difference between clinical and non-clinical samples. The interaction was not observed in the clinical group, one explanation might be that the deleterious effects attributed to high metacognitions is not moderated or offset by attentional control in those who have clinical disorder because their dysfunctional metacognitions are so much greater or these individuals use less effective mental regulation strategies. Such effects would be consistent with the S-REF model where attention control and flexibility is considered a general purpose processing resource that is compromised by high dysfunctional metacognitions and strategy selection (i.e., using extended negative thinking to deal with stress) (Wells and Matthews, 1994).

When exploring the relative contribution from the individual ACS-C and MCQ-C<sup>30</sup> subscales in the community sample, attention focusing and shifting, positive metacognitive beliefs, negative metacognitive beliefs, low cognitive confidence, cognitive self-consciousness and the additional interaction effect between metacognitions and attention control explained unique variance in symptoms. In the clinical sample, attention

focusing, negative metacognitive beliefs and beliefs about the need to control thoughts were the only significant independent contributors to symptoms, indicating that greater negative beliefs about the uncontrollability and danger of thoughts, need to control thoughts, and lower levels of attention focusing contribute individually to greater symptoms in clinically anxious children. While positive metacognitive beliefs are suggested to be an important disposition to anxiety disorder specific in metacognitive models of anxiety (Wells, 2009), we found that there was no independent effect of positive metacognitive beliefs on anxiety in the clinical group. One explanation could be that the effects of positive beliefs was masked by the substantial contribution from negative metacognitive beliefs and need for control, that are a more proximal contributor to disorder. Moreover, it could be that different domains of metacognitions may serve as maintenance factors (i.e., negative metacognitive beliefs and need for control) and as causal factors constituting vulnerability (i.e., positive metacognitive beliefs) as reported by others (e.g., Nordahl et al., 2019), but this possibility cannot be tested given the cross-sectional data-set in the present study.

In sum, our findings suggest that the metacognitive model might offer a useful framework to conceptualize psychopathology and psychological vulnerability in children with the implication that metacognitive therapy techniques for the preventionand treatment of disorder could be applicable. Metacognitive therapy (Wells, 2009) interventions aim to modify maladaptive metacognitive knowledge and strengthen flexible control over attention and they should be investigated in this group. While cognitive-behavioral therapy (CBT) is an effective treatment for anxiety in children (e.g., Ewing et al., 2015; James et al., 2015), regulatory processes (i.e., metacognition) and executive function aspects are in large overlooked in these models and treatments, which may account for the fact that a substantial number of patients are non-responders (James et al., 2005). For example, Reinholdt-Dunne et al. (2015) found that attention control did not significantly change in anxious children following CBT. Furthermore, the effect size of anxiety prevention programs for children has been reported as small (e.g., Fisak et al., 2011), indicating a need for further therapeutic developments. Applications of metacognitive therapy and techniques for children have begun and show promising results (Simons et al., 2006; Esbjørn et al., 2015; Murray et al., 2016, 2018; Simons and Kursawe, 2019). However, more studies are needed before any firm conclusions on its effect can be drawn.

The present study has several limitations that should be acknowledged. First, a cross-sectional design was used, and therefore no causal inferences can be made. Second, the clinical sample in this study predominantly consisted of children with primary GAD, and our study should therefore be replicated in a wider clinical context. Furthermore, the clinical sample was

a convenience sample of preadolescents referred for treatment, which resulted in a heterogeneous sample in terms of both primary diagnoses and age. While this is a limitation in some respect, our study has external validity as the sample consisted of patients referred to a clinic setting. Third, selfreport symptom assessment is a limitation of the study. Fourth, an important question concerns the measurement of attention control. In a recent study, the ACS was largely unrelated to behavioral performance measures of attention control (Williams et al., 2017), indicating that the ACS may represent subjective judgments of attention (metacognitive knowledge) rather than actual cognitive ability. However, metacognitive beliefs have been associated with objective shifting ability after controlling for symptoms and general cognitive function (Kraft et al., 2017) and improved neuropsychological functioning has been observed following MCT for depression (Groves et al., 2015) in adults, suggesting that there is a link between metacognitions (including beliefs about attention) and some aspects of objective executive functioning. Further research should utilize longitudinal and experimental designs with objective measures of attention control to better address the relation and direction of relations among metacognitive beliefs, objective attention control and psychopathology symptoms. In addition, testing the contribution of attentional control and metacognitive knowledge to symptoms in more specific clinical groups of children may further enhance our understanding. Further research should take account of potential age differences when exploring the influence of metacognitive knowledge and executive functions on psychological disorder and vulnerability in children.

# REFERENCES


# CONCLUSION

In conclusion, metacognitive beliefs and attention control appear to contribute to emotion disorder symptoms in both clinical and non-clinical children samples. This suggests that prevention strategies and treatment interventions should aim to modify maladaptive metacognitive knowledge and enhance judgments of attention control as recommended in metacognitive therapy. But the nature of the relationship between objective attention performance, beliefs about attention control and disorder symptoms remains to be differentiated.

# ETHICS STATEMENT

Ethical approval for the study was obtained from the Institutional Review Board at the Department of Psychology, University of Copenhagen. The study complies with ethical standards in Denmark regarding assessment and treatment for children enrolled in psychological research studies. Written informed consent to participate was obtained from all parents of participating youth, and assent was obtained from the youth.

# AUTHOR CONTRIBUTIONS

All authors were part of the design of the study. MR-D, AB, NN, and BE were part of data collection and writing the manuscript. HN and AW were part of analyzing the data and writing the manuscript.



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Reinholdt-Dunne, Blicher, Nordahl, Normann, Esbjørn and Wells. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Metacognitive Beliefs Predict Greater Mental Contamination Severity After an Evoking Source

Thomas A. Fergus\*, Kelsi A. Clayson and Sara L. Dolan

Department of Psychology and Neuroscience, Baylor University, Waco, TX, United States

Mental contamination occurs when individuals experience feelings of internal dirtiness and distress in the absence of physical contact with a contaminant. Women who experience sexual trauma frequently report mental contamination. The self-regulatory executive function (S-REF) model proposes that metacognitive beliefs contribute to the appraisal and regulation of thinking, leading to expectations that metacognitive beliefs would predict greater mental contamination severity following an evoking source. Women who reported directly experiencing sexual trauma (N = 102) completed selfreport measures of metacognitive beliefs and covariates during an online study session, and subsequently completed a task that evoked mental contamination during a followup in-person study session. Metacognitive beliefs surrounding the uncontrollability and danger of thoughts, cognitive confidence, and the need to control thoughts positively correlated with mental contamination severity following the evoking source. Metacognitive beliefs surrounding the uncontrollability and danger of thoughts predicted greater mental contamination severity following the evoking source in multivariate analyses that statistically controlled for baseline mental contamination severity, trait anxiety, and overlap among the metacognitive beliefs. The present results provide preliminary support for the S-REF model as a potential framework for conceptualizing mental contamination.

Keywords: mental contamination, metacognitive beliefs, posttraumatic stress, self-regulatory executive function (S-REF) model, sexual trauma

# INTRODUCTION

Contamination is a near universal unpleasant feeling that can be separated into two distinct, albeit related, domains (Rachman, 2004; Coughtrey et al., 2012; Rachman et al., 2015). Contact contamination occurs when there are concerns of dirtiness, endangerment, infection, or pollution following physical contact with a source. Mental contamination—the focus of the present research—typically arises in the absence of direct physical contact with a source (Rachman, 2004; Rachman et al., 2015). Images, memories, and thoughts are common sources of mental contamination (e.g., Fairbrother et al., 2005; Herba and Rachman, 2007; Elliott and Radomsky, 2009, 2012; Rachman et al., 2012). Mental contamination ranges along a continuum of severity and, thus, typically is best conceptualized dimensionally (Radomsky et al., 2018), with prior investigations using a full range of severity scores (e.g., Elliott and Radomsky, 2009, 2013; Radomsky and Elliott, 2009; Rachman et al., 2012; Brake et al., 2018; Jacoby et al., 2018; Ojserkis et al., 2018).

#### Edited by:

Adrian Wells, The University of Manchester, United Kingdom

#### Reviewed by:

Henrik Nordahl, Norwegian University of Science and Technology, Norway Costas Papageorgiou, Priory Hospital Altrincham, United Kingdom Samuel Myers, Bar-Ilan University, Israel

#### \*Correspondence:

Thomas A. Fergus thomas\_fergus@baylor.edu

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 03 July 2018 Accepted: 04 September 2018 Published: 23 October 2018

#### Citation:

Fergus TA, Clayson KA and Dolan SL (2018) Metacognitive Beliefs Predict Greater Mental Contamination Severity After an Evoking Source. Front. Psychol. 9:1784. doi: 10.3389/fpsyg.2018.01784

Much of the extant research has focused on mental contamination in the context of obsessive-compulsive symptoms, in which individuals report experiencing internal dirtiness following ego-dystonic images or thoughts (e.g., Rachman, 2004; Elliott and Radomsky, 2009; Rachman et al., 2015). Cleansing behavior often is reported in such situations; yet, cleansing behavior ultimately contributes to the persistence of perceptions of dirtiness (Rachman et al., 2015). Despite associations with obsessive-compulsive symptoms, mental contamination likely spans across multiple forms of psychopathology (Blakey and Jacoby, 2018). The relevance of mental contamination to posttraumatic stress following sexual trauma has garnered attention, with existing study findings indicating the relatively common experience of mental contamination among women who survive sexual trauma (e.g., Fairbrother and Rachman, 2004; Fairbrother et al., 2005; Herba and Rachman, 2007; Olatunji et al., 2008; Badour et al., 2013a,b). Mental contamination subsequent to sexual trauma relates to greater posttraumatic stress symptoms (Fairbrother and Rachman, 2004; Olatunji et al., 2008; Badour et al., 2013a,b) and may be particularly relevant to understanding intrusion-related distress associated with traumatic events. For example, feelings of dirtiness may contribute to avoidant coping that maintains mental contamination and distress surrounding images, memories, and thoughts (Coughtrey et al., 2014). Indeed, following sexual trauma, women report mental contamination and avoidant coping (e.g., cleansing behavior) after trauma recall (Fairbrother and Rachman, 2004; Badour et al., 2013a). Identifying factors contributing to mental contamination holds promise for improving our understanding of posttraumatic stress following sexual trauma.

Conceptual models of mental contamination have yet to be fully developed and the purpose of the present research is to provide a preliminary examination as to whether the self-regulatory executive function (S-REF) model (Wells and Matthews, 1994) could serve as a framework for conceptualizing mental contamination. The S-REF model proposes that selfknowledge about coping guides self-regulatory efforts that ultimately maintain and worsen emotional distress (Wells and Matthews, 1994). Metacognitive beliefs (i.e., beliefs about thinking) underlie a particularly deleterious form of selfregulation known as the cognitive attentional syndrome (CAS) within the S-REF model. Threat monitoring and negatively valenced, self-referential thinking (e.g., worry) are hallmark features of the CAS (Wells and Matthews, 1994). The S-REF model has been applied to specific symptomatology, including posttraumatic stress (Wells and Sembi, 2004). The S-REF model proposes that posttraumatic stress symptoms are a normative part of an adaptation process in the acute aftermath of trauma exposure. For example, women commonly experience posttraumatic stress symptoms in the acute aftermath of sexual trauma (Shevlin et al., 2014). Mental contamination commonly is experienced by women following sexual trauma (e.g., Fairbrother and Rachman, 2004) and, thus, could be a relatively normative part of an adaptation process in the acute aftermath of sexual trauma as well.

An important consideration pertains to processes that help maintain mental contamination following sexual trauma, with the S-REF model leading to expectations that metacognitive beliefs contribute to greater mental contamination severity. For example, metacognitive beliefs activate the CAS and, thus, are responsible for "trauma-lock" (Wells and Sembi, 2004; Wells, 2009). Trauma-lock is a byproduct of the CAS that involves trauma perseveration. Supporting the potential relevance of this process to mental contamination are results that trauma reminders evoke mental contamination (Fairbrother and Rachman, 2004; Badour et al., 2013a) and re-evoking mental contamination contributes to its persistence (Coughtrey et al., 2014). Greater mental contamination severity would thus be expected to occur following trauma perseveration, which, according to the S-REF model, occurs because of metacognitive beliefs.

The S-REF model further holds that trauma-lock contributes to negative interpretations of symptoms, which often take the form of metacognitive beliefs (Wells and Sembi, 2004; Wells, 2009). For example, individuals may endorse beliefs such as "It's not normal to keep thinking about the trauma" or "I could lose my mind if I continue to think this way" (Wells, 2009). Such beliefs commonly are termed negative metacognitive beliefs because the beliefs relate to the uncontrollability or danger of thinking (Wells, 2000). Extant research supports negative metacognitive beliefs as being particularly relevant to posttraumatic stress (Bennett and Wells, 2010; Fergus and Bardeen, 2017a), thus highlighting the possible relevance of those specific metacognitive beliefs to mental contamination following trauma exposure. Other researchers have similarly raised the possibility that negative metacognitive beliefs underlie mental contamination (e.g., "If I cannot control my repugnant, repulsive thoughts I will go crazy," Radomsky et al., 2018). The S-REF model posits that negative metacognitive beliefs contribute to negative appraisals of symptomatology, thereby contributing to responses (e.g., worry and other CAS-relevant avoidant coping) that block emotional processing and result in greater threat perception. Underlying metacognitive beliefs are strengthened (e.g., about threat detection, danger of thinking) and traumalock is, thus, maintained (Wells and Sembi, 2004; Wells, 2009). The individual consequently experiences heightened emotional distress, possibly inclusive of mental contamination, because underlying metacognitive beliefs continue to fuel the process (e.g., trauma perseveration, negative interpretation of symptoms).

As noted, conceptual models of mental contamination have yet to be fully developed. Other researchers offered a preliminary cognitive conceptualization of mental contamination, which chiefly focuses on content-based self-appraisals related to responsibility and violation (Rachman et al., 2015; Radomsky et al., 2018). That conceptualization diverges from the S-REF model, as the S-REF model proposes that the impact of such appraisals on emotional distress is the result of metacognitive beliefs and the CAS (Wells, 2000). The present study sought to provide preliminary support for the S-REF model as a framework for conceptualizing mental contamination by examining metacognitive beliefs as a predictor of mental contamination severity following an evoking source among women experiencing sexual trauma. This particular sample

composition was chosen because of the reviewed literature indicating that mental contamination is particularly salient for women experiencing sexual trauma.

It was expected that metacognitive beliefs would positively relate to mental contamination severity following the evoking source. In addition to examining bivariate relations, multivariate analyses examined the robustness of those relations by statistically controlling for the effects of theoretically relevant covariates. Examined covariates included trait anxiety, disgust proneness, and posttraumatic stress symptom severity, each of which has shown relevance to mental contamination in prior research (e.g., Badour et al., 2014; Ojserkis et al., 2018). Including covariates in multivariate analyses allowed for an examination as to the incremental explanatory power of metacognitive beliefs in accounting for mental contamination severity. Multivariate analyses also statistically controlled for mental contamination severity before the evoking source to ensure observed effects captured something beyond baseline severity. Following from extant data that beliefs surrounding the danger and uncontrollability of thinking are metacognitive beliefs particularly relevant to posttraumatic stress (Bennett and Wells, 2010; Fergus and Bardeen, 2017a), as well as other indices of emotional distress (e.g., Spada et al., 2008), those metacognitive beliefs were expected to emerge as particularly relevant to mental contamination severity within multivariate analyses.

# MATERIALS AND METHODS

# Participants

A total of 713 undergraduate women at a private Southern United States university were screened for potential participation. Eligibility criteria were women who reported personally experiencing sexual trauma on the Life Events Checklist for DSM-5 (LEC-5; Weathers et al., 2013a). More precisely, eligibility involved women who endorsed directly experiencing sexual assault or another unwanted or uncomfortable sexual experience on the LEC-5. A broad definition of sexual trauma was used following findings that women may resist endorsing experiencing sexual trauma when questions contain stigmatized terminology, such as "rape" (e.g., Resnick et al., 1993). A total of 206 women were eligible for participation (28.9% of the total screened sample), a percentage consistent with the lifetime prevalence of sexual trauma found in undergraduate samples of women (Frazier et al., 2009).

Of those 206 eligible participants, 102 participated in the lab-based session (49.5% of eligible participants). The average age of those 102 women was 19.4 years (SD = 3.1, range 18–38), with 56.9% self-identifying as White, 16.7% as Latina, 9.8% as multi-racial, 8.8% as Black, 5.9% as Asian, and 1.9% as "other" ethnicity or race. There were no significant age (t(204) = 0.76, p = 0.451) or ethnoracial (χ 2 (5) = 1.99, p = 0.851) differences between women who were eligible and did versus did not participate. There also were no significant differences on any of the study variable scores reported below between women who were eligible and did versus did not participate (magnitude of t(204) ranged from 0.66 to 1.31, ps > 0.193).

# Measures

## Metacognitions Questionnaire-30 (MCQ-30; Wells and Cartwright-Hatton, 2004)

The MCQ-30 is a 30-item short form of the 65-item MCQ (Cartwright-Hatton and Wells, 1997). Both MCQ versions assess the same five metacognitive beliefs: (a) positive beliefs about worry; (b) negative beliefs about the uncontrollability and danger of thoughts; (c) cognitive confidence; (d) need for control; and (e) cognitive self-consciousness. The distinctiveness of the five metacognitive beliefs of the MCQ-30 has since been replicated (Spada et al., 2008; Fergus and Bardeen, 2017b). MCQ-30 items are rated using a 4-point scale (ranging from 1 to 4). The MCQ-30 scales show approximately 5-week test-retest correlation coefficients ranging from 0.59 to 0.79 (Wells and Cartwright-Hatton, 2004). The MCQ-30 scales showed adequate to good internal consistency in the present study (Cronbach's αs ranging from 0.76 to 0.91).

# State Trait Inventory for Cognitive and Somatic Anxiety (STICSA; Ree et al., 2008)

The STICSA is a 21-item self-report measure of anxiety using separate state and trait versions. In regards to trait anxiety, participants rate the degree to which each item indicates how they "generally feel." STICSA items are rated using a 4-point scale (ranging from 1 to 4). The STICSA assesses cognitive and somatic anxiety, with a total score derived by summing the 21 item scores. Higher scores reflect greater trait anxiety. The STICSA shows approximately 7-week test-retest correlation coefficients of 0.60 and 0.66 (Ree et al., 2008). The STICSA showed good internal consistency in the present study (α = 0.88).

# Disgust Propensity and Sensitivity Scale-Revised (DPSS-R; van Overveld et al., 2006)

The DPSS-R is a 16-item self-report measure of disgust proneness, conceptualized as the propensity to experience disgust and negative appraisals of disgust. DPSS-R items are rated using a 5-point scale (ranging from 1 to 5). A 12-item version that improves upon the factorial validity of the measure was used (Fergus and Valentiner, 2009). Higher scores reflect greater disgust proneness. The DPSS-R shows approximately 8-week test-retest correlation coefficients of 0.69 and 0.67 (van Overveld et al., 2006). The DPSS-R showed good internal consistency in the present study (α = 0.85).

# PTSD Checklist for DSM-5 (PCL-5; Weathers et al., 2013b)

The PCL-5 is a 20-item self-report measure that assesses posttraumatic stress symptoms following PTSD criteria in the DSM-5 (American Psychiatric Association, 2013). PCL-5 items are endorsed using a 5-point scale (ranging from 0 to 4). A total score is derived by summing intrusion, hyperarousal, avoidance, and negative alterations in cognition and mood symptoms over the past month. Higher scores reflect greater

symptom severity. The PCL-5 shows an approximately 1 week test-retest correlation coefficient of 0.82 (Blevins et al., 2015). The PCL-5 showed good internal consistency in the present study (α = 0.93). Participants completed the PCL-5 in relation to the LEC-5 event that currently bothered them the most<sup>1</sup> .

## State Mental Contamination Scale (SMCS; Lorona et al., 2018)

The SMCS is a 15-item self-report measure that assesses state mental contamination and was developed to parallel the items of the trait measure of mental contamination known as the Vancouver Obsessive-Compulsive Inventory-Mental Contamination Scale (VOCI-MC; Radomsky et al., 2014). Lorona et al. (2018) reworded 15, of the 20, items from the VOCI-MC so that the timeframe of the SMCS items reflected the present moment. The remaining five VOCI-MC items were not conducive to rewording to the present moment and were dropped from the item pool. SMCS items are rated using a 5-point scale (ranging from 1 to 5). A total score is derived by summing the 15 item scores. Higher scores indicate greater state mental contamination and the SMCS showed good internal consistency in the present study (α = 0.94).

# Procedure

The local institutional review board approved the study protocol. Separate informed consent processes were completed before the online and lab-based session. The LEC-5 was completed online to determine study eligibility. Participants also completed the MCQ-30, STICSA, DPSS-R, and PCL-5 during the online session. The self-report measures were completed during a separate session to ensure the study activities in the labbased session did not inadvertently influence responses to the self-report measures. Moreover, completing the self-report measures during the separate online session helped reduce the likelihood that responses to those activities influenced responses in the lab-based session. There was an average of 22 days (SD = 16) between the online and the individual lab-based session<sup>2</sup> . Each eligible participant was invited to participate in the lab-based session through an e-mail, of which, as noted, only a subset of eligible participants signedup for the later study session. Eligible participants who attended the lab-based session initially completed an item asking about current feelings of dirtiness rated using a 0 to 100 scale, with 100 representing the greatest severity, to assess baseline mental contamination severity. Participants completed that same item again following the evoking task for purposes of a manipulation check (e.g., Elliott and Radomsky, 2009).

For the evoking task, participants completed the "dirty-kiss" task (Elliott et al., 2008) in which they listened to an audio recording through headphones that instructed them to imagine attending a party with a friend. At the party, participants imagined receiving a non-consensual kiss from a male described as possessing disgusting qualities. The recording ends with the friend asking, "How did you end up kissing that guy?" and participants then take off the headphones. This task has been used in prior research to evoke mental contamination (e.g., Elliott and Radomsky, 2012). Immediately following task completion, participants completed the SMCS. Participants then completed items related to the ease of imagining, vividness, and realism of the scenario using a 0 to 100 scale, with higher scores indicating greater ease imagining, vividness, and realism. Ratings indicated a high degree of ease imagining (M = 84.25, SD = 22.09), vividness (M = 82.41, SD = 19.22), and realism (M = 75.74, SD = 25.41) in the present study. Participants were then debriefed. Participants received partial course credit for their participation in both the online and labbased session.

# RESULTS

# Preliminary Analyses

A paired-samples t-test was used to examine the effectiveness of the dirty-kiss task by comparing feelings of dirtiness from before, M = 18.32, SD = 22.14, and after, M = 65.95, SD = 28.82, the task. That analysis indicated a significant increase in dirtiness ratings, t(101) = 15.03, p < 0.001, and the effect was large in magnitude, Cohen's d = 1.86. The task had its intended effect.

Descriptive statistics and zero-order correlations among the study variables are presented in **Table 1**. The maximum magnitude values for skewness (baseline feelings of dirtiness: 1.09) and kurtosis (PCL-5: 0.94) of the study variables were below levels typically considered elevated (i.e., | 2| ; Bandalos, 2018). As such, the distributions of scores did not appear to substantively deviate from normality. The metacognitive beliefs generally significantly intercorrelated with the covariates, save for baseline feelings of dirtiness only correlating with cognitive confidence. As predicted, metacognitive beliefs generally positively correlated with mental contamination severity following the evoking source (i.e., SMCS scores). Those correlations were small-tomoderate in magnitude. Associations with mental contamination severity following the evoking source were found in relation to negative metacognitive beliefs, cognitive confidence, and the need for control. Because positive metacognitive beliefs and cognitive self-consciousness did not correlate with mental contamination severity following the evoking source, the positive metacognitive beliefs and cognitive self-consciousness scales of the MCQ-30 were dropped from multivariate analyses (e.g., Thielsch et al., 2015). In addition, among the covariates,

<sup>1</sup>Approximately half of the sample (i.e., 48%) indicated that sexual trauma was the most bothersome LEC-5 event and no other LEC-5 event was endorsed as the most bothersome by more than 10% of the sample. SMCS scores following the evoking task did not significantly differ based upon whether sexual trauma was the most distressing LEC-5 event (t(100) = 1.32, p = 0.192). Negative metacognitive beliefs continued to share an association with SMCS scores when including the most distressing LEC-5 event (sexual trauma versus non-sexual trauma) as a covariate (β = 0.34, p = 0.015, in final block of regression analysis). The most distressing LEC-5 event (sexual trauma versus non-sexual trauma) did not moderate the association between negative metacognitive beliefs and SMCS scores (β = -0.18, p = 0.590, for the interaction term).

<sup>2</sup>Days between study sessions did not correlate with any of the study variables (magnitude of rs ranging from 0.01 to 0.12, ps > 0.252).

#### TABLE 1 | Descriptive statistics and zero-order correlations.

fpsyg-09-01784 October 22, 2018 Time: 17:26 # 5


N = 102. ∗∗p < 0.01, <sup>∗</sup>p < 0.05 (two-tailed). MCQ, Metacognitions Questionnaire (P, Positive; N, Negative; CC, Cognitive Confidence; NC, Need for Control; CSC, Cognitive Self-Consciousness); STICSA, State-Trait Inventory of Cognitive and Somatic Anxiety; DPSS-R, Disgust Propensity and Sensitivity Scale-Revised; PCL-5, PTSD Checklist for DSM-5; SMCS, State Mental Contamination Scale.

only trait anxiety and baseline feelings of dirtiness correlated with mental contamination severity following the evoking source. As such, disgust proneness and posttraumatic stress symptoms were dropped as covariates from multivariate analyses.

## Block 4, baseline feelings of dirtiness and negative metacognitive beliefs were the only significant statistical predictors.

# DISCUSSION

# Regression Analyses

A hierarchical multiple linear regression was used to examine the unique variance accounted for by metacognitive beliefs in mental contamination severity following the evoking source (i.e., SMCS scores). The retained covariates (trait STICSA, baseline feelings of dirtiness) were entered into Block 1 of the model. The retained metacognitive variables were entered into subsequent blocks in descending order based upon the magnitude of zeroorder correlations with mental contamination severity following the evoking source. As such, negative metacognitive beliefs from the MCQ-30 were entered into Block 2, need for control from the MCQ-30 was entered into Block 3, and cognitive confidence from the MCQ-30 was entered into Block 4. The maximum variance inflation factor (VIF) among the predictors in the regression analysis was 2.31, well below conventional guidelines for indicating problems with multicollinearity (>10; Cohen et al., 2003). The maximum Cook's D value was 0.08, well below conventional guidelines for indicating the presence of an overly influential case on the regression model (>1.0; Cohen et al., 2003). The maximum Mahalanobis distance value was 14.95 and, thus, there were no values at or above the respective critical value for indicating multivariate outliers (χ 2 (5) = 20.52, p < 0.001; Mertler and Vannatta, 2005).

The variance accounted for in mental contamination severity following the evoking source and standardized beta weights from the regression analysis are presented in **Table 2**. As shown, the covariates collectively accounted for 14% of variance in mental contamination severity in Block 1. Adding negative metacognitive beliefs to the model in Block 2 accounted for an additional 8% of variance in mental contamination severity. Adding metacognitive beliefs related to need for control and cognitive confidence in Block 3 and Block 4, respectively, did not explain additional variance in mental contamination severity. In The present study sought to provide a preliminary examination of the S-REF model (Wells and Matthews, 1994) as a framework for conceptualizing mental contamination by investigating metacognitive beliefs as predictors of mental contamination severity. An S-REF model applied to posttraumatic stress (Wells and Sembi, 2004) was the selected framework for the present study given the frequent occurrence of mental contamination following sexual trauma. Women who experienced sexual trauma completed a self-report measure of metacognitive beliefs and later completed a task that evoked mental contamination. Consistent with study predictions, metacognitive beliefs generally positively correlated with mental contamination severity following the evoking task. In bivariate analysis, metacognitive beliefs related to the uncontrollability and danger of thoughts, cognitive confidence, and need for control shared small-to-moderate correlations with mental contamination severity. However, only negative metacognitive beliefs (i.e., uncontrollability and danger of thoughts) related to mental contamination severity in multivariate analyses, suggesting, as predicted, that those metacognitive beliefs are particularly relevant to mental contamination.

The association between negative metacognitive beliefs and mental contamination severity is notable because it was found even while statistically controlling for baseline mental contamination severity, trait anxiety, and interrelations among other metacognitive beliefs. Disgust proneness and posttraumatic stress symptom severity were not included as covariates in multivariate analyses, as those two variables unexpectedly did not correlate with mental contamination severity following the evoking source in the present study. That pattern of findings stands in contrast to prior findings that disgust and posttraumatic stress symptoms are associated with changes in feelings of dirtiness from before to after an evoking source among women experiencing sexual trauma (Badour et al., 2014). Sample



N = 102. ∗∗p < 0.01, <sup>∗</sup>p < 0.05 (two-tailed). STICSA, State-Trait Inventory of Cognitive and Somatic Anxiety; MCQ, Metacognitions Questionnaire (N, Negative; NC, Need for Control; CC, Cognitive Confidence).

composition could be one reason for the discrepant findings, as Badour et al. (2014) used a narrower group of respondents experiencing sexual trauma (i.e., women reporting sexual trauma exposure and denied history of physical assault) than the present study. Assessment method could be another reason for discrepant findings, as Badour et al. assessed mental contamination via feelings of dirtiness alone. Although that assessment method is common (e.g., Elliott and Radomsky, 2009), and was included as a manipulation check in the present study, Radomsky et al. (2014) contend that mental contamination is more fully represented through aspects other than feelings of dirtiness. The state measure of mental contamination used in the present study follows Radomsky et al.'s contention via conceptualizing mental contamination as broader than feelings of dirtiness alone (Lorona et al., 2018). Although tenable possibilities, future research examining the contribution of disgust proneness and posttraumatic stress symptoms to in-vivo experiences of mental contamination appears warranted before firmer conclusions about those interrelations are drawn.

Negative metacognitive beliefs were expected to be the metacognitive beliefs particularly relevant to mental contamination following from extant findings linking those metacognitive beliefs to posttraumatic stress (Bennett and Wells, 2010; Fergus and Bardeen, 2017a). In the context of posttraumatic stress, beliefs about the uncontrollability and danger of thoughts putatively lead to threatening interpretations of symptoms that contribute to emotional distress (Wells and Sembi, 2004; Wells, 2009). Images, memories, and thoughts are common sources of mental contamination (e.g., Fairbrother et al., 2005; Herba and Rachman, 2007; Elliott and Radomsky, 2009, 2013; Rachman et al., 2012). Following from the S-REF model, negative interpretations of symptoms would be expected to increase the likelihood of unwanted images, memories, and thoughts occurring (Wells and Sembi, 2004; Wells, 2009) and, thereby, re-evoke mental contamination. Prior research indicates that re-evoking mental contamination contributes to its persistence (Coughtrey et al., 2014). Additionally, negative metacognitive beliefs could contribute to the engagement in avoidant behavior in an attempt to regulate thoughts, with that behavior ultimately blocking emotional processing and maintaining distress (Wells, 2000). Cleansing behavior is a common type of avoidant behavior reported in the context of mental contamination (Rachman et al., 2015) and future research should seek to examine whether negative metacognitive beliefs are associated with cleansing behavior following a mental contamination provocation.

Conceptual models of mental contamination have yet to be fully developed, with existing cognitive conceptualizations emphasizing the role of negative appraisals in relation to mental contamination (Rachman et al., 2015; Radomsky et al., 2018). Whereas such conceptualizations do not preclude the consideration of metacognitive beliefs, existing examinations of the relevance of cognitive variables to mental contamination have tended to focus on content-based self-appraisals related to responsibility and violation (e.g., Radomsky and Elliott, 2009; Elliott and Radomsky, 2013). The S-REF model would lead to predictions that mental contamination is not the result of such content-based self-appraisals (e.g., "I am pathetic, weak, hopeless," Radomsky et al., 2018), but is the result of metacognitive beliefs and the CAS (Wells, 2000). Content-based self-appraisals, unfortunately, were unexamined. Future research that concurrently examines content-based self-appraisals and metacognitive beliefs will aid in elucidating the degree to which those variables incrementally contribute to our understanding of mental contamination.

Additional support for the S-REF model as a tenable framework for conceptualizing mental contamination could come from future research findings that content-based selfappraisals do not account for unique variance in mental contamination severity once statistically controlling for metacognitive beliefs, such as the uncontrollability and danger of thoughts, or the CAS. Such patterns of findings have emerged in prior studies examining obsessive-compulsive symptoms (e.g., Myers et al., 2009; Solem et al., 2010). It is important to note that content-based self-appraisals can initiate self-regulatory efforts in the form of the CAS (Wells, 2000). It is thus possible that the

relationship between content-based self-appraisals and mental contamination depends upon metacognitive beliefs or the CAS serving as moderators. Indeed, extant research supports that possibility when considering the relation between content-based self-appraisals and the frequency of ego-dystonic intrusive thoughts (Fergus and Wu, 2010). Another possibility is that the impact of the types of beliefs on mental contamination differs across time, which could be examined in future longitudinal research.

Future research supporting the relevance of metacognitive beliefs to mental contamination would point to potential treatment strategies when seeking to reduce mental contamination. For example, such patterns of findings may point to a focus on the metacognitive mode in which intervention strategies chiefly target how one relates to cognitive events (Wells, 2000). A greater focus on the metacognitive mode, rather than the object mode, in which the content of appraisals are evaluated for their accuracy, could be preferred when seeking to reduce mental contamination. The present results indicate the relevance of negative metacognitive beliefs to mental contamination. Intervention strategies relevant to mitigating negative metacognitive beliefs about the uncontrollability and danger of thoughts include verbal reattribution, behavioral experiments, and detached mindfulness (Wells, 2009). Detached mindfulness seeks to promote the metacognitive mode by having individuals consider themselves as an observer separate from their thoughts to facilitate suspension of conceptual processing and the alteration of metacognitive beliefs (Wells, 2009). Current treatment efforts for mental contamination are in their relative infancy (Coughtrey et al., 2013) and future research may seek to examine the usefulness of metacognitive intervention strategies in the reduction of mental contamination severity.

Study limitations must be considered. As previously reviewed, the present study focused on women experiencing sexual trauma given that mental contamination is particularly salient for these individuals. Indeed, the evoking task produced a large increase in mental contamination severity among study participants. However, the inclusion criteria was broad in that other types of trauma exposure were not restricted. In addition, there was lack of available information on the nature of the sexual trauma and the frequency of sexual trauma was not assessed. The generality of the present findings to women experiencing sexual trauma would thus be strengthened through determining study eligibility following a more in-depth assessment of sexual trauma exposure. A large number of eligible participants did not participate in the lab-based session. Although eligible participants who did versus did not participate in the lab-based session did not differ on demographic information or study variable scores, it is possible that the subset of participants who attended the labbased session differed in some unknown ways from participants who did not attend that study session. Trauma exposure is common among college students (Frazier et al., 2009) and, yet, the generality of the findings would be further strengthened by examining the relation between metacognitive beliefs and mental contamination among community respondents. The present study was adequately powered (1 – β = 0.80) to detect smallsized effects in the examined regression model (Cohen's f 2 ≈0.08; Aiken and West, 1991), as determined using a post hoc power analysis (Faul et al., 2009). Future research replicating and extending the findings with larger samples nonetheless appears warranted (e.g., Schönbrodt and Perugini, 2013).

Studies commonly examine mental contamination among women (e.g., Elliott and Radomsky, 2009, 2013; Radomsky and Elliott, 2009; Badour et al., 2013a,b, 2014). Nevertheless, men experience mental contamination as well (e.g., Coughtrey et al., 2012). A limitation of the dirty-kiss task is that it is most appropriate for women (Elliott and Radomsky, 2009). Future research should thus seek to use alternative methods for evoking mental contamination (De Putter et al., 2017) in order to replicate the present findings among samples consisting of both sexes. By using alternative evoking sources and other samples, future research can help address whether metacognitive beliefs generally account for mental contamination severity or whether the impact of those beliefs seems most relevant in the context of posttraumatic stress. The study methods precluded the consideration of causal relations between metacognitive beliefs and mental contamination. Future longitudinal and experimental research is needed to address if negative metacognitive beliefs causally influence mental contamination. The self-report measures of the statistical predictors were completed, on average, 22 days before the completion of the lab-based session, which was done to ensure the study activities in the lab-based session did not inadvertently influence responses to the self-report measures or vice versa. As discussed, scores on the self-report measures of the statistical predictors have evidenced stability estimates considered moderate to high for trait variables (e.g., Roberts et al., 2008) in prior research. Nonetheless, interrelations between mental contamination and the other study variables may have been impacted by the gap between study sessions.

The examined variables, collectively, accounted for about 22% of the variance in mental contamination severity, with negative metacognitive beliefs accounting for about 8% of unique variance. Additional variables to consider in future research include content-based self-appraisals (e.g., Radomsky and Elliott, 2009) and markers of the CAS (e.g., rumination, worry; Wells, 2000). It is possible that variables from other metacognitive models could be useful in accounting for additional variance in mental contamination severity. Links between mental contamination and obsessive-compulsive symptoms (e.g., Rachman, 2004; Rachman et al., 2015) highlight the possibility that variables from the metacognitive model of obsessive-compulsive symptoms (Wells, 2000) warrant consideration. Potentially relevant variables from that model include thought-fusion beliefs, beliefs about rituals, and stop signals.

Limitations notwithstanding, the present results provide support for the relevance of metacognitive beliefs to mental contamination. Negative metacognitive beliefs surrounding the uncontrollability and danger of thoughts accounted for unique variance in mental contamination severity following an evoking source. Continued support for a link between metacognitive beliefs and mental contamination could further support the S-REF model as a potential framework for conceptualizing mental contamination, and may ultimately lead to the use of intervention strategies that target those beliefs when seeking to reduce mental contamination.

# ETHICS STATEMENT

fpsyg-09-01784 October 22, 2018 Time: 17:26 # 8

This study was carried out in accordance with the recommendations of the Committee for Protection of Human Subjects at Baylor University with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the Committee for Protection of Human Subjects at Baylor University.

# REFERENCES


# AUTHOR CONTRIBUTIONS

TF, KC, and SD conceptualized the study. KC oversaw the data collection. TF completed the study analyses and wrote the first draft of the manuscript. KC and SD provided the feedback on that draft. TF incorporated that feedback into the submitted version, with KC and SD agreeing on the submitted version.

# ACKNOWLEDGMENTS

Publication was made possible, in part, by support from the Open Access Fund sponsored by the Baylor University Libraries.



of female victims of sexual trauma. Psychol. Assess. 26, 1155–1161. doi: 10.1037/ a0037272


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Fergus, Clayson and Dolan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Executive Control Deficits Potentiate the Effect of Maladaptive Metacognitive Beliefs on Posttraumatic Stress Symptoms

Joseph R. Bardeen<sup>1</sup> \* and Thomas A. Fergus<sup>2</sup>

<sup>1</sup> Department of Psychology, Auburn University, Auburn, AL, United States, <sup>2</sup> Baylor Psychology and Neuroscience Department, Baylor University, Waco, TX, United States

#### Edited by:

Adrian Wells, The University of Manchester, United Kingdom

#### Reviewed by:

David Rosenbaum, Universitätsklinikum Tübingen – Universität Tübingen, Germany Odin Hjemdal, Norwegian University of Science and Technology, Norway

> \*Correspondence: Joseph R. Bardeen jbardeen@auburn.edu

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 06 June 2018 Accepted: 18 September 2018 Published: 08 October 2018

#### Citation:

Bardeen JR and Fergus TA (2018) Executive Control Deficits Potentiate the Effect of Maladaptive Metacognitive Beliefs on Posttraumatic Stress Symptoms. Front. Psychol. 9:1898. doi: 10.3389/fpsyg.2018.01898 The metacognitive model and recent preliminary research suggests that metacognitive beliefs (i.e., beliefs about thinking) may be particularly important for understanding the pathogenesis of posttraumatic stress (PTS). The metacognitive model also suggests that deficits in executive control (i.e., metacognitive control) may increase the impact of metacognitive beliefs on PTS symptoms. Trauma-exposed adult participants (N = 469), recruited through an online crowdsourcing website, completed a battery of measures assessing the constructs of interest. As predicted, deficits in executive control strengthened the positive association between metacognitive beliefs and PTS symptoms. This effect was found in relation to positive (e.g., "Worrying will keep me safe"), but not negative (e.g., "My thoughts are uncontrollable"), metacognitive beliefs. Supplemental analyses, indicated that the interaction between positive metacognitive beliefs and executive control significantly predicted all PTS cluster scores (i.e., Intrusion, Cognition, Arousal, Avoidance). Taken together, results support the proposal that executive control deficits potentiate the effect of metacognitive beliefs on PTS symptoms. Intervention strategies designed to strengthen executive control (e.g., the attention training technique) may be useful in treating individuals with PTS.

Keywords: metacognition, posttraumatic stress, trauma, cognitive control, executive control

# INTRODUCTION

A large majority of the U.S. population will be exposed to one or more traumatic events at some point in the lifespan (Breslau, 2009), and approximately 6 to 8% of the U.S. population (i.e., Kessler et al., 2005; Kilpatrick et al., 2013), as well as 5–20% of returning military personnel (Ramchand et al., 2010), will develop posttraumatic stress disorder (PTSD; Diagnostic and Statistical Manual of Mental Disorders [DSM-5]; American Psychiatric Association [APA], 2013) following trauma exposure. Cognitive-behavioral therapy (CBT) and its underlying theoretical model has resulted in advances to our understanding and treatment of PTSD, thereby aiding in the reduction of the substantial personal and societal burden associated with PTSD (Amaya-Jackson et al., 1999; Brady et al., 2000). Nevertheless, approximately 41% of individuals with PTSD will be classified as non-responders following CBT (McDonagh et al., 2005), and this number may be as high as 72% in community clinical settings (Zayfert et al., 2005).

Whereas cognitive-behavioral models of PTSD suggest that the content of one's thoughts (e.g., beliefs about the self, others, and the world) play a central role in conceptualizing and treating PTSD (e.g., Foa and Rothbaum, 1998; Ehlers and Clark, 2000), the metacognitive model posits that PTSD develops as a function of one's beliefs about thinking (i.e., metacognitive beliefs), and subsequent maladaptive coping (i.e., the cognitive attentional syndrome [CAS]), rather than content-based cognitive themes (Wells and Sembi, 2004; Wells, 2009). Consistent with this conceptualization of PTSD, empirical evidence suggests that metacognitive beliefs may be more important than traumarelated thought content in the pathogenesis of PTS. For example, Fergus and Bardeen (2017a) found that associations between content-specific beliefs and PTS symptoms were attenuated or rendered non-significant after accounting for metacognitive beliefs in a sample trauma-exposed community adults (N = 299).

Within the metacognitive model of PTSD (Wells and Sembi, 2004; Wells, 2009), PTS symptoms are viewed as a normative part of an adaptation process in the acute aftermath of trauma exposure. The CAS (i.e., heightened self-focused attention and threat monitoring, as well as use of rumination, worry, or other avoidant coping strategies), which is initiated and maintained by metacognitive beliefs, is thought to account for the duration and severity of PTS symptoms following trauma exposure. Positive metacognitive beliefs (e.g., "Worrying helps me to avoid problems in the future") are thought to lead to the use of the CAS following trauma exposure. Positive metacognitive beliefs are strengthened when feared outcomes do not occur. Negative metacognitive beliefs surrounding the uncontrollability and danger of thoughts (e.g., "If I could not control my thoughts, I would not be able to function") are thought to increase attention toward internal experience (e.g., monitoring of thought content) and the likelihood that thought processes will be perceived as distressing. Negative metacognitive beliefs are maintained and strengthened because avoidant coping increases vigilance toward internal experience and PTS symptoms, thereby strengthening beliefs about the uncontrollability and danger of thoughts.

Despite promising preliminary findings that metacognitive beliefs may be more relevant to PTS symptoms than contentspecific beliefs (Fergus and Bardeen, 2017a), relatively few studies have reported associations between metacognitive beliefs and PTS symptoms and findings have been mixed. In two studies, significant positive associations that were medium to large in size were reported between metacognitive beliefs and PTS symptoms (Roussis and Wells, 2006; Fergus and Bardeen, 2017a). In contrast, Bennett and Wells (2010) did not observe an association between positive metacognitive beliefs and PTS symptoms (r = −0.01), while a medium-large sized association was observed between negative metacognitive beliefs and PTS symptoms (r = 0.41). These discrepancies may be the result of using different measures of metacognitive beliefs. Specifically, the measure used by Bennett and Wells (2010) focuses on metacognitive beliefs related to memory, the measure used by Fergus and Bardeen (2017a) assesses metacognitive beliefs broadly, but in relation to trauma, and the measure used by Roussis and Wells (2006) broadly assesses metacognitive beliefs without on an emphasis on trauma. Apparent discrepancies in these associations may also be a function of failing to account for a third variable (i.e., moderator variable) that alters the strength of the relationship between PTS symptoms and metacognitive beliefs. A moderator impacts the strength of the relation between two other variables and can help explicate under what conditions the two variables relate to one another (Hayes, 2018). As described below, executive control (i.e., metacognitive control: Wells and Matthews, 1996; Wells, 2009) may be one such potential moderator that helps explicate when metacognitive beliefs relate to PTS symptoms.

Executive control relies on of a variety of top-down cognitive abilities that are associated with activation in the prefrontal cortex (e.g., inhibition, set shifting, working memory updating, error detection, and strategy formulation; Fernandez-Duque et al., 2000). Within the metacognitive model, the excessive conceptual processing that characterizes the CAS is thought to be exacerbated by deficits in executive control that reduce the likelihood that one can effectively disengage from internal experience (e.g., worry, rumination, and other forms of selffocused attentional processes) and maintain attentional focus on adaptive goal-relevant pursuits (i.e., value-driven behavior). The importance of considering deficits in executive control when conceptualizing PTSD from a metacognitive perspective is highlighted by the fact that a technique was developed to specifically address these deficits in metacognitive therapy. Specifically, the attention training technique (Wells, 1990) was developed to strengthen executive control processes that can be used to interrupt the excessive self-focused, threat-based processing that characterizes the CAS (Wells, 2009). Despite the conceptual importance of executive control to the metacognitive model, the impact of this construct on the relationship between metacognitive beliefs and PTS symptoms has yet to be empirically examined.

# PRESENT STUDY

The purpose of the present study was to examine executive control as a moderator of the relationship between metacognitive beliefs and PTS symptoms in a trauma-exposed sample of adults. Following from the empirical evidence described above, we predicted that both positive and negative maladaptive metacognitive beliefs would be associated with PTS symptoms, and the association between negative metacognitive beliefs and PTS symptoms would be the largest in magnitude (Roussis and Wells, 2006; Fergus and Bardeen, 2017a). Additionally, based on evidence showing that individuals with PTSD exhibit relative deficits in the cognitive abilities associated with executive control (Deppermann et al., 2014; Scott et al., 2015), we predicted that executive control deficits would be positively associated with PTS symptoms. Importantly, based on metacognitive theory (Wells, 2009), which suggests that CAS-based coping is exacerbated by deficits in executive control, we predicted that the magnitude of the positive association between maladaptive metacognitive beliefs (i.e., positive and negative) and PTS symptoms would become significantly stronger as deficits in executive control increased. Finally, we conducted an exploratory

analysis examining significant domain-specific interaction effects in the context of PTS cluster scores (i.e., clusters B [Intrusion], C [Avoidance], D [Cognition], and E [Arousal]: DSM-5 PTSD, American Psychiatric Association [APA], 2013). Given the exploratory nature of these analyses, no a priori hypotheses were made.

# MATERIALS AND METHODS

# Participants and Procedure

A total of 597 adults were recruited via Amazon Mechanical Turk (MTurk). MTurk is an online labor market where adults from the general population can be recruited to complete questionnaires in exchange for payment. MTurk samples tend to be more demographically diverse than American undergraduate samples (Buhrmester et al., 2011) and a number of studies support the quality of data collected via MTurk (e.g., Behrend et al., 2011; Buhrmester et al., 2011; Shapiro et al., 2013; Paolacci and Chandler, 2014). Recruitment was limited to MTurk users located within the United States and between the ages of 18–65. Additionally, to be included in the present study, participants had to report exposure to a traumatic event (Criterion A: exposure to actual or threatened death, serious injury, or sexual violence) as defined in the DSM-5 (American Psychiatric Association [APA], 2013). The final sample (n = 469) consisted of adults who had experience at least one traumatic event. The average age of the final sample was 35.9 years (SD = 11.0) and the majority were female (61.4%). In regard to race and ethnicity, 83.6% self-identified as White, 7.0% as Black, 6.6% as Asian, 1.7% as American Indian or Alaska Native, 1.1% endorsed "other," and 7.2% of the final sample identified their ethnicity as Hispanic.

A secure online survey program was used to administer informed consent and self-report measures. Participants were informed (via the electronic consent form) of the costs/benefits of study participation, that their responses were confidential, and that they were free to withdraw from the study at any time. After reading the consent form, participants were able to consent to, or opt out of, continued participation by clicking on one of two radio buttons that offered these choices. Upon study completion, participants were debriefed and paid in full. Participants were compensated \$1.50 for completing study questionnaires, an amount consistent with precedence for paying MTurk workers in similar studies (Buhrmester et al., 2011). This study was approved by the local university-based institutional review board.

# Measures

# Metacognitive Questionnaire-30 (MCQ-30)

The MCQ-30 (Wells and Cartwright-Hatton, 2004) is a 30 item measure inclusive of positive metacognitive beliefs about CAS-based coping (e.g., "worrying helps me to avoid problems in the future") and negative metacognitive beliefs about uncontrollability and danger of thinking (e.g., "my worrying is dangerous for me"). Items of the MCQ-30 are rated on a 4-point scale ranging from 1 (do not agree) to 4 (agree very much). Higher scores indicate higher levels of maladaptive metacognitive beliefs. The MCQ-30 has exhibited adequate psychometric properties, including internal consistency, retest reliability, and construct validity (Wells and Cartwright-Hatton, 2004; Spada et al., 2008). Additionally, factor analytic results support use of a total score and subscale scores, and measurement invariance has been observed between men and women (Fergus and Bardeen, 2017b). Internal consistency of the positive and negative MCQ-30 scales was adequate in the present study (α = 0.92 and 0.91, respectively).

# Barkley Deficits in Executive Functioning Scale-Short Form (BDEFS-SF)

The BDEFS-SF (Barkley, 2011) is a 20-item self-report measure designed to identify deficits in executive functioning. Participants are asked to use a 4-point scale (1 = never or rarely to 4 = very often) to indicate how often they exhibit behaviors associated with daily activities that are indicative of executive functioning deficits across five domains (i.e., time management, organization and problem solving, self-restraint, self-motivation, and self-regulation of emotions). The BDEFS has exhibited adequate psychometric properties in previous research, including evidence of internal consistency (Feldman et al., 2013) and criterion-related validity in relation to both self-report (e.g., Attention Deficit/Hyperactivity Disorder; Gray et al., 2014) and performance-based measures (e.g., working memory; Gray et al., 2015). Internal consistency of the BDEFS-SF total score was adequate in the present study (α = 0.95).

# Life Events Checklist for DSM-5 (LEC-5) Extended Version

The LEC-5 (Weathers et al., 2013a) assesses exposure to 17 potentially traumatic events (e.g., sexual assault, motor vehicle accident, and combat). For each event, respondents are asked to indicate whether the event happened to them, they witnessed it, they learned about it, it was part of their job, they are unsure, or the event did not apply to them. For the extended version of the LEC-5, participants are asked to provide a brief narrative of the events endorsed on the screening page. They then answer a series of follow-up questions designed to clarify whether the endorsed events meet Criterion A (e.g., exposure to actual or threatened death, serious injury, or sexual violence; American Psychiatric Association [APA], 2013).

# PTSD Checklist for DSM5-Civilian Version (PCL-5)

The PCL-5 (Weathers et al., 2013b) is a 20-item self-report measure designed to assess symptoms in clusters B (Intrusion), C (Avoidance), D (Cognition), and E (Arousal) of the DSM-5 PTSD criteria (American Psychiatric Association [APA], 2013). Participants were asked to rate how much they have been bothered by each symptom in the past month (0 = not at all to 4 = extremely), with higher scores indicating greater PTS symptoms. Cluster scores were calculated by summing ratings for each item within a particular symptom cluster. Consistent with evidence suggesting that PTSD is a dimensional construct rather than a discrete clinical syndrome (e.g., Ruscio et al., 2002; Forbes et al., 2005; Broman-Fulks et al., 2006), items were summed to create both total and cluster scores. The PCL-5 has demonstrated adequate psychometric properties, including internal consistency, retest reliability over a 1-week period, and convergent and discriminant validity (Blevins et al., 2015). Internal consistency of the total score and subscale scores was adequate in the present study (i.e., total score α = 0.97, subscale scores from 0.88 to 0.93).

# Data Analytic Strategy

fpsyg-09-01898 October 4, 2018 Time: 15:25 # 4

Meng et al. (1992) test for dependent correlations was used to test the hypothesis that the association between negative metacognitive beliefs and PTS symptoms would be larger in magnitude than the association between positive metacognitive beliefs and PTS symptoms. Next, SPSS version 24 (SPSS IBM, New York) was used to conduct a hierarchical regression to test the hypothesized interactive effects. Consistent with Aiken and West (1991), the predictor (i.e., metacognitive beliefs) and moderator (i.e., executive functioning) variables were mean centered and interaction terms were calculated as the product of the moderator and predictor variables. The predictor variables were entered into the first step of the model (negative and positive metacognitive beliefs), the moderator was entered into the second step of the model (executive functioning), and the interaction terms were entered into the third step of the model (executive functioning by negative and positive metacognitive beliefs). PTS symptoms served as the outcome variable in each model. Simple slopes analysis was used to further examine significant interaction effects (Aiken and West, 1991). Simple slopes analysis helps to explicate under what conditions two variables relate to one another (Hayes, 2018). More specifically, simple slopes analysis consists of constructing two simple regression equations in which the relationship between the independent variable and the dependent variable is tested at both high (+1 SD) and low (−1 SD) levels of the moderating variable (i.e., executive functioning).

Next, interaction effects (i.e., positive and/or negative metacognitive beliefs by executive functioning) were examined in the context of PTS clusters scores. Structural equation modeling (SEM) and path analysis were used to conduct this examination, instead of standard regression analysis, because multiple outcome variables (i.e., PTS cluster scores) can be modeled simultaneously in SEM. For each of the two path models, metacognitive beliefs (i.e., positive or negative), executive functioning, and an interaction term (i.e., metacognitive beliefs by executive functioning) served as predictor variables in the model. The four PTS cluster scores served as outcome variables. Each model was tested using Amos software (Version 24; Arbuckle, 2010) and maximum likelihood estimation. All variables were modeled as manifest indicators. Fit statistics were not computed because just-identified models provide perfect fit to the data (Kline, 2016).

# RESULTS

# Bivariate Correlations

Both positive and negative metacognitive beliefs positively correlated with PTS symptoms (see **Table 1**). As predicted, a test of dependent correlations revealed that PTS symptoms correlated significantly more strongly with negative metacognitive beliefs (r = 0.54, p < 0.001) than positive metacognitive beliefs (r = 0.46, p < 0.001, z = 2.07, and p = 0.02). Also of note, a positive association between executive functioning deficits and PTS symptoms was observed (r = 0.57, p < 0.001).

# Predicting Total Posttraumatic Stress Symptoms

An examination of scatterplots (refer to **Supplementary Figures S1**, **S2**) and the Durbin–Watson statistic indicated that the regression assumptions [i.e., additivity and linearity, independent errors (Durbin–Watson statistic = 1.85), homoscedasticity, and normally distributed errors] were met (see Cohen et al., 2003). Moreover, an examination of multivariate outliers suggested that none of the cases exhibited undue influence on the estimates within the regression model (defined as >1 DFFITS<sup>i</sup> ; Cohen et al., 2003). Additionally, multicollinearity statistics were all above recommended levels (tolerance statistics >0.10 and VIF<10; Cohen et al., 2003), thus indicating no robust problems related to multicollinearity.

In the first step of the regression model (adjusted R<sup>2</sup> = 0.34, p < 0.001), positive and negative metacognitive beliefs significantly predicted PTS symptoms (βs = 0.26 and 0.42, respectively, ps < 0.001). In the second step of the model (1R <sup>2</sup> = 0.07, p < 0.001), executive functioning significantly predicted PTS symptoms (β = 0.33, p < 0.001). In the third step of the model (1R <sup>2</sup> = 0.03, p < 0.001), the interaction between positive metacognitive beliefs and executive functioning significantly predicted PTS symptoms (β = 0.18, p < 0.001), but the interaction between negative metacognitive beliefs and executive functioning did not (β = −0.02, p = 0.65). The non-significant interaction term (negative metacognitive beliefs by executive functioning) was removed from the model to provide an accurate interpretation of simple effects for the significant interaction (positive metacognitive beliefs by executive functioning) in simple slopes analysis. Simple slopes analysis revealed a positive association between positive metacognitive beliefs and PTS symptoms that was significant at higher (β = 0.43, p < 0.001), but not lower (β = 0.08, p = 0.18), levels of executive functioning deficits (see **Figure 1**).

# Predicting Posttraumatic Stress Symptom Cluster Scores

Because the positive negative metacognitive beliefs by executive functioning interaction was significant in our primary analytic model, we conducted a path analysis in which the interaction between positive metacognitive beliefs and executive functioning predicted PTS cluster scores. Standardized path coefficients are presented in **Figure 2**. As can be seen in **Figure 2**, positive metacognitive beliefs, executive functioning, and the interaction term (positive metacognitive beliefs by executive functioning) significantly predicted each of the four PTS cluster scores (all ps < 0.05). Following from our primary analysis, the interaction effect was further explored using simple slopes analysis (Aiken and West, 1991). Simple slopes analysis revealed significant positive associations between positive metacognitive beliefs and each PTS cluster score at higher (Intrusion: β = 0.34,


n = 469 trauma-exposed adults; all ps < 0.001. MCQ-30 = Metacognitive Questionnaire-30; BDEFS-SF = Barkley Deficits in Executive Functioning Scale-Short Form; and PCL-5 = PTSD Checklist for DSM-5–Civilian Version.

Avoidance: β = 0.28, Cognition: β = 0.34, and Arousal: β = 0.34, ps < 0.001), but not lower (Intrusion: β = 0.05, p = 0.46, Avoidance: β = 0.10, p = 0.14, Cognition: β = 0.08, p = 0.22, and Arousal: β = 0.09, p = 0.16), levels of executive functioning deficits.

Although the negative metacognitive beliefs by executive functioning interaction was not significant in our primary analytic model, we conducted a second path analysis in which the interaction between negative metacognitive beliefs and executive functioning predicted PTS cluster scores to ensure that the interaction terms did not exhibit significant associations with specific PTS clusters. Negative metacognitive beliefs and executive functioning significantly predicted each of the four PTS cluster scores (Intrusion: β = 0.28 and 0.32, Avoidance: β = 0.33 and 0.16, Cognition: β = 0.27 and 0.37, and Arousal: β = 0.27 and 0.37, ps < 0.01). Consistent with our primary analytic model, the interaction term (i.e., negative metacognitive beliefs by executive functioning) did not significantly predict any of the PTS cluster scores (Intrusion: β = 0.05, Avoidance: β = 0.03, Cognition: β = 0.07, and Arousal: β = 0.07, ps > 0.05).

# DISCUSSION

As predicted, executive control deficits moderated the relationship between maladaptive metacognitive beliefs and PTS symptoms in a sample of trauma-exposed adults. Specifically, as executive control deficits increased, the strength of the association between positive metacognitive beliefs and PTS symptoms also increased. This pattern of findings is consistent with the metacognitive model (Wells, 2009), which suggests that deficits in executive control reduce the likelihood of successfully disengaging from CAS-based coping in response to internal experience (e.g., worry, rumination, and other forms of self-focused attention). These findings are also consistent with evidence that suggests that top-down executive control processes (e.g., inhibition, set shifting, working memory updating) can be used to protect those who are vulnerable to maladaptive psychological outcomes from experiencing such outcomes (Fergus et al., 2012; Jones et al., 2012; Bardeen and Fergus, 2016).

An examination of raw correlations in the present study was consistent with previous research showing that the association between negative metacognitive beliefs and PTS symptoms is larger in magnitude than the association between positive metacognitive beliefs and PTS symptoms (Roussis and Wells, 2006; Fergus and Bardeen, 2017a). However, an aggregate effect of negative metacognitive beliefs and executive control deficits on PTS symptoms was not observed. One explanation for this null result is that the relationship between negative metacognitive beliefs and PTS symptoms is more direct than the relationship between positive metacognitive beliefs and PTS symptoms. That is, the amount of time one has to enact top-down regulatory processes before distress is experienced could be shorter in duration for negative, versus positive, metacognitive beliefs. In support of the proposition, Roussis and Wells (2006) found that the association between positive metacognitive beliefs and PTS symptoms was accounted for by CAS-based coping (i.e.,

use of worry as a thought control strategy), whereas negative metacognitive beliefs had a direct effect on PTS symptoms independent of such coping. Put more succinctly, positive metacognitive beliefs, such as "worrying helps me to avoid problems in the future," are likely to lead to continued processing, but do not necessarily pose an immediate threat. In contrast, negative metacognitive beliefs, such as "my worrying could make me go mad," have a clear sense of urgency, and thus, the buffering effect of executive control may be of less benefit for those whose metacognitive beliefs are primarily about the uncontrollability and danger of thinking.

Another plausible explanation is that the diverse content of the MCQ-30 negative metacognitive beliefs subscale may be partially responsible for the null finding. The negative metacognitive beliefs subscale consists of items denoting either danger or uncontrollability. Following from the hypothesis above, executive control may have an impact on the relationship between negative metacognitive beliefs related to uncontrollability, but not danger, and PTS symptoms. Separately assessing uncontrollability and danger metacognitive beliefs in future research may be beneficial.

An exploratory aim of the present study was to examine significant domain-specific interaction effects in the context of PTS cluster scores. Results of a path analysis indicated that the interaction between positive metacognitive beliefs and executive control significantly predicted all PTS cluster scores (i.e., Intrusion, Cognition, Arousal, and Avoidance). At higher levels of executive control deficits, the magnitude of the associations between positive metacognitive beliefs and each PTS cluster score were similar in size (i.e., 0.28 to 0.34). Given its emphasis on distress associated with intrusive cognitive content, one might hypothesize that the observed interaction effect might be particularly relevant to the Intrusion cluster. However, cognitive content is present in some form for all four of the PTS symptom clusters. Trauma-related thoughts are referenced in the avoidance cluster. Memory difficulties, negative expectations about one's self, others and the world, self- or other-blame, and other internal content make up the Cognitions cluster. And finally, the Arousal cluster references concentration difficulties, as well as hypervigilance toward perceived threat (i.e., CAS threat monitoring; DSM-5: American Psychiatric Association [APA], 2013).

The present results should be considered in light of study limitations. Internet samples of community adults have been used to examine trauma and PTS symptoms in prior research (e.g., Seligowski and Orcutt, 2016). Moreover, evidence supports MTurk as a viable method for collecting data for clinical research (Chandler and Shapiro, 2016) and established quality control methods were used in the present study to improve study data

(e.g., using high reputation MTurk workers; Peer et al., 2014). Nonetheless, MTurk samples are not representative of the general population. As such, replicating study findings in samples with more racial/ethnic diversity, male representation, and higher levels of psychological distress (i.e., clinical samples) will be important in the future to ensure that study findings generalize. Despite utilization of a sample unselected based upon symptom severity, it is important to note that a considerable proportion of the trauma-exposed sample reported the presence of clinically relevant PTS symptoms (i.e., 29% using the more liberal PCL-5 cut score of 28 and 19.2% using the most conservative PCL-5 cut score of 37; Blevins et al., 2015).

The cross-sectional study design may also be considered a study limitation. Future research using longitudinal study designs will help clarify the temporal nature of relations among metacognitive beliefs, executive control deficits, and PTS symptoms. Additionally, experimental designs will be helpful in determining temporal precedence, as well as in determining whether executive functioning deficits are a moderator of the relationship between positive metacognitive beliefs and PTS symptoms, or vice versa. As described, executive control consists of a variety of top-down cognitive processes that are associated with activation in the prefrontal cortex (e.g., inhibition, set shifting, working memory updating, error detection, strategy formulation; Fernandez-Duque et al., 2000). The use of multiple objective measures (e.g., established behavioral assessments) to assess these cognitive processes will be important in future research to determine whether one or more of these specific processes is primarily responsible for the effects observed in the present study. Identification of the specific cognitive deficits that exacerbate the effect of metacognitive beliefs on PTS symptoms may aid in the development of a treatment for PTSD that has a narrower target.

To our knowledge, the present study is the first to provide evidence that executive control modulates the effect of metacognitive beliefs on PTS symptoms. Although evidence supports the use of metacognitive therapy for treating individuals with PTSD (Wells et al., 2015), the attention training technique (Wells, 1990, 2009) remains underutilized as a component of this larger treatment package. Results of the present study, in combination with evidence that the attention training technique reduces symptoms of emotional disorders as a standalone intervention (e.g., Fergus and Bardeen, 2016; Knowles et al., 2016), suggest that using the attention

# REFERENCES


Arbuckle, J. L. (2010). Amos (Version 24.0) [Computer software]. Chicago, IL: SPSS.

training technique to directly target executive control deficits may be an important adjunct to more established PTSD interventions. Moreover, given the applicability of the observed interaction to all four PTS symptom clusters, metacognitive therapy, including the attention training technique, may be particularly well-suited for treating individuals with PTSD.

# ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the Institutional Review Board (IRB) in the Office of Research Compliance at Auburn University. The protocol was approved by the IRB at Auburn University. All participants were provided with a one page description of the study in order to make an informed choice about their participation. Participants were then given the option to participate in the study by clicking a "yes" or "'no" button to indicate their consent. A waiver of documentation of written consent was approved by the IRB at Auburn University.

# AUTHOR CONTRIBUTIONS

JB was involved in study conceptualization, data collection, data analysis and interpretation, and manuscript preparation. TF was involved in study conceptualization and manuscript preparation.

# FUNDING

The authors received no financial support for the research, authorship, and/or publication of this article.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg. 2018.01898/full#supplementary-material

FIGURE S1 | Examination of assumption of normally distributed errors.




**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Bardeen and Fergus. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Measuring the Cognitive Attentional Syndrome in Cardiac Patients With Anxiety and Depression Symptoms: Psychometric Properties of the CAS-1R

Cintia L. Faija<sup>1</sup> , David Reeves <sup>2</sup> , Calvin Heal <sup>3</sup> , Lora Capobianco4,5, Rebecca Anderson<sup>4</sup> and Adrian Wells 4,5 \*

#### Edited by:

Gianluca Castelnuovo, Catholic University of the Sacred Heart, Italy

#### Reviewed by:

Ana Nikcevic, Kingston University, United Kingdom Daniele Di Lernia, Catholic University of the Sacred Heart, Italy Juan V. Luciano, Parc Sanitari Sant Joan de Déu, Spain

\*Correspondence:

Adrian Wells adrian.wells@manchester.ac.uk

#### Specialty section:

This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology

Received: 10 May 2019 Accepted: 30 August 2019 Published: 18 September 2019

#### Citation:

Faija CL, Reeves D, Heal C, Capobianco L, Anderson R and Wells A (2019) Measuring the Cognitive Attentional Syndrome in Cardiac Patients With Anxiety and Depression Symptoms: Psychometric Properties of the CAS-1R. Front. Psychol. 10:2109. doi: 10.3389/fpsyg.2019.02109 <sup>1</sup> Division of Nursing, Midwifery & Social Work, Faculty of Biology, Medicine and Health, Manchester Academic Health Science Centre, The University of Manchester, Manchester, United Kingdom, <sup>2</sup> Manchester Academic Health Science Centre, NIHR School for Primary Care Research, The University of Manchester, Manchester, United Kingdom, <sup>3</sup> Faculty of Biology, Medicine and Health, Centre for Biostatistics, Manchester Academic Health Science Centre, The University of Manchester, Manchester, United Kingdom, <sup>4</sup> Faculty of Biology, Medicine and Health, School of Psychological Sciences, Manchester Academic Health Science Centre, The University of Manchester, Manchester, United Kingdom, <sup>5</sup> Greater Manchester Mental Health NHS Foundation Trust, Manchester Royal Infirmary, Manchester, United Kingdom

Metacognitive Therapy (MCT) is a recent treatment with established efficacy in mental health settings. MCT is grounded in the Self-Regulatory Executive Function (S-REF) model of emotional disorders and treats a negative perseverative style of thinking called the cognitive attentional syndrome (CAS), thought to maintain psychological disorders, such as anxiety and depression. The evaluation of effective psychological therapies for anxiety and depression in chronic physical illness is a priority and research in this area depends on the suitability and validity of measures assessing key psychological constructs. The present study examined the psychometric performance of a tenitem scale measuring the CAS, the CAS-1R, in a sample of cardiac rehabilitation patients experiencing mild to severe symptoms of anxiety and/or depression (N = 440). Participants completed the CAS scale, the Hospital Anxiety and Depression Scale and the Metacognitions Questionnaire 30 (MCQ-30). The latent structure of the CAS-1R was assessed using confirmatory factor analyses (CFA). In addition, the validity of the measure in explaining anxiety and depression was assessed using hierarchical regression. CFA supported a three-factor solution (i.e., coping strategies, negative metacognitive beliefs and positive metacognitive beliefs). CFA demonstrated a good fit, with a CFI = 0.988 and an RMSEA = 0.041 (90% CI = 0.017–0.063). Internal consistency was acceptable for the first two factors but low for the third, though all three demonstrated construct validity and the measure accounted for additional variance in anxiety and depression beyond age and gender. Results support the multi-factorial assessment of the CAS using this instrument, and demonstrate suitability for use in cardiac patients who are psychologically distressed.

Keywords: cardiac, cognitive attentional syndrome, anxiety, depression, metacognitive therapy

# INTRODUCTION

Coronary heart disease is the leading cause of death for adult men and women worldwide in developed countries (World Health Organization, 2017). Prevalence of anxiety and depression among patients with cardiovascular disease is up to 3-fold higher than in the general population (Thombs et al., 2008; Tully et al., 2014). Anxiety and depression have been associated with adverse outcomes, such as increased risk of mortality and increased risk of future cardiac problems, poorer quality of life, poorer treatment adherence, and greater health care use (Thombs et al., 2008; Frasure-Smith and Lesperance, 2010; Palacios et al., 2018). Furthermore, anxiety and depression were found to be risk factors for cardiac comorbidity (Halaris, 2009). Following a cardiac event or procedure, patients are offered cardiac rehabilitation (CR) to improve health outcomes and prevent future cardiac problems (Lesperance and Frasure-Smith, 2000). The European Association of Preventive Cardiology has emphasized that symptoms of anxiety and depression in heart disease patients play a key role in the success of CR programmes (Piepoli et al., 2014). CR programmes usually include elements aiming to influence psychological and/or psychosocial outcomes. However, an audit of CR in the United Kingdom (2018) showed that when patients enter the CR programme, 27.5% experienced borderline or clinical levels of anxiety and after the programme 21% remained in those categories. In relation to depression, 18% experienced borderline or clinical levels of depression before starting CR and 12% continued to report depression afterwards (British Heart Foundation, 2018). The variation of improvement across CR programmes ranged from −13 to 43.6% for anxiety and from −12.5 to 36.4% for depression, suggesting that some patients got worse and a substantial number of them did not achieve the national average change in levels of anxiety and/or depression after CR (British Heart Foundation, 2018).

A Cochrane review including 24 randomized controlled trials evaluating effectiveness of psychological interventions vs. usual care, administered by trained staff among coronary heart disease patients, reported small to moderate improvements in depression (d = 0.21) and anxiety (d = 0.25) (Whalley et al., 2011). Furthermore, other studies highlighted that attempts to treat psychological distress in cardiac patients have shown non-significant improvements in anxiety and depression (Dickens et al., 2013; Reid et al., 2013; Jiang et al., 2017). The National Institute for Health and Care Excellence (NICE) currently recommends cognitive behavioral therapy (CBT) as the first-line treatment for anxiety (NICE, 2014) and depression (NICE, 2016). The CBT model suggests that anxiety and depression are maintained by cognitive distortions and unhelpful behaviors; CBT adopts a range of strategies to challenge the content of negative automatic thoughts to overcome negative emotions (Beck, 1967, 1976). A recent meta-analysis including 12 randomized controlled trials of CBT in cardiac patients showed small to moderate effects in improving anxiety (d = 0.34) and depression (d = 0.35) compared mainly to usual care (Reavell et al., 2018). Evidence suggests that there is considerable scope for improving outcomes of psychological interventions aimed at reducing anxiety and/or depression in the cardiac population. This has led to a recent National Institute for Health Research (NIHR) funded research programme, called PATHWAY, to examine the effects of a newer form of treatment, metacognitive therapy (MCT: Wells, 2009). A recent meta-analysis evaluating MCT has shown that this therapeutic approach is highly effective in adult mental health settings (Normann and Morina, 2018). The treatment is based on the Self-Regulatory Executive Function (S-REF) model (Wells and Matthews, 1994, 1996). The S-REF model proposes that a particular style of responding to negative thoughts called the cognitive attentional syndrome (CAS) contributes to and maintains emotional disorders and symptoms of distress (e.g., anxiety, depression) (Wells and Matthews, 1994, 1996). The CAS consists of "a perseverative thinking style that takes the form of worry and rumination, attentional focusing on threat, and unhelpful coping behaviors (e.g., thought suppression, avoidance, substance use)" (Wells, 2009, p. 10). It is problematic because it maintains negative processing and a sense of current and future threat. The CAS is thought to be caused, in part, by metacognitive beliefs that individuals hold about their thinking, such as the belief that worrying is useful for coping with threats and the belief that worrying is uncontrollable and dangerous. These positive and negative metacognitions give rise to extended negative thinking patterns that maintain awareness of threat and consequent emotional distress. In sum, the CAS locks the individual into prolonged negative emotional experiences and interferes with adaptive self-regulation, leading to feelings of hopelessness, loss of subjective control over cognition and emotion, and lack of flexibility in implementing alternative thinking styles (Wells, 2009). In contrast to CBT that challenges the content of negative thoughts, MCT aims to interrupt the CAS and challenges metacognitive beliefs using techniques, such as the metacognitive Socratic dialogue, detached mindfulness and attention training techniques (Wells, 2009).

Although trait measures exist to assess metacognitions (Wells and Cartwright-Hatton, 2004) and some dimensions of thinking style (Wells, 1994, 2009; Nolen-Hoeksema, 2000; Ehring et al., 2011), for purposes of assessment of change in treatment it is useful to measure these factors as state variables and to have one instrument that can assess all elements/factors of the CAS simultaneously. Then, changes in these key mechanisms can be monitored over time in an easy and accessible way. With this objective in mind, Wells developed the Cognitive Attentional Syndrome Scale-1 (CAS-1) (Wells, 2009) which includes items to assess maladaptive coping strategies (e.g., dwelling, worrying, focusing attention on threat, avoidance, use of alcohol/drugs) to cope with negative thoughts, and underlying negative and positive metacognitive beliefs.

The CAS-1 (Wells, 2009) has been used by clinicians delivering MCT to measure weekly changes in the CAS, and it has been recently used in research among a non-clinical sample (Fergus et al., 2012; Fergus and Scullin, 2017) and a clinical sample with primary mood or anxiety disorder (Fergus et al., 2013). The CAS-1 has been recently used in medical samples (e.g., cancer, multiple sclerosis, cardiac) (Cook et al., 2015; Heffer-Rahn and Fisher, 2015; Fisher et al., 2017; Wells et al., 2018a,b) but the factorial structure of it has not been explored.

The NIHR UK recently funded a programme of research to examine MCT for anxiety and depression in CR patients (trial protocols: Wells et al., 2018a,b). As MCT aims to target the CAS, it is necessary to assess and monitor changes in this construct. The ethical committee strongly advised reducing respondent burden in the context of the PATHWAY research study. Therefore, the CAS-1 (Wells, 2009) was revised, resulting in a shortened version of 10-items. The revised version of the instrument, Cognitive Attentional Syndrome Scale-1 Revised (CAS-1R) differs from the CAS-1 (Wells, 2009) in using a reduced number of items to assess the CAS and a different rating scale (0–100) rather than (0–8) for all the items. Six items were identified for removal from the original scale by its developer (Adrian Wells), based on clinical and research expertise using the scale. The goal was to produce a revised version incorporating the minimum number of items required to reliably assess all important elements of the CAS (e.g., worry/rumination and other coping strategies, and metacognitive beliefs).

# Aims

The aim of the present study was to investigate the psychometric properties of the CAS-1R in cardiac patients with co-morbid anxiety and/or depression. Specifically, we investigated four theoretically based models of the latent structure of the measure. Each of the structural models was derived from the S-REF model (Wells and Matthews, 1994, 1996). As shown in **Figure 1** each model introduces incremental refinement to the factor structure. A unidimensional model (**Figure 1A**) was set as the baseline model in which all items load on a general factor called CAS making no distinctions between subcomponents. The two-factor model (**Figure 1B**) differentiates between proximal and distal causative mechanisms of emotional disorders. Specifically, proximal mechanisms that maintain negative emotional experiences are included in the factor named Coping Strategies, which combines conceptual and attentional processes in the form of worry, rumination, focusing attention on threat and also strategies, such as thought suppression, avoidance (emotional and cognitive) and alcohol use. Distal mechanisms underlying anxiety and depression disorders are the metacognitive beliefs that people hold about their thinking, thus, the second factor in this model was labeled metacognitive beliefs. The three-factor model (**Figure 1C**) includes a separation between negative and positive metacognitive beliefs. From a theoretical and clinical perspective it is relevant to examine these two content domains of metacognitive beliefs separately. Specifically, negative metacognitive beliefs lead individuals into a sense of threat from thoughts themselves, unhelpful types of mental control or diminished control attempts, whilst positive metacognitive beliefs contribute to worrying and rumination as strategies to cope with distressing negative thoughts (Wells, 2009). Empirical evidence has shown that negative metacognitive beliefs are a strong predictor of anxiety and depression in mental health, physical health, student and community samples (Sun et al., 2017) and positive metacognitive beliefs are associated with rumination in depression (Papageorgiou and Wells, 2003). Thus, maintaining a differentiation between positive and negative metacognitive beliefs might help to identify whether different metacognitive beliefs predict anxiety, and/or depression symptoms. Finally, a bi-factor model consisting of the same three factors depicted in **Figure 1C** with the addition of a general factor contributing to all the individual items was hypothesized, in order to explore if a general factor would carry additional information beyond that conveyed by the three factors alone.

The primary aim of the study was to identify which factor structure fitted the underlying data best in order to evaluate theoretically derived construct validity of the instrument among a CR sample experiencing symptoms of anxiety and/or depression. The secondary aims of the study were: (i) to assess convergent and discriminant validity of the CAS-1R; (ii) to examine whether the CAS-1R explains variance in anxiety and/or depressive symptoms in cardiac patients after controlling for age and gender. Gender was controlled following evidence highlighting that depression and anxiety disorders are more prevalent in women than in men (Nolen-Hoeksema, 2001; Simonds and Whiffen, 2003; McLean and Anderson, 2009; Jalnapurkar et al., 2018). Age was controlled because anxiety and depression varies across the lifespan (Jorm, 2000; Lenze and Wetherell, 2011). Moreover, research studies examining the S-REF model and effectiveness of MCT using other measures of metacognition, such as the MCQ-30 (Wells and Cartwright-Hatton, 2004) have controlled for age and gender (e.g., Yilmaz et al., 2011; Hjemdal et al., 2013; Ryum et al., 2018). It is therefore important to explore if the results are consistent with previous findings when using a measure that assesses different elements of the CAS and not only metacognitive beliefs.

# MATERIALS AND METHODS

# Ethics Statement

This study draws on data collected under a five years programme of research funded by the National Institute for Health Research (NIHR) and sponsored by Greater Manchester Mental Health NHS Foundation Trust. The research programme is called PATHWAY and the Chief Investigator is Professor Adrian Wells. The aim of the programme is to improve effectiveness of psychological interventions for anxiety and depression in CR services. The psychological intervention delivered is MCT (Wells, 2009). Ethical approval for the PATHWAY programme has been granted by the NHS Research Ethics Committee, UK. The Group-MCT Trial (Wells et al., 2018a) received ethical approval from Preston Research Ethics Committee (Ref: 15/NW/0163) and the Home-based MCT Feasibility Trial (Wells et al., 2018b) received ethical approval from the North West-Greater Manchester West Research Ethics Committee (Ref: 16/NW/0786).

# Participants and Procedure

Participants were recruited from CR services at seven National Health Services (NHS) Trusts in the North-West of England. Participants were invited to take part in the PATHWAY Programme if they met the eligibility criteria presented on **Table 1**. In the present study, anxiety and/or depression symptoms were defined by a score of 8 or more on either of the subscales of the Hospital Anxiety and Depression Scale (HADS) which corresponds to at least a mild category (HADS; Zigmond

TABLE 1 | Participant's eligibility: inclusion and exclusion criteria.

#### INCLUSION CRITERIA

(i) Patients were referred to the cardiac rehabilitation services

(ii) A score of ≥8 on the depression and/or anxiety subscale of the Hospital Anxiety and Depression Scale (Zigmond and Snaith, 1983)

(iii) Minimum of 18 years old

(iv) Competent level of English language skills

#### EXCLUSION CRITERIA

(i) Cognitive impairment precluding informed consent or ability to participate (ii) Acute suicidality

(iii) Active psychotic disorder

(iv) Current drug/alcohol abuse

(v) Concurrent psychological intervention for emotional distress that is not part of usual care

(vi) Antidepressant or anxiolytic medications initiated in the previous 8 weeks (vii) Life expectancy of <12 months

and Snaith, 1983). In a general population, a score of 8 provides 82% sensitivity and 74% specificity for detecting major depressive disorder, and 78% sensitivity and 74% specificity for identifying generalized anxiety disorder (Brennan et al., 2010). Fifty-three percent of the patients invited to take part in the PATHWAY programme agreed to participate.

Patients meeting the inclusion criteria were identified by NHS CR staff that also provided an invitation flier and the patient information sheet to interested patients. All eligible and interested patients were asked to provide written informed consent prior to participating in the study and were then asked to complete the study questionnaires at baseline, 4 and 12 months follow up. Data for the present study include baseline measures only (before receiving any treatment).

The sample consisted of 440 participants experiencing mild to severe symptoms of anxiety and/or depression referred to CR services. The sample mean age was 60.24 (SD = 10.76, age range from 27 to 87), the majority of the sample were male (65.5%), white (90.8%), with almost half of the participants married (48.9%), and 78% reported having achieved an educational qualification (e.g., GCSE, diploma, degree).

## Measures

#### The Cognitive Attentional Syndrome Scale-1 Revised (CAS-1R) (Wells, 2015)

The original CAS-1 questionnaire is a 16-item self-report questionnaire developed to assess the different elements of the cognitive attentional syndrome (Wells, 2009, p. 268). The CAS-1 has demonstrated adequate internal consistency (Cronbach's Alpha between 0.78 and 0.86) (Fergus et al., 2012, 2013) and has shown good convergent validity with a measure assessing psychological inflexibility (r = 0.63) (Fergus et al., 2013). The CAS-1 was revised by shortening it to 10-items and changing the response scale to increase consistency of responses across items. Examples of the CAS-1R items are: "How much time in the last week have you found yourself dwelling on or worrying about problems [e.g., health, family, finances]?," "How much do you believe that worrying or dwelling on thoughts is uncontrollable?" Items are rated based on the past 7 days on an 11-point response scale ranging from 0 (none of the time/not at all true) to 100 (all of the time/completely certain this is true) in steps of 10. This is the first study assessing the factorial structure of the CAS-1R measure.

# The Hospital Anxiety and Depression Scale (HADS;

Zigmond and Snaith, 1983)

The HADS is a 14-item self-report scale assessing anxiety (7 items) and depression (7 items). Respondents rate the items based on the past 7 days using a four-point scale (from 0 to 3). High scores indicate greater anxiety, depression, and general emotional distress. The HADS is a widely used measure and has shown good internal consistency for both subscales (Cronbach's alpha for anxiety = 0.85 and 0.80 for depression) and for the total scale (Cronbach's alpha = 0.89) (Roberts et al., 2001). The HADS is used in CR services as part of routine assessment in the UK (Stafford et al., 2007; Tesio et al., 2014, 2017; British Heart Foundation, 2018). The Cronbach alpha values for the present sample were as follows: 0.81 for anxiety, 0.76 for depression, and 0.84 for the total score.

# The Metacognitions Questionnaire 30 (MCQ-30; Wells and Cartwright-Hatton, 2004)

The MCQ-30 is a 30-item self-report scale that measures different dimensions of metacognitive beliefs. The questionnaire assess five domains: (i) Cognitive Confidence (e.g., My memory can mislead me at times), (ii) Positive Beliefs about Worry (e.g., Worrying helps me cope), (iii) Negative Beliefs about Uncontrollability and Danger (e.g., When I start worrying I cannot stop, My worrying is dangerous for me), (iv) Cognitive Self-Consciousness (e.g., I pay close attention to the way my mind works), and (v) Need to Control Thoughts (e.g., Not being able to control my thoughts is a sign of weakness). Each domain is a subscale with six items. Respondents rate how much they "generally agree or disagree" with the statements presented on a four-point scale (from 1 to 4). The MCQ-30 has good internal consistency and good test–retest reliability (Wells and Cartwright-Hatton, 2004; Spada et al., 2008; Fergus and Bardeen, 2017). In addition, a five-factor solution of the MCQ-30 was confirmed in medical samples (i.e., cancer and epilepsy) (Cook et al., 2015; Fisher et al., 2016), non-clinical samples (Wells and Cartwright-Hatton, 2004; Spada et al., 2008; Fergus and Bardeen, 2017) and psychiatric disorder samples (Martin et al., 2014; Grötte et al., 2016). Furthermore, a bi-factor solution of the MCQ-30 (i.e., a general factor named metacognitions and five factors representing each subscale) demonstrated good fit in a non-clinical sample (Fergus and Bardeen, 2017).

The Cronbach alpha values for the present sample were as follows: 0.91 for Cognitive Confidence, 0.88 for Positive Beliefs about Worry, 0.83 for Negative Beliefs about Uncontrollability and Danger, 0.81 for Cognitive Self-Consciousness, 0.73 for Need to Control Thoughts, and 0.91 for the Total Score.

# Statistical Analyses

## Descriptive Statistics

Descriptive statistics included means, standard deviations, and score distributions for the individual CAS-1R items. In addition, mean and standard deviations are reported for the CAS-1R, MCQ-30, and HADS.

# Measurement Models

The factor structure of the CAS-1R was investigated using confirmatory factor analysis (CFA). If none of the hypothesized factor structures demonstrated an adequate fit to the data, exploratory factor analysis was planned to determine whether a different, non-hypothesized model could be identified.

Four different models for the factor structure of the CAS-1R were hypothesized based on the S-REF model and were compared. The unidimensional model (**Figure 1A**) was fitted first, principally to provide a baseline for comparison of the more complex models as the expectation was that this model would not fit the data well. In sequence we then fitted the two-factor model, discriminating between coping strategies (6 items) and metacognitions (4 items) (**Figure 1B**); the threefactor model with a further differentiation between positive (2 items) and negative metacognitive beliefs (2 items) (**Figure 1C**); and finally the bi-factor model including a general factor on which all items loaded independently of the different specific domains. Factors consisting of only two items are generally not recommended as this can cause problems of model identification and the items may not adequately tap the latent construct (Hair et al., 2010). However, the distinction between positive and negative metacognitive beliefs is theoretically and clinically relevant (Wells and Matthews, 1994, 1996; Wells, 2009).

The hypothesized models were each specified with no correlated errors between the observed variables (Byrne, 2001), with the intention that if no model demonstrated an adequate fit, correlated error terms between observed variables within the same factor would be added based on modification indices (Aish and Joreskog, 1990). The analysis sample of 440 individuals was well in excess of a generally accepted rule of thumb of a minimum sample of 200 for CFA (Kline, 2011; Koran, 2016).

# Model Estimation and Evaluation

AMOS Version 22 (Arbuckle, 2014) was used to conduct CFA within a structural equation modeling framework using maximum likelihood (ML) estimation. Although the CAS-1R items demonstrated non-normal distributions (see below) for which weighted least squares (WLS) is often advocated, simulation studies have demonstrated that ML in fact strongly outperforms WLS (and Generalized Least Squares) under such conditions, including when data is ordinal, and that WLS tends to over-estimate goodness-of-fit (Olsson et al., 2000). The adequacy and parsimony of the models was assessed using a set of commonly-recommended fit statistic indices (Hu and Bentler, 1999; Kline, 2011; Brown, 2015): the Comparative Fit Index (CFI), the Root Mean Square Error of Approximation (RMSEA), the Goodness of Fit Index (GFI), and the Parsimony Goodness of Fit Index (PGFI). We assessed goodness of fit principally on the basis of the CFI and RMSEA, as these indices are least sensitive to sample size and parameter estimates (Hu and Bentler, 1998), using the modern criteria of CFI greater or equal to 0.95 (Hu and Bentler, 1999) and RMSEA <0.08 indicate an acceptable fit and 0.05 a good fit, with an upper 90% confidence limit of 0.1 or less (Browne and Cudeck, 1993). The additional indices were computed to provide a broader picture of model performance and were a GFI value close to 1 and a PGFI above 0.5 which indicate good fit (Mulaik et al., 1989; Hu and Bentler, 1999). We also report the Chi-square statistic, but goodness-of-fit decisions were not based on this as it is known to be sensitive to sample size and to large correlations between factors within the model, making it an unreliable criterion for detecting well-fitting models (Tanaka, 1987). However, the Chi-square difference test was used to statistically compare models according to overall fit, for which it appropriately preserves the alpha-level regardless of sample size (Marsh et al., 2004).

# Assessing Reliability and Validity

The internal consistency of each factor in the resulting model for the CAS-1R was assessed using Cronbach's Alpha and McDonald's Omega coefficient. Alpha is reported, being the commonly accepted standard measure of scale reliability. However, when factor loadings are not equal, alpha underestimates true reliability (Trizano-Hermosilla and Alvarado, 2016) and we therefore also report omega—which is computed directly from the factor loadings—as a generally less biased measure (Trizano-Hermosilla and Alvarado, 2016). Factor uniqueness was assessed using inter-correlations. Convergent and discriminant validity were assessed on the basis of directions and strengths of correlations between the CAS-1R and the MCQ-30, and the HADS. Specifically, a number of relationships were evaluated in line with theoretical expectations: (i) subscales assessing negative metacognitions in the CAS-1R and in the MCQ-30 would correlate positively, and at a higher level than negative CAS-1R metacognitions with MCQ positive beliefs; (ii) subscales assessing positive metacognitions in the CAS-1R and in the MCQ-30 would correlate positively, and at a higher level than positive CAS-1R metacognitions with MCQ negative metacognitions; and (iii) all the CAS-1R subscales would show a positive correlation with the HADS subscales.

# T-Tests and Regression Analysis

Independent sample t-tests were conducted to explore gender differences in the CAS-1R and the HADS. Although there are no published studies exploring the role of the CAS in CR patients, we hypothesized on theoretical grounds that the CAS-1R would explain anxiety and/or depression above and beyond the variation accounted for by age and gender. To this end, hierarchical regression analysis was conducted. At Step 1, age and gender were entered and at Step 2 all the CAS-1R subscales were included using forced entry.

Assumptions of linearity, homoscedasticity, independence of residuals and the normality of distributed errors were examined to determine whether regression analyses were appropriate (Field, 2013). Regression plots were reviewed to confirm linearity, correlation coefficients between variables were reviewed for multicollinearity, and values of the tolerance and variance inflation factors (VIF) were examined; tolerance values lower than 0.10 or 0.25 are considered a cause of concern (Tabachnick and Fidell, 2001); and VIF values should not exceed 10 (Field, 2013).

# RESULTS

# Descriptive Statistics

There were no missing values for the CAS-1R. The response distributions on each CAS-1R item are given in **Table 2**. Mean values for items ranged from 23.73 (item 9) to 53.43 (item 1), except for item 6 (M = 3.60). Item 6 assesses the use of alcohol to cope with thoughts and feelings, and was the only item substantially skewed, with 88.2% of participants reporting a score of 0 on this item. The lack of score variation on item 6 negatively impacts on estimates of correlation with other items and may be specific to this sub-population. We therefore decided to exclude item 6 from the subsequent structural modeling.

Responses were missing for 20 items on the MCQ-30, and two on the HADS, with no more than two missing responses for any single participant. As the amount of missing data were very small (<0.1% in total) missing values on each scale were replaced with participant means across the completed items.

# CAS-1R Measurement Models

Standardized factor loadings (regression weights) for each of the hypothesized models are presented in **Figure 2**. Standardized factor loadings on the different models ranged from 0.34 to 0.89.

Goodness-of-fit statistics for each of the measurement models are presented in **Table 3**. As anticipated, the unidimensional model did not reach our primary criteria (CFI and RMSEA) for adequate fit. The two-factor model showed a significant improvement over the unidimensional solution according to the Chi-squared difference test, and was borderline with regard to our primary fit indices [CFI = 0.953; RMSEA = 0.082 (95% CI 0.066–0.099)]. The three-factor model represented a further significant improvement, with substantially improved fit in terms of CFI and RMSEA and its confidence interval [CFI = 0.988; RMSEA = 0.043 (95%CI 0.022–0.064)]. When attempting to fit a bi-factor model, we experienced problems of identification and negative variance estimates, only solvable by adding additional parameter constraints into the model. Even then, parameter estimates were unstable under different constraint assumptions. We took this as evidence that a bi-factor solution did not fit the data and do not report any further on this model.

On the basis of these results we selected the three-factor model as the optimal solution, displaying as it did a good fit on all the criteria and a statistically significant improvement over the two-factor model. The three-factor model discriminates between coping strategies (5 items), negative (2 items), and positive (2 items) metacognitive beliefs; in addition, the correlation between the negative and positive belief factors was 0.46, suggesting that these are reasonably distinct constructs. Therefore, the threefactor model was used for all subsequent analyses.

Patient scores were computed on each sub-scale (factor) as a total score across the included items (rather than applying item weights from the CFA) to reflect how the instrument is used in practice. These sub-scale scores were then used for assessing validity.

# CAS-1R Reliability, Convergent, and Discriminant Validity

Cronbach alpha values were 0.88 for Coping Strategies, 0.65 for Negative Metacognitive Beliefs, and 0.58 for Positive Metacognitive Beliefs. Corresponding omega values were 0.88 for Coping Strategies, 0.70 for Negative Metacognitive Beliefs, and 0.59 for Positive Metacognitive Beliefs. Correlations between the three CAS-1R subscales were all moderate, with the highest being 0.55 between Coping Strategies and Negative Metacognitive Beliefs (**Table 4**).

Results relating to assessment of convergent and discriminant validity are summarized in **Table 5**. As hypothesized, each CAS-1R subscale was found to correlate more highly with similar constructs than with dissimilar constructs. CAS-1R Negative Metacognitive Beliefs correlated highly with MCQ-30 Negative Beliefs and showed a significantly lower correlation TABLE 2 | Descriptive statistics for the CAS-1R Items: Mean, Standard Deviation, frequency and percentage per scale category (N=440).



<sup>a</sup>Reduction in χ 2 from previous model.

with MCQ-30 Positive Beliefs (r = 0.62 vs. r = 0.17; p < 0.001); similarly, CAS-1R Positive Metacognitive Beliefs correlated moderately with MCQ-30 Positive Beliefs and had a significantly lower correlation with MCQ-30 Negative Beliefs (r = 0.53 vs. r = 0.25; p < 0.001). All the CAS-1R subscales were positively correlated with HADS anxiety and HADS depression, though associations with the latter were all lower.

# T-Tests and Regression Analyses

Independent sample t-tests exploring gender differences in the CAS-1R subscale scores did not show significant differences. However, gender differences were found to be significant only for HADS-Anxiety scores: males (M = 9.81, SD = 3.85) and females (M = 11.29, SD = 3.67); t(320) = −3.95, p = < 0.001.

Assumptions of linearity, homoscedasticity, independence of residuals, and normally distributed errors were met for regression analyses. The Durbin-Watson test values for all the regression models were all close to 2, indicating that the assumption of independent errors was met (Field, 2013). Tolerance statistics for all regression models were all above 0.62 and the VIF values were all below 2, suggesting collinearity was not a problem (Tabachnick and Fidell, 2001; Field, 2013).

The regression models examined whether as a block the three subscales of the CAS-1R explained variance in anxiety and depression after controlling for age and gender. As shown in **Table 6**, when predicting HADS-Anxiety and HADS-Depression, the inclusion of the CAS-1R subscales (step 2) was significant and accounted for additional variance: 37% in anxiety and 21% in depression, respectively. At a subscale level, all the three CAS-1R subscales were unique predictors of anxiety; whilst Coping Strategies alone was a significant individual predictor of depression.

# DISCUSSION

The assessment and monitoring of change in purported underlying causal mechanisms of anxiety and depression in patients with medical conditions is a priority for evaluating and interpreting psychological treatment outcomes. This is the first study investigating the factor structure of a measure assessing the CAS in a sample of cardiac patients with mild to severe symptoms of anxiety and/or depression. The measure is grounded in the S-REF model (Wells and Matthews, 1994, 1996) which proposes that the CAS is a key construct in explaining the maintenance of psychological disorders.

Results of the CFA showed that the best fit for the CAS-1R data in cardiac patients experiencing emotional distress corresponded to a three-factor model distinguishing between unhelpful coping strategies (e.g., worry, rumination, avoidance), negative and positive metacognitive beliefs, supporting the value in separating these constructs. This separation of factors maps neatly onto the focus of metacognitive therapy that aims to increase patient awareness of CAS processes, bring them under control and challenge negative and positive metacognitive beliefs (Wells, 2009).


TABLE 4 | Descriptive statistics and correlations for psychological measures (i.e., CAS-1R, MCQ-30, and HADS).

TABLE 5 | Summary of investigations of CAS-1R convergent and discriminant validity.


\*CAS NEG, negative metacognitive beliefs; CAS-POS, positive metacognitive beliefs.

\$Z-score relating to comparison of (a) with (b) controlling for (c).

This study found positive associations between the CAS-1R and anxiety and depression symptoms, which is consistent with previous findings using the CAS-1 in clinical (Fergus et al., 2013) and non-clinical samples (Fergus et al., 2012; Fergus and Scullin, 2017). These positive relationships were also found among samples with physical conditions, i.e., patients with cancer (McNicol et al., 2013; Cook et al., 2015; Fisher et al., 2017) and multiple sclerosis (Heffer-Rahn and Fisher, 2015).

The results of the regression analyses provide evidence that the three components of CAS-1R are significant statistical predictors of anxiety among cardiac patients after controlling for age and gender. The CAS-1R was also a predictor of depression symptoms, although the only significant contributing factor was coping strategies. This could be related to the sample being more anxious than depressed.

The alcohol use item of the CAS-1R was very highly skewed and was removed from the analysis. Participants' answers to this item may reflect CR patients being asked to stop unhealthy behaviors, such as smoking and drinking, and some responses may have been aspirational rather than actual. It is anticipated that this item may perform differently in other populations and it may retrieve valuable information in other samples.

The CFA yielded a good fit for one of the CAS-1R hypothesized model, i.e., the three-factor model. Internal consistency was excellent for the coping strategies factor, acceptable for negative metacognitions, but well below the TABLE 6 | Cognitive Attentional Syndrome Scale-1 Revised (CAS-1R) subscales predicting anxiety and depression symptoms, after controlling for age and gender.

#### (A) CAS-1R SUBSCALES PREDICTING SYMPTOMS OF ANXIETY


#### (B) CAS-1R SUBSCALES PREDICTING SYMPTOMS OF DEPRESSION


Bold values represent a significant p-value.

conventional threshold of 0.70 for positive metacognitions. The latter two factors each included just two items, which may be contributing to lower internal consistency. However, derived subscales scores showed good convergent and discriminant validity with the MCQ-30 subscales (i.e., positive beliefs about worry and negative beliefs about uncontrollability and danger), suggesting that these subscales have practical utility in spite of this.

The present study provides support for the use of the multiple dimensions of the CAS-1R in research settings and its continued use in clinical settings. Findings suggest that psychological treatments for anxiety and depression in cardiac patients should target both unhelpful thinking styles and coping strategies and metacognitive beliefs. Generalization of the psychometric properties of the CAS-1R to populations with different mental health diagnoses and other psychical illnesses warrants further research.

# Strengths and Limitations

Strengths of this study include a reasonably large sample of more than 400 participants used to test the theoretical models, no missing data on the CAS-1R, and only a very small amount of missing data for the MCQ-30 and the HADS (<0.01%). However, some limitations warrant discussion. Data were not collected to examine test-retest reliability of the CAS-1R, meaning that this area remains unexplored and should be considered in future research. The measure is intended to be a state measure that is sensitive to variation in the CAS, but a limitation at the present time is a lack of data on responsivity of the measure. It is important to highlight that two of the factors are measured by just two items which may provide limited coverage of these constructs. If more comprehensive assessment of negative and positive metacognitive beliefs is required, the MCQ-30 could be used alongside the CAS-1R.

## Conclusion

This study investigated the factor structure and some of the psychometric properties of a measure of the CAS. Findings provide preliminary evidence supporting a theoretically consistent and well-fitting three-factor solution. Given these findings it is recommended that the measure be used to evaluate change in putative maintenance factors during the course of psychological therapy for anxiety and depression in cardiac samples. The use of the CAS-1R measure in future research could help to enhance understanding of psychological processes involved in treatment response and maintenance of emotional distress in cardiac patients and other populations.

# DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

# ETHICS STATEMENT

Ethical approval for the PATHWAY programme has been granted by the NHS Research Ethics Committee, UK (References: 15/NW/0136, 16/NW/0786). All patients provided written informed consent.

# AUTHOR CONTRIBUTIONS

AW designed the CAS-1R and CF, DR, and AW designed the study. CF, LC, and RA recruited, consented and administered the baseline measures to participants. CF and CH conducted the statistical analysis supervised by DR and AW. CF drafted the initial manuscript. DR and AW revised the manuscript. All authors contributed and agreed the final draft.

# REFERENCES


# FUNDING

This study was funded by the National Institute for Health Research (NIHR) under its Programme Grants for Applied Research scheme (RP-PG-1211 20011) awarded to AW. The views and opinions expressed are those of the authors and do not necessarily reflect those of the NIHR, NHS or the Department of Health and Social Care. Greater Manchester Mental Health NHS Foundation Trust is the sponsor.

# ACKNOWLEDGMENTS

We gratefully acknowledge all our colleagues of the PATHWAY Team based at the different seven NHS sites, including nurses, researchers, and administrators. We would like to thank the Manchester Academic Health Science Centre Clinical Trials Unit for managing and releasing data for this study. Thank you very much to the people who took time to complete the questionnaires for the study.


of patients with cardiovascular disease. Core components, standards and outcome measures for referral and delivery. Eur. J. Prev. Cardiol. 21, 664–681. doi: 10.1177/2047487312449597


**Conflict of Interest Statement:** AW is the developer of metacognitive therapy and a co-director of the Metacognitive Therapy Institute.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Faija, Reeves, Heal, Capobianco, Anderson and Wells. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Neural Correlates of Cognitive-Attentional Syndrome: An fMRI Study on Repetitive Negative Thinking Induction and Resting State Functional Connectivity

#### Joachim Kowalski<sup>1</sup> \*, Marek Wypych<sup>2</sup> , Artur Marchewka<sup>2</sup> and Małgorzata Dragan<sup>1</sup> <sup>1</sup> Faculty of Psychology, University of Warsaw, Warsaw, Poland, <sup>2</sup> Laboratory of Brain Imaging, Nencki Institute of

Experimental Biology, Polish Academy of Sciences, Warsaw, Poland

negative thinking, abstract thinking, and resting states.

#### Edited by:

Gerald Matthews, University of Central Florida, United States

#### Reviewed by:

Almira M. Kustubayeva, Al-Farabi Kazakh National University, Kazakhstan Francisco J. Ruiz, Fundación Universitaria Konrad Lorenz, Colombia

\*Correspondence: Joachim Kowalski joachim.kowalski@psych.uw.edu.pl

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 07 September 2018 Accepted: 08 March 2019 Published: 26 March 2019

#### Citation:

Kowalski J, Wypych M, Marchewka A and Dragan M (2019) Neural Correlates of Cognitive-Attentional Syndrome: An fMRI Study on Repetitive Negative Thinking Induction and Resting State Functional Connectivity. Front. Psychol. 10:648. doi: 10.3389/fpsyg.2019.00648 Aim: Cognitive-attentional syndrome (CAS) is the main factor underlying depressive and anxiety disorders in the metacognitive approach to psychopathology and psychotherapy. This study explore neural correlates of this syndrome during induced

Methods: n = 25 people with high levels of CAS and n = 33 people with low levels of CAS were chosen from a population-based sample (N = 1225). These groups filled-in a series of measures of CAS, negative affect, and psychopathology; they also underwent a modified rumination induction procedure and a resting state fMRI session. Resonance imaging data were analyzed using static general linear model and functional connectivity approaches.

Results: The two groups differed with large effect sizes on all used measures of CAS, negative affect, and psychopathology. We did not find any group differences in general linear model analyses. Functional connectivity analyses showed that high levels of CAS were related to disrupted patterns of connectivity within and between various brain networks: the default mode network, the salience network, and the central executive network.

Conclusion: We showed that low- and high-CAS groups differed in functional connectivity during induced negative and abstract thinking and also in resting state fMRI. Overall, our results suggest that people with high levels of CAS tend to have disrupted neural processing related to self-referential processing, task-oriented processing, and emotional processing.

Keywords: repetitive negative thinking, cognitive-attentional syndrome, rumination, resting state, fMRI, neural correlates

# INTRODUCTION

fpsyg-10-00648 March 22, 2019 Time: 17:57 # 2

Cognitive-attentional syndrome (CAS) is a key construct in Wells' metacognitive theory of emotional disorders (Wells and Matthews, 1994; Wells, 2009). In the Self-Regulatory Executive Function (S-REF) model, CAS is a set of psychological processes that includes repetitive negative thinking (worry and rumination), threat monitoring, and associated unhelpful behavioral and cognitive strategies; it is derived from metacognitive beliefs, either positive (e.g., "If I ruminate I will understand my situation") or negative (e.g., "I cannot control my ruminative thoughts"). While moments of negative self-appraisal are relatively brief in most people, the prolonged occurrence of negative emotions and negative self-appraisal in some people is due to recurring activation of CAS. This specific style of responding to negative thoughts is considered a transdiagnostic factor which underlies emotional disorders. Many studies have confirmed the relationship of CAS with emotional distress as well as symptoms of mood and anxiety disorders (Fergus et al., 2012, 2013). According to the metacognitive model, CAS is a prominent factor in the development of mood disorders, e.g., major depressive disorder (MDD; Papageorgiou and Wells, 2001, 2003, 2009; Wells, 2009), anxiety disorders, e.g., generalized anxiety disorder (GAD; Wells, 1999, 2005, 2007, 2009), posttraumatic stress disorder (PTSD; Wells and Sembi, 2004; Wells, 2009; Bennett and Wells, 2010), and obsessive-compulsive disorder (OCD; Fisher and Wells, 2005; Myers et al., 2009a,b; Wells, 2009; Solem et al., 2010).

A fundamental element of CAS is a pattern of negative, pervasive, and recurring thoughts. Rumination is associated with decreased attentional resources (Donaldson et al., 2007; Koster et al., 2011), the occurrence of negative emotions, and difficulties with problem solving (Nolen-Hoeksema et al., 2008). A ruminative thinking style is most often associated with mood disorders, as it is a risk factor for the development of depression (Nolen-Hoeksema et al., 2008) and is generally associated with dysphoric and depressive mood (Mor and Winquist, 2002). However, rumination is not only present in mood disorders – it also plays a prominent role in the symptomatology of other emotional and psychiatric disorders, such as anxiety or eating disorders (Olatunji et al., 2013). Pathological worry, another form of extended thinking, is considered a key feature of GAD; however, many researchers have shown that it also occurs in other types of emotional disorders (e.g., Starcevic et al., 2007; Spinhoven et al., 2015).

To date, there have been no studies on brain functioning in people with high levels of CAS – i.e., elevated levels of CAS-related symptomatology: repetitive negative thinking, attention to threats, unhelpful coping behaviors, and maladaptive metacognitive beliefs. There are, however, some studies using functional magnetic resonance imaging (fMRI) methods in which induction of core aspects of CAS – rumination (state rumination rather than trait rumination; Cooney et al., 2010; Berman et al., 2014; Burkhouse et al., 2017) or worry (Paulesu et al., 2010) – has been employed. The first two of the aforementioned studies on rumination induction compared depressed participants to healthy controls, while the third compared adolescents with remitted MDD to healthy controls. The Rumination Induction task used in an fMRI setting by Cooney et al. (2010) consisted of alternating blocks of ruminative, concrete, and abstract sentences which participants were asked to think about (e.g., "think about the expectations people have for you"). In this procedure, ruminative sentences, in comparison to concrete/abstract sentences, were associated with altered activity in brain regions involved in emotion processing and regulation in depressed patients: the dorsolateral prefrontal cortices, cingulate cortices, amygdalae, and parahippocampi (Cooney et al., 2010). Another study compared resting state functional connectivity with functional connectivity during negative mood induction using personalized cues created by ruminating on negative autobiographical events (e.g., "Please recall a specific time when you were very embarrassed"; Berman et al., 2014). This study showed that depressed patients had stronger connections within brain regions belonging to the default mode network (DMN), like the cingulate cortex. It was suggested that these results may be understood as difficulty in down-regulating self-oriented emotional and cognitive processing after rumination induction (Berman et al., 2014). A fourth study (Burkhouse et al., 2017) found that rumination induction with prior negative mood induction (e.g., "Remember when you failed badly at something") elicits stronger neural activations in regions involved in the DMN and emotion processing in remitted MDD adolescents. A study by Paulesu et al. (2010) explored differences in worrying between patients with GAD and healthy controls. Sentences which induce worrying (e.g., "Mull over what worries you about your future") were related to activation in the anterior cingulate and dorsal medial prefrontal cortex in the GAD group.

Several recent meta-analyses on neuronal functioning in people with depression (Hamilton et al., 2012; Palmer et al., 2015), specific phobias (Ipser et al., 2013), and PTSD (Simmons and Matthews, 2012) show, in general, that emotional disorders are most prominently connected to the dysregulation of subcortical brain areas involved in emotion processing, i.e., the amygdalae and hippocampi, as well as the striatum. This dysregulation is interpreted as the overdeveloped salience of threatening or saddening stimuli. Also, several cortical regions are involved in this type of processing, like the insulae and dorsolateral prefrontal cortices. Studies on repetitive negative thinking induction and large meta-analyses on emotional disorders have found that people experiencing mood and anxiety disorders exhibit dysregulation of the default mode, salience, and executive networks. Overall, people with emotional disorders demonstrate a pattern of disrupted neural processing in the areas of self-referential, task-oriented, and emotional processing (Hamilton et al., 2012; Simmons and Matthews, 2012; Ipser et al., 2013; Palmer et al., 2015).

In the current study, we aimed to explore differences in neural functioning between people with high and low levels of CAS symptoms. Given that there are no previous studies on the neural correlates of CAS, we decided to base our hypotheses on available work on repetitive negative thinking induction and meta-analytical results regarding emotional disorders which, according to metacognitive theory, are undergirded by CAS. We hypothesized that people with high levels of CAS symptoms

will show similar patterns of cortical activations to those found in studies on neural correlates of depressive and anxiety disorders, as described above. To test these hypotheses, we employed a modified Rumination Induction procedure and resting state functional magnetic resonance imaging (rsfMRI). We expected that differences in neural activation in people with high levels of CAS symptoms (HCAS) would be comparable to the patterns of activation reported by Cooney et al. (2010) in depressed patients, with greater neural activity in the amygdalae, hippocampi, and cingulate and dorsolateral cortices in the rumination condition as compared to the abstract condition. We also hypothesized that the cortical regions associated with rumination and which show aberrant activity in emotional disorders will show different patterns of functional connectivity in the HCAS group in comparison to the group with low levels of CAS symptoms (LCAS). We expected to find disrupted patterns of connectivity within and between several neural networks: the DMN, the salience network, and the central executive network (CEN).

# MATERIALS AND METHODS

# Procedure and Sample Selection

Participation in the study was voluntary and participants gave their informed consent. The study was approved by the Research Ethics Committee at the Faculty of Psychology, University of Warsaw. The study was conducted in two stages. The first stage took place through an Internet survey panel and was conducted by an external company. A large sample was gathered for the purpose of an fMRI study, so there were standard strict exclusion criteria related to the fMRI procedure (left-handedness, metal objects within the body, irremovable piercings, etc.) as well as any history of neurological or serious mental disorders or substance abuse disorders. Participants were also required to live in the Warsaw area to ensure their ability to participate in the second stage of the study. A total of 1,225 participants were eligible and completed the first stage of the study. Participants were selected based on quotas mirroring the population of Warsaw (Central Statistical Office, 2017) in terms of sex, age, and education. **Figure 1** depicts the selection procedure from the first to the final stage of the study.

From the first stage participants, two extreme groups were selected. As the results of previous studies (Kowalski and Dragan, 2019) have suggested that combining different measures of aspects of CAS is best for predicting levels of psychopathology, several measures were used in forming the two groups. The cutoff criterion was a score above the 66th percentile or below the 33rd percentile of the sum of results on the following measures: the CAS-1 questionnaire, the Brooding subscale of the RRS (as this aspect of rumination is most robustly associated with depressive and anxiety disorders, cf. Olatunji et al., 2013), and the Need to Control Thoughts as well as the Uncontrollability and Danger subscales from the MCQ-30, as these aspects of metacognitive beliefs are most prominently connected to levels of anxiety and depression (cf. Wells and Cartwright-Hatton, 2004; Spada et al., 2008; Dragan and Dragan, 2011; Sarisoy

et al., 2014). Finally two extreme groups, each consisting of 134 subjects, were formed.

The second stage of the study took part in the Laboratory of Brain Imaging, Neurobiology Center, Nencki Institute of Experimental Biology, Polish Academy of Sciences. Participants were invited to the laboratory in a random order by a person from an external company. Researchers were blinded to the participants' group affiliation. A total of 89 participants took part in the study – 43 in the HCAS group and 46 in the LCAS group. Participants who underwent the whole fMRI procedure were given a sum of money equivalent to about 50 EUR.

The second stage of the study occurred 4–22 weeks after the first stage, depending on the timing of the participants' second stage appointment. Despite the acceptable time-stability of the questionnaire results between the first and second stages of the study (correlations of results at these two time points: CAS-1: r = 0.83, p < 0.001, RRS – Brooding: r = 0.82, p < 0.001, MCQ – Need to Control Thoughts: r = 0.76, p < 0.001, MCQ – Uncontrollability and Danger: r = 0.82, p < 0.001) some shift in individual results was observed. To ensure that both groups had extreme characteristics, participants had to have results above or below median on all four measures used in the study. As a result, 31 participants were excluded: 30 had mixed results and 1 "changed groups" as this participant had HCAS results



<sup>∗</sup>Chi-squared test; ∗∗Cramer's Phi; CAS-1, Cognitive-Attentional Syndrome Questionnaire; RRS-brooding, Ruminative Response Scale - brooding subscale; MCQ-30, Metacognitions Questionnaire - Short Version; SCL-27 plus, Symptoms Checklist 27-plus.

on the internet measures but LCAS results on the day of the fMRI scan. Ultimately, data from 58 participants (HCAS = 25, LCAS = 33) were analyzed and are presented in this paper. Group demographic characteristics are presented in **Table 1**. These groups were also clinically diagnosed with a SCID-I interview but full results are presented elsewhere (Kowalski and Dragan, 2019; Dragan and Kowalski, unpublished). A total of 45% of participants from the HCAS group and none from LCAS group met the diagnostic criteria for a current diagnosis of a psychological disorder. In the HCAS group, 12 participants were diagnosed according to DSM-IV-TR criteria with: MDD (1), dysthymic disorder (1), GAD (2), GAD comorbid with social phobia (1), GAD comorbid with social phobia and dysthymic disorder (1), PTSD comorbid with MDD (1), PTSD comorbid with social phobia (1), PTSD comorbid with binge eating (1), cyclothymic disorder comorbid with bulimia nervosa (1), depressive disorder NOS (1), and anxiety disorder NOS (1). All participants were treatment-naive and diagnosis-naive at the beginning of the study. The second stage of the procedure consisted of filling-in questionnaires (CAS-1, RRS, MCQ-30, SCL-27) followed by the MRI procedure, including: a T1 weighted structural scan, rsfMRI, and a Rumination Induction procedure. This MRI procedure lasted approximately 40 min in total and constituted a part of a larger MRI study. After the MRI procedure, participants filled-in PANAS and STAI questionnaires. A schematic representation of the procedure is displayed in **Figure 2**.

# Measures and Materials

## The Cognitive-Attentional Syndrome Questionnaire (CAS-1)

The CAS-1 questionnaire (Wells, 2009) consists of 16 items measuring aspects of CAS: worry/rumination, attention to threat, maladaptive behaviors, and metacognitive beliefs. The results of the questionnaire were calculated as in the paper by Fergus et al. (2012) – the last eight items were recalculated to range between 0 and 8 before summing them up. The total results range from 0 to 128, where a higher result indicates a greater

level of CAS. The psychometric qualities of the Polish version of CAS-1 are presented elsewhere (Kowalski and Dragan, 2019). In the current study, CAS-1 had excellent internal consistency of Cronbach's α = 0.91.

### Ruminative Response Scale (RRS)

The 22-item Ruminative Response Scale focuses on one's responses to depressive mood: concentration on the self, symptoms, and the causes and consequences of depressive mood. A newer approach (Treynor et al., 2003) distinguishes two subscales: "Reflection" and "Brooding." Only the results of the latter are presented in this study. This subscale consists of five items with results ranging from 5 to 20, where a higher result indicates a greater tendency to respond to depressed mood with brooding. The Polish version of the RRS has generally good psychometric qualities (Kornacka et al., 2016). In the current study, the Brooding subscale had internal consistency of Cronbach's α = 0.88.

### Metacognitions Questionnaire – Short Version (MCQ-30)

The short version of the Metacognitions Questionnaire, developed by Wells and Cartwright-Hatton (2004), consists of five subscales and 30 items. It concerns metacognitive beliefs:

monitoring techniques, judgments, and beliefs about one's thoughts and cognitive abilities central to the metacognitive model of psychopathology. Two subscales are of interest in present study: the "Uncontrollability and Danger" scale explores the negative aspects of worry, e.g., "My worrying is dangerous for me" and the "Need to Control Thoughts" scale deals with beliefs about the negative consequences of not controlling one's thoughts, e.g., "Not being able to control my thoughts is a sign of weakness." The Polish version of this questionnaire exhibits good psychometric qualities and is considered equivalent to the English version (Dragan and Dragan, 2011). In this study, these two MCQ-30 subscales had good internal consistencies of α = 0.89 and α = 0.84, respectively.

### Symptom Checklist 27 Plus (SCL-27-Plus)

This is a checklist-type questionnaire that measures depressive, vegetative, agoraphobic, sociophobic, and pain symptoms (Hardt, 2008), and it allows the calculation of a global severity index (GSI). The results on each scale can range from 0 to 20, where higher scores indicate higher levels of a given symptom. In this study, the Polish adaptation of the questionnaire was used (Kuncewicz et al., 2014) and it had an excellent internal consistency of Cronbach's α = 0.93.

### Positive and Negative Affect Schedule (PANAS)

This is a comprehensive measure of emotions with two distinct subscales of positive and negative affect (Watson et al., 1988). In this study, a Polish adaptation of the 30-item PANASstate questionnaire, which has good psychometric qualities, was used (Brzozowski et al., 2010). In the current study, the internal consistencies of its subscales were α = 0.82 and α = 0.80, respectively.

### State-Trait Anxiety Inventory (STAI)

A widely used measurement of anxiety and its cognitive and vegetative components (Spielberger et al., 1970). In this study, a Polish adaptation of the STAI-state questionnaire, which has good psychometric qualities, was used (Wrze´sniewski et al., 2002). In the current study, the internal consistency was Cronbach's α = 0.93.

### Resting State fMRI

The resting state procedure consisted of a fixation cross being shown for 10 min on the MRI display (cf. Birn et al., 2013; Patriat et al., 2013). Subjects were instructed to fix their gaze on the cross and to not move.

Modified Rumination Induction (RumInd-M) fMRI Task

During rumination induction, participants are asked to think about sentences that are designed to induce the process of rumination (Nolen-Hoeksema and Morrow, 1993). The sentences deal with themes of the reader's own emotions, appraisals, and experiences. In this task, we used the mix of stimuli used by Cooney et al. (2010; rumination induction) and by Paulesu et al. (2010; worry induction) to obtain a robust repetitive negative thinking effect in participants. We used the modified procedure from Cooney et al. (2010) with ruminative/worrying sentences (e.g., "Think about the opportunities you didn't take in your life," "Think about what worries you have about your health"; RUM), and abstract sentences (e.g., "Think about how a plant grows"; ABS) as a control condition (see **Appendix 1** for all stimuli used). Participants were asked to think about sentences presented on screen and to try to clear their minds when a cross appeared on screen. Each sentence was presented on screen for 30 s and sentences were separated by 10 s of a fixation cross. Four blocks of five sentences were presented in a non-consecutive order (RUM-ABS-RUM-ABS). After each block, participants assessed their sadness, anxiety, and engagement in thinking on a 1–5 Likert scale. Results from this task are the totals of the assessments from both blocks of the same type. The task lasted about 15 min. Two parallel versions of rumination induction were used. Versions did not differ on any of the results (all values of p > 0.05) and administration of the versions did not differ between HCAS and LCAS groups, χ <sup>2</sup> = 0.43, p = 0.51.

# Behavior Analysis

Internal consistency was calculated with Cronbach's α. Group differences were analyzed with Student's t-test for independent samples or χ 2 for nominal data, group differences were calculated to demonstrate effect sizes using Cohen's d. Data were analyzed with IBM SPSS 24, effect sizes were calculated using an online calculator<sup>1</sup> .

# MRI Data Acquisition and Analysis

Data were acquired using a 3T Siemens MAGNETOM Trio system (Siemens Medical Solutions) equipped with a 12-channel head coil: structural T1-weighted image (TR: 2,530 ms, TE: 3.32 ms, flip angle: 7◦ , voxel size: 1 × 1 × 1 mm, field of view: 256 mm, measurements: 1), rsfMRI (TR: 2,000 ms, TE: 28 ms, flip angle: 80◦ , voxel size: 3 × 3 × 3 mm, field of view: 216 mm, measurements: 200), and task fMRI (TR: 2,500 ms, TE: 28 ms, flip angle: 80◦ , voxel size: 3 × 3 × 3 mm, field of view: 216, measurements: 364). After the rsfMRI and rumination induction tasks, B0 inhomogeneity field maps were collected (TR: 400 ms, TE: 4.5 ms/6.96 ms, flip angle: 60◦ , voxel size: 3 × 3 × 3 mm, field of view: 216 mm, measurements: 1).

The DICOM series were converted to NIfTI and BIDS data formats with Horos Bids Output<sup>2</sup> . Spatial preprocessing was performed using Statistical Parametric Mapping (SPM12<sup>3</sup> ). Functional images were corrected for distortions related to magnetic field inhomogeneity, corrected for motion by realignment to the first acquired image, slice-timed, normalized to the MNI space, and resliced to obtain a resolution of 2 × 2 × 2 mm, and smoothed with the 6 mm FWHM Gaussian kernel. Before normalization, structural images were coregistered to the mean functional image and segmented into separate tissues using the default tissue probability maps. Functional data were also analyzed with the Artifact Detection Toolbox (ART<sup>4</sup> ). Any EPI which deviated from the previous one by 3SD, 1.6 mm, or

<sup>4</sup>https://www.nitrc.org/projects/artifact\_detect

<sup>1</sup>https://www.psychometrica.de/effect\_size.html

<sup>2</sup>https://github.com/mslw/horos-bids-output

<sup>3</sup>http://www.fil.ion.ucl.ac.uk/spm/

0.04 rad was considered an outlier and such EPIs were regressed out in the 1st level models. Averages of 4.12%, SD = 2.64%, of scans for the rumination induction task and of 4.74%, SD = 4.13%, of scans for rsfMRI were regressed out. Participants with more than 20% outliers were excluded from the analyses. Based on these criteria no participants were excluded. There were no differences between groups in the number of outliers in the rumination induction task (t = 0.23, p = 0.82) or in the resting state (t = −1.76, p = 0.08), there were also no differences in the number of outliers between RUM and ABS conditions (t = 0.23, p = 0.82). Functional data were high pass filtered (1,000 s for rumination induction and 128 s for rsfMRI), and fixation crosses in the rumination induction task were modeled as baseline. Data were analyzed as a flexible factorial model of group × condition activation and with a two sample t-test of RUM > ABS and ABS > RUM contrasts. A regressor with a mock variable for gender was added to the second level models. On a group level, a voxel-wise height threshold of p < 0.05 corrected for multiple comparisons using the family wise error (FWE) rate was employed for whole brain analyses. Thresholded fMRI maps and raw data are available to any researcher upon request.

#### Functional Connectivity Analyses

The CONN (ver. 18<sup>5</sup> ) toolbox was used to perform functional connectivity analyses. First level SPM files and functional data for the resting state and rumination induction were imported into the software. Data were denoised with use of the respective T1 weighted scans, normalized to MNI-space, with eight regressors for WM and seven regressors for CSF, and with movement parameters obtained with the ART toolbox. The acceptance threshold for denoised signal voxel-to-voxel correlations was on average r ≤ 0.1. Resting state connectivity was calculated as HRF modulated pairwise correlations with seed-to-voxel analyses with a regressor for gender. RumInd connectivity was calculated as HRF modulated pairwise regressions with seed-to-voxel analyses of the generalized psychophysiological interaction (gPPI; McLaren et al., 2012) of group (HCAS and LCAS) versus condition (RUM and ABS) interactions with a regressor for gender. To make things clearer, η 2 , the effect size for the interaction analysis, was transformed into Cohen's d using an online calculator (see footnote 1). The threshold for significance was set at p ≤ 0.05 with false discovery rate cluster correction (FDRc). Figures depicting the connectivity analyses were made with use of MRIcroGL<sup>6</sup> .

## Seed Definitions

ROIs (regions of interest) chosen for functional connectivity seeds were based on main effects of the RUM condition from the rumination induction task and analysis of meta-analytic literature on the neural correlates of emotional disorders (i.e., depression and anxiety), these being conceptually most similar to CAS activation. Spheres of r = 6 mm were created over the obtained peak activations or the coordinates of peak activations provided by other authors. The MarsBar toolbox<sup>7</sup> was used to create ROIs. Talairach coordinates from meta-analyses were converted to MNI coordinates with the mni2tal calculator<sup>8</sup> . Nine ROIs were extracted from the RUM > ABS contrast from the rumination induction task: left and right precunei [−4 −58 32, −8 −52 28 and 6 −52 26], middle cingulate cortex [0 −18 36], L-paracingulate gyrus [−6 52 8], L- and R-superior frontal gyri [−2 56 38 and 6 52 28] and L- and R-frontal poles [−4 62 24 and 4 56 10]. Task-based ROI labels were based on an Harvard–Oxford anatomical atlas. Nine ROIs were extracted from meta-analyses on depressive and anxiety disorders: sub-callosal gyrus [2 16 −12], R-anterior cingulate cortex [10 30 −4] (Depression; Palmer et al., 2015), L-insula [−41 −3 −14], R-dorsal anterior cingulate cortex [−2 32 21], R-dorsolateral prefrontal cortex [30 10 50], and L-dorsolateral prefrontal cortex [−23 25 46] (Depression; Hamilton et al., 2012), L-insula [−42 14 −1] (Social anxiety disorder; Ipser et al., 2013), R-anterior cingulate [5 28 18], and R-middle frontal gyrus [41 9 40] (PTSD; Simmons and Matthews, 2012). Literaturebased ROI labels were based on nomenclature used by the authors of meta-analyses. Due to the long-block nature of the rumination induction task, we limited these analyses to cortical regions chosen as ROIs.

# RESULTS

# Behavioral Results

HCAS and LCAS groups differed strongly on all CAS measures (CAS-1, RRS-brooding, and MCQ-30 subscales) and all the subscales of SCL-27-plus used in this study. All differences were large in effect size with values of d > 3.5 for CAS measures and values of d > 1.3 for measures of psychopathology. There were more women in the HCAS group, for this reason, a mock variable for gender was added to the second levels of the fMRI and functional connectivity analyses. The groups also differed significantly with medium-to-large effect sizes on their assessments during rumination induction, both in RUM and ABS conditions as well as post-scan measurements of anxiety and negative emotions – for details see **Table 2**.

# Neuroimaging Results

Significant neural activations in the whole sample for RUM > ABS and ABS > RUM contrasts are presented in **Figure 3** and **Table 3**. The RUM > ABS condition yielded activations in bilateral precunei, bilateral superior frontal cortices, bilateral frontal poles, and the middle cingulate cortex. The ABS > RUM condition yielded several cortical activations: bilateral middle temporal gyri, bilateral supramarginal gyri, L-precentral gyrus, R-middle and inferior frontal gyri, and bilateral frontal poles. We did not find any differences between groups in neuronal activity in contrasts between RUM and ABS conditions in the rumination induction task, in the flexible factorial model, or in the two sample t-test models.

<sup>5</sup>https://www.nitrc.org/projects/conn

<sup>6</sup>https://www.mccauslandcenter.sc.edu/mricrogl/

<sup>7</sup>http://marsbar.sourceforge.net/

<sup>8</sup>http://sprout022.sprout.yale.edu/mni2tal/mni2tal.html

#### TABLE 2 | Behavioral results of RumInd-M task and post-scan assessments.


RUM, ABS, conditions in RumInd-M task; STAI, State-Trait Anxiety Inventory - State Version; PANAS, Positive and Negative Affect Schedule.

# gPPI Results

**Table 4** and **Figure 4** displays results of gPPI of group and condition interactions. The L-precuneus [−4 −58 32] showed increased connectivity with parts of the L-lateral occipital cortex and supramarginal gyrus in the HCAS group in the RUM condition in comparison to the LCAS group and decreased connectivity with bilateral parts of the precunei in the RUM condition in comparison to the LCAS group; opposite effects were observed in the ABS condition. The L-superior frontal gyrus showed decreased connectivity with parts of the L-superior parietal lobule and postcentral gyrus in the HCAS group in the ABS condition in comparison to the LCAS group and increased connectivity with the R-precuneus in this group in the ABS condition in comparison to LCAS group; opposite effects were seen in the RUM condition. Also, the L-precuneus [−8 −52 28] showed

TABLE 3 | Structure activations for both groups in RUM > ABS and ABS > RUM contrasts with FWE correction (p ≤ 0.05).


R, right hemisphere; L, left hemisphere; <sup>∗</sup>one cluster containing parts of two structures.

increased connectivity with bilateral frontal poles in the HCAS group in the RUM condition in comparison to the LCAS group and the opposite effect was found in the ABS condition. There was also increased connectivity in the HCAS group in the RUM condition between the R-precuneus and parts of the L-angular gyrus and supramarginal gyrus in comparison to the LCAS group; the opposite effect was observed in the ABS condition. The R-frontal pole showed decreased connectivity in the HCAS group in the RUM condition with four effect clusters in the right

TABLE 4 | Group differences in gPPI rumination induction functional connectivity.


L, left hemisphere; R, right hemisphere; HCAS, high-CAS group; LCAS, low-CAS group; RUM, rumination condition in RumInd-M; ABS, abstract condition in RumInd-M.

FIGURE 4 | Seed and effect clusters for gPPI analyses. Yellow clusters depict increased connectivity in the HCAS group in the RUM condition and/or decreased connectivity in the ABS condition in comparison to the LCAS group, cyan clusters depict decreased connectivity in the HCAS group in the RUM condition and/or increased connectivity in the ABS condition in comparison to the LCAS group. Green clusters depict seeds with bidirectional effects. Beginnings of arrows mark the seeds and ends mark the effects. For details of seeds, see Table 4.

temporal and right parietal lobes (see **Table 4** for details) in comparison to the LCAS group; opposite effects were observed in the ABS condition. A similar pattern of connectivity was observed in the R-anterior cingulate cortex and its effect clusters – bilateral precentral and R-postcentral gyri, and R-pre- and postcentral gyri. All presented interaction effects are significant with large effect sizes of Cohen's d > 1.

# Resting State Functional Connectivity Results

The between-group differences in rsfMRI functional connectivity are presented in **Table 5** and **Figure 5**. The HCAS group showed increased connectivity in comparison to the LCAS group between the L-insula and the L-central opercular cortex and planum temporale. Similarly, stronger connectivity in the HCAS group was found for the seed in the R-dorsolateral prefrontal cortex leading to three resulting clusters in the R-occipital pole and intracalcarine cortex, R-occipital pole and lingual gyrus, and the L-intracalcarine cortex and lingual gyrus. On the other hand, there was decreased connectivity in the HCAS group in comparison to the LCAS group between the R-anterior cingulate cortex and the L-frontal pole. All differences were large in effect with all values of d > 1.

# DISCUSSION

The present study used the rumination induction fMRI task and rsfMRI method to disentangle differences in the neural functioning of people with elevated levels of CAS in comparison to people with low levels of CAS. We ensured that the groups had extreme characteristics by pre-selecting two subsamples of people with low and high results on various measures of CAS and, additionally, by excluding participants with non-extreme and inconclusive results on the day of the study. A series of self-assessment questionnaires before, during, and after the fMRI procedure was used to address different levels of CAS, psychopathology symptoms, and negative emotions.

# Group Differences in Self-Assessment

By their construction, the studied groups differed significantly on all used measures of CAS – the CAS-1 questionnaire, rumination, and metacognitive beliefs concerning the need to control thoughts as well as the perceived inability to control thoughts and the associated dangers. Nevertheless, both groups also differed in levels of psychopathology symptoms – both depressive (Papageorgiou and Wells, 2003, 2009; Fergus et al., 2012, 2013) and anxiety symptoms (Wells, 2005; Fergus et al., 2012, 2013), as well as pain symptoms. This result is in line with numerous studies on the relationships of psychopathology with somatic symptoms and complaints (Bair et al., 2003; Kroenke, 2003; Tsang et al., 2008). It is noteworthy that the groups did not differ in terms of physical illnesses and concerns reported in SCID-I (cf. Dragan and Kowalski, unpublished). The discrepancy between lack of difference in number of physical illnesses and concerns in SCID-I and large difference in self-reported levels of pain symptoms may be due to self-focused attention and threat monitoring in people with high levels of CAS, resulting in fixation of attention on bodily sensations that would otherwise go unnoticed. Such a mechanism would be consistent with an understanding of health anxiety based on the metacognitive model (Melli et al., 2018).

There were medium to large group differences in reported assessments of sadness and anxiety during rumination induction, but not in assessments of engagement. The HCAS group scored significantly higher on levels of these negative emotions not only when assessing their mood after the rumination condition but also, with smaller effect size, after reading the abstract sentences. Results from previous studies on patients with depression are mixed: in one study there were no differences in negative affect between MDD patients and controls during rumination induction despite initial differences (Berman et al., 2014), and another study (Burkhouse et al., 2017) found a significant effect of group, as remitted MDD adolescents had higher sadness ratings during both rumination and abstract conditions. Our study dealt with people with time-persistent high or low levels of CAS, so these results may indicate that CAS levels are a prominent characteristic related to experiencing negative affect during rumination induction. This could serve as an explanation of remitted MDD adolescents having higher negative affect scores at all times (Burkhouse et al., 2017) and current MDD patients (Berman et al., 2014) having such scores only initially, before rumination induction. This hypothesis needs to be verified by further studies which take these results about CAS levels into account. The large-effect group differences in levels of postfMRI assessments of anxiety and negative emotions are also in line with this interpretation. Unfortunately, we did not collect pre-rumination-induction assessments of affect, which would enable the comparison of effects of group as well as group and time interactions.

# Effects of Negative and Abstract Thinking

The results pertaining to main effects of conditions are partially in line with previous results about rumination induction (Cooney et al., 2010). The RUM > ABS direct comparison in our study revealed neural activations in the bilateral precunei, middle cingulate cortex, L-paracingulate gyrus, bilateral superior frontal gyri, and bilateral frontal poles. Cooney et al. (2010) reported a similar pattern of activations with larger parts of the frontal cortices as well as the occipital and temporal gyri, but using a lenient statistical threshold. This indicates engagement of the DMN (Greicius et al., 2003) with the most prominent activation in both precunei (Zhang and Chiang-shan, 2012). Precuneal activity is often linked to self-referential processing (Kjaer et al., 2002; Lou et al., 2004) and depressive rumination (Johnson et al., 2009; Cooney et al., 2010; Milazzo et al., 2014; Burkhouse et al., 2017). The medial parts of the prefrontal cortex are also associated with self focused attention (Gusnard et al., 2001) and emotional responses (Lane et al., 1997). Such a pattern

#### TABLE 5 | Group differences in resting state functional connectivity.


L, left hemisphere; R, right hemisphere; HCAS, high-CAS group; LCAS, low-CAS group.

of activation during negative thinking induction may reflect cognitive components of negative thinking, specifically selffocused attention and self-referential processing. There were no significant brain activations in regions involved in emotional processing in the RUM > ABS comparison, i.e., in the amygdalae, parahippocampal gyri, or insulae.

Interestingly the ABS > RUM contrast (not reported by Cooney et al., 2010) revealed strong activations in the bilateral

middle temporal gyri, bilateral supramarginal gyri, L-precentral gyrus, R-middle and inferior frontal gyri, L-precentral gyrus and bilateral frontal poles. Widely distributed cortical activations in parts of the frontal poles (considered functionally as the dorsolateral prefrontal cortex) and parts of the parietal lobes can be identified as parts of the CEN (Corbetta and Shulman, 2002). The activity of the CEN, in opposition to the DMN, is associated with performing cognitive tasks, attention functioning, and working memory. The CEN as well as middle temporal regions and supplementary motor areas are also part of the "task-positive network" (Fox et al., 2005), which is a net of functionally correlated regions engaged in attention and working memory. This may indicate that abstract sentences engaged participants in tasks that required their attentional resources and were cognitively demanding.

The obtained patterns of neural activity specific to negative and abstract sentences are different and emphasize cognitive differences between these two types of thinking. It is also worth noting that both the DMN and CEN are engaged in the process of mind wandering (Christoff et al., 2009). In light of our results, this may indicate that mind wandering is comprised of self-referential rumination and dwelling on abstract cognitions.

# Group Differences in Modified Rumination Induction

As rumination induction has scarcely been used to-date in fMRI studies, we based our hypotheses concerning group differences on results obtained by Cooney et al. (2010) in a group of depressed patients. We did not replicate these results, i.e., we did not uncover any significant group differences between HCAS and LCAS groups in rumination induction in the basic fMRI analysis. There may be several reasons for this. The first reason may be the very design of the rumination induction task: it is comprised of blocks of five sentences which each last 30 s and are divided by 10 s fixation crosses, which gives almost 200 s per block. This may subject the obtained data to physiological noise (Liu, 2016) or noise due to the instabilities of the magnetic field inside the scanner (Smith et al., 1999). As such, long blocks prevent the filtering of low-frequency changes in the fMRI signal. Thus, it would be recommended to use shorter blocks or event-related paradigms in future studies. The second reason may be that the sentences used in our study did not directly tap into the individual experiences of participants, but were more general, aiming to evoke rumination or worry in every person, regardless of their personal experiences. This may have resulted in weaker responses to the stimuli used. It may be expected that personalized ruminative sentences would evoke much higher responses in participants (cf. Berman et al., 2014; Burkhouse et al., 2017). Another reason may be the heterogeneity of obtained results, as high levels of CAS can manifest in different ways, with a person developing mood or anxiety disorders or comorbid disorders, producing differences on the cognitive level which could result in high variability of the fMRI signal across the whole brain. However, it is also possible that the results of Cooney et al. (2010) are not replicable. The authors used a rather liberal statistical threshold. Moreover they employed AFNI and AlphaSim software, in which a bug which elevates levels of false positive results has been identified (Eklund et al., 2016). Taking all the above into account, it is possible that in the rumination induction task used, brain activity related to repetitive negative thinking is similar in both sub-populations and potential between-group differences are not detectable with 'static' general linear model analysis. Thus we decided to seek possible between-group differences, delving into more dynamic temporal characteristics of brain activity, i.e., applying functional connectivity analyses.

# Generalized Psychophysiological Interactions

The results of this study provide the first evidence that high levels of CAS are related to disrupted patterns of functional neural connectivity. Moreover, the between-group differences were found not only during rumination and worry, but also in abstract thinking. We conducted a gPPI functional connectivity analysis using areas found to be active in the RUM condition as seeds as well as ROIs based on metaanalytical literature on mood and anxiety disorders. The results show disrupted functional connectivity in the HCAS group within the DMN – the precunei, the medial parts of the prefrontal cortices, and parts of the occipital cortex (Greicius et al., 2003; Zhang and Chiang-shan, 2012) – during evoked negative thoughts. This may indicate a heightened tendency toward self-referential thinking and focusing attention on the self (Raichle et al., 2001; Buckner et al., 2008). A similar pattern of functional connectivity was also found in depression and interpreted as an inability of MDD patients to downregulate cognitive activity broadly associated with the DMN (Sheline et al., 2009).

There was also an interaction indicating a pattern of heightened connectivity in the RUM condition and/or lowered connectivity in the ABS condition in the HCAS group in comparison to the LCAS group between the L-precuneus and bilateral ventrolateral prefrontal cortices (vlPFC), which play a role in emotion processing in MDD (Keedwell et al., 2005). Furthermore the vlPFC are associated with anxiety (in primates; Agustín-Pavón et al., 2012) and, more specifically, attention bias to both threatening and neutral stimuli in anxiety and anxiety related disorders (Sylvester et al., 2012) and PTSD (Fani et al., 2012). Previous research on adolescents (Guyer et al., 2008; Monk et al., 2008) has shown that functioning of the ventrolateral prefrontal cortex may be modulated by the amygdala in social phobia and GAD. Current results suggest that the functioning of the vlPFC is modulated by disrupted functioning of the DMN, particularly the precuneus, which may "override" the regulatory role of the vlPFC in emotional processing and indicates the proneness of HCAS subjects to attention bias in self-referential processing (Wells, 2009).

We also observed a disrupted connectivity pattern in parts of the DMN during the abstract condition in the HCAS group. Interaction indicating increased connectivity was found between medial parts of the frontal cortex and R-precuneus, as well as within frontal and parietal parts of the DMN, and also within the precunei. Diminished connectivity of the anterior part of the cingulate cortex, interpreted as part of the salience network (Peters et al., 2016), with medial parts of the somatosensory cortex was found in the HCAS group in both RUM and ABS conditions, as compared to the LCAS group. A similar pattern of connectivity was also found between part of the DMN – the medial part of the prefrontal cortex (mPFC) – and the medial part of the somatosensory cortex. The rostral part of the anterior cingulate cortex (ACC), which plays a role in the symptomatology of various emotional disorders (Etkin et al., 2006), was shown to modulate the activity of the amygdala in task (Etkin et al., 2006) and resting state (Margulies et al., 2007) fMRI. Diminished connectivity between the ACC, mPFC, and somatosensory cortex in the HCAS group may indicate the mechanism of disrupted regulation of perception of bodily sensations. This result may be in line with the higher scores on the pain and vegetative symptoms subscale of the SCL-27 plus in the HCAS group. Perhaps the disrupted connectivity of the ACC, mPFC, and somatosensory cortex is related to one of the core mechanisms of CAS – heightened vigilance and monitoring for threatening stimuli, including threatening bodily sensations, which is characteristic of anxiety and anxiety-related disorders (Wells and Carter, 2001; Esteve and Camacho, 2008; Ginzburg et al., 2014).

There was also an interaction indicating a decreased connectivity pattern in the RUM condition and/or increased connectivity pattern in the ABS condition in the HCAS group in comparison to the LCAS group between part of the mPFC, part of the DMN, and R-Heschl's gyrus, insular cortex, and R-planum temporale, which have been shown to be engaged in auditory (Storti et al., 2013) and language (Nakada et al., 2001; Buchsbaum et al., 2005) processing. These results are also consistent with diminished resting state connectivity in Heschl's gyrus and the planum temporale in high trait-anxiety participants (Modi et al., 2015). Taking into account that Heschl's gyrus is engaged in both task-elicited and spontaneous inner speech (Hurlburt et al., 2016), it may be hypothesized that the disrupted connectivity of the DMN, mPFC in this case, and parts of auditory and language circuitries reflects the tendency for repetitive negative thinking typical of HCAS participants (Wells, 2009).

These results may not only serve as evidence for difficulty in down-regulating DMN activity in HCAS subjects during ruminative and abstract thinking, but also suggest a more global pattern of functional connectivity during various types of thinking and diminished cognitive control (Peters et al., 2016). This conclusion is supported by higher amplitudes of changes in connectivity between conditions in the HCAS group in comparison to the control group (see beta values in **Table 3**). Different patterns of connectivity in the more cognitively demanding ABS condition between groups also suggests that high levels of CAS may be associated with disturbances in the performance of cognitive tasks observed in clinical groups (Austin et al., 2001; Bishop et al., 2004; Eysenck et al., 2007; Hammar and Årdal, 2009; Murrough et al., 2011), which is in line with the S-REF model and the metacognitive theory of psychological disorders (Wells and Matthews, 1994; Wells, 2009).

The described results are also in line with those showing connectivity disruptions in rsfMRI and task-based fMRI in MDD patients (Zhang et al., 2011; Sambataro et al., 2014; Palmer et al., 2015) and anxiety disorder patients (Ding et al., 2011; Lei et al., 2015). This suggests that clinical levels of psychopathology and clinical diagnoses may not be necessary to observe disrupted patterns of functional connectivity in the brain. High levels of CAS may serve as an underlying factor not only for the symptoms observed in various clinical afflictions, but also can be associated with corresponding patterns of neural functioning.

# Resting State Functional Connectivity

In the current study, we also examined functional connectivity from brain activity recorded during a 10-min-long resting state fMRI procedure. We found the HCAS group to be characterized by stronger connectivity between several brain regions as compared to the LCAS group. First, the HCAS group showed stronger functional connectivity between the posterior part of the insula, a region involved, inter alia, in emotional processing during memory retrieval (Phan et al., 2002) and part of the opercular cortex in the left hemisphere, which is associated with auditory imagery (Lima et al., 2015). This pattern of connectivity could reflect the process of repetitive negative thinking occurring in the HCAS group – with interplay between parts of brain associated with emotion processing during memory retrieval (Phan et al., 2002) and verbal imagery. Increased connectivity was also found between the R-dorsolateral prefrontal cortex, which is associated with working memory and a part of the CEN (Corbetta and Shulman, 2002), and medial parts of the occipital lobe cortex associated with word recognition and processing (Mechelli et al., 2000) and visual processing (Kozlovskiy et al., 2014). Perhaps this increased connectivity may reflect common activations of these structures on a daily basis during the frequent rumination, worry, and reflection of the participants in the HCAS group. This is consistent with the results of the questionnaires they filled-in immediately before the fMRI study. It is noteworthy that diminished, not increased, connectivity was found between frontal and occipital brain regions in patients with social anxiety disorder (Ding et al., 2011). This result was interpreted by the authors as disrupted processing of visual stimuli in social contexts. Similarly, our results may suggest that CAS is an underlying factor of the heightened salience of threatening social cues in social anxiety disorder. This calls for investigation in further studies, as the results of this and other studies are mixed.

There was also a pattern of decreased connectivity found in the HCAS group as compared to the control group. This pattern was observed between part of the ACC and part of the ventral frontal pole which, again, are parts of the salience and CENs, respectively. Disruption in this connection was found in patients with GAD and interpreted as a dysfunction of topdown control over emotion regulation (Mochcovitch et al., 2014).

In general, the obtained results can be understood as altered interplay between different brain networks in people with high levels of CAS. Similar abnormalities were reported in studies on different clinical disorders such as depression (Zhang et al., 2011; Mulders et al., 2015; Peters et al., 2016) and social anxiety (Ding et al., 2011; Liu et al., 2015). This points to CAS as a probable factor underlying the clinical symptomatology and disrupted neural functional connectivity in people with different clinical afflictions, or even in people without a current diagnosis but with a high risk of developing emotional disorders.

# CONCLUSION

To our knowledge, this is the first study to explore the neural correlates of CAS. In this study we showed that treatmentand diagnosis-naive people with high levels of CAS differ substantially from people with low levels of this syndrome on various psychopathology and affect measures. Nearly half of the HCAS group was diagnosed with at least one current psychiatric disorder, predominantly mood and anxiety disorders as well as PTSD. We also demonstrated a large difference in self-assessment in these groups during repeated induction of negative thinking. These serve as proof-of-concept results of the metacognitive theory of emotional disorders (Wells, 2009). Contrary to our first hypothesis, we had no success in replicating rumination induction results in depressed participants (Cooney et al., 2010), for which there may be methodological and theoretical reasons. Irrespective of previous results, we demonstrated that neuronal activity during negative thinking is strongly related to neural activation of the DMN and that brain activity patterns during abstract thinking resemble the CEN. We were able to demonstrate evidence for our two hypotheses regarding differences in functional connectivity between groups. We showed, that low- and high-CAS groups differed in measures of functional connectivity during rumination and worry as well as during abstract thinking and resting state fMRI: high levels of CAS were related to disrupted patterns of connectivity within and between various brain networks – the DMN, the salience network, and the CEN. Overall, our results suggest that people with high levels of CAS tend to have disrupted neural processing in the areas of self-referential, task-oriented, and emotional processing. The obtained results are broadly analogous to results

# REFERENCES


obtained in fMRI studies of different clinical groups with mood, anxiety, and PTSDs, which serves as an argument for recognizing high levels of CAS as an underlying factor of emotional disorders and their neural correlates. These results are consistent with the theoretical underpinnings of the metacognitive theory of psychopathology, suggesting a common mechanism of emotional disorders originating in CAS and laying the foundations for further exploration of neural correlates of CAS. Future studies should use different, better-established fMRI paradigms and more differentiated groups, such as people with high levels of CAS with and without clinical diagnoses.

# AUTHOR CONTRIBUTIONS

JK, MW, AM, and MD wrote the manuscript. JK and MD conducted the research. JK, MW, and AM analyzed the MRI data. MW and AM supervised the MRI part of the study. MD supervised the research and analyses.

# FUNDING

The authors were financed by the Polish National Science Centre OPUS grant 2015/17/B/HS6/04157. The project was realized with the aid of CePT research infrastructure purchased with funds from the European Regional Development Fund as part of the Innovative Economy Operational Programme, 2007–2013.

# ACKNOWLEDGMENTS

The authors would like to thank the entire personnel of the Laboratory of Brain Imaging, Nencki Institute of Experimental Biology, Polish Academy of Sciences for their support, especially Bartosz Kossowski, Dawid Drozdziel, and Jacek Matuszewski. ´

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg. 2019.00648/full#supplementary-material



functional connectivity. Brain Struct. Funct. 220, 101–115. doi: 10.1007/s00429- 013-0641-4



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Kowalski, Wypych, Marchewka and Dragan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Single Dose of the Attention Training Technique Increases Resting Alpha and Beta-Oscillations in Frontoparietal Brain Networks: A Randomized Controlled Comparison

Mark M. Knowles1,2 and Adrian Wells1,2,3 \*

<sup>1</sup> Manchester University Hospitals NHS Foundation Trust, Manchester, United Kingdom, <sup>2</sup> Division of Psychology and Mental Health, The University of Manchester, Manchester, United Kingdom, <sup>3</sup> Greater Manchester Mental Health NHS Foundation Trust, Manchester, United Kingdom

#### Edited by:

Nuno Barbosa Rocha, Politécnico do Porto, Portugal

# Reviewed by:

Anna Contardi, Università Europea di Roma, Italy Xu Lei, Southwest University, China Lora Capobianco, Manchester Mental Health and Social Care Trust, United Kingdom

> \*Correspondence: Adrian Wells adrian.wells@manchester.ac.uk

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 11 May 2018 Accepted: 31 August 2018 Published: 20 September 2018

#### Citation:

Knowles MM and Wells A (2018) Single Dose of the Attention Training Technique Increases Resting Alpha and Beta-Oscillations in Frontoparietal Brain Networks: A Randomized Controlled Comparison. Front. Psychol. 9:1768. doi: 10.3389/fpsyg.2018.01768 The Attention Training Technique (ATT) was developed with the aim of reducing selffocused attention and increasing executive control as part of metacognitive therapy. So far there is a paucity of data on the neurophysiological effects of ATT. In the present study we tested for specific effects to determine if attention control components of ATT elicit a specific signature that is different from passive listening. Thirty-six healthy volunteers were randomized to an active (follow instructions) or control (ignore instructions) condition. Resting state EEG was recorded for 3 min with eyes open and eyes closed before and after exposure to training, and the power of the theta, alpha, and beta-bands were analyzed in frontal, midline, and posterior electrodes. The active ATT condition enhanced alpha and beta-band activity during eyes-open, and frontal alpha during eyes-closed (p < 0.005). Frontoparietal changes in Alpha were generally accompanied by changes in Beta in the same brain regions of interest. However, these associations were largely significant in the active ATT rather than the control condition. No between-group differences were observed in the Theta-band. These results suggest a single dose of attention training increases alpha and beta-oscillations in frontoparietal networks. These networks are associated with top-down attentional or executive control.

Keywords: attention, attention training technique, therapeutics, psychophysiology, electroencephalography, executive control, metacognitive therapy

# INTRODUCTION

The Attention Training Technique (ATT; Wells, 1990) is a metacognitive treatment strategy grounded in the Self-Regulatory Executive Function Model (S-REF; Wells and Matthews, 1994, 1996) of psychological disorder. According to S-REF theory, a specific pattern of thinking called the 'Cognitive Attentional Syndrome' (CAS) is assumed responsible for the maintenance of emotional distress. The CAS consists of unhelpful modes of processing including inflexible self-focused attention, threat-orientated attention biases, and worry and rumination. When activated, the CAS leads to a loss of cognitive resources and locks individuals into extended patterns of negative

processing of threat. As a result, psychological change is impeded because the processing resources required for efficient topdown self-regulation are reduced. A key feature of metacognitive therapy (MCT; Wells, 2000, 2009) is the explicit modification of maladaptive attentional strategies and the knowledge concerning them. ATT was developed as part of MCT to help moderate CAS activation by increasing top-down attentional control and flexibility. ATT consists of auditory attentional exercises that require individuals to engage in executive control skills including selective attention, divided attention, and attention switching (for a comprehensive overview of ATT, see Wells, 2009).

A large body of experimental and clinical data supports the contention that components associated with the CAS are linked to negative emotional outcomes. Studies include those examining worry and rumination (e.g., Nolen-Hoeksema, 1991; Capobianco et al., 2018), inflexible attention (e.g., Liu et al., 2002; Kolur et al., 2006), attentional biases (e.g., Mogg and Bradley, 2005; Bar-Haim et al., 2007), and inefficient cognitive control (e.g., Arnsten and Rubia, 2012; Wiers et al., 2013). Aside from supporting the S-REF model, such studies also complement wider views within neuroscience highlighting the critical role that executive control processes play within psychopathology. For example, decreased prefrontal function has been observed across multiple psychiatric conditions and is thought to reflect inefficiency in top-down regulatory processes including attentional flexibility, working memory, response inhibition, and the planning and execution of adaptive responses (Miller, 2000; Miyake et al., 2000; Porter et al., 2007). An implication of these results is that treatments integrating strategies specifically designed to attenuate deficits in executive control are more likely to prove efficacious (Siegle et al., 2007). In particular, treatments such as ATT which aim to increase aspects of attentional control and flexibility are predicted to yield improved functional and neurocognitive outcomes (Wells, 1990; Wells and Matthews, 1996; Siegle, 1999; Ottowitz et al., 2002).

Although originally developed as part of MCT, ATT has since been recognized as an effective stand-alone treatment for both anxiety and depressive disorders (e.g., Fergus and Bardeen, 2016; Knowles et al., 2016). Furthermore, a number of efficacy trials have demonstrated that the specific attentional processes targeted by the technique (e.g., inflexible selffocused attention, attentional bias) are associated with improved executive control and symptom relief (e.g., Sharpe et al., 2010; Callinan et al., 2014; Fergus et al., 2014; Nassif and Wells, 2014). In addition to this clinical and experimental data, a small number of studies are also beginning to uncover the neurophysiological effects of the technique. For example, initial neuropsychological and functional magnetic resonance imaging (fMRI) data suggests that Cognitive Control Training, which combines ATT with a working memory task, enhances activity within the dorsolateral prefrontal cortex (dlPFC), improves executive control, and disrupts amygdala activity in unipolar depression (Siegle et al., 2007, 2014). Furthermore, initial data from functional near-infrared spectroscopy (fNIRS) studies has also demonstrated increased blood oxygenation in the right inferior frontal gyrus, the right dorsolateral prefrontal cortex, and the superior parietal lobule during ATT in comparison to a control condition (Rosenbaum et al., 2018).

In the present study, we sought to provide further insight regarding the neurophysiological effects of ATT by using electroencephalography (EEG) to evaluate change in oscillatory activity across the scalp. EEG methodology was selected for two primary reasons: first, we were interested to see whether change in tonic frequency power following exposure to ATT would yield increased activity in known areas associated with top-down executive control. For example, it is well established that alpha and beta oscillations are generated by frontoparietal executive control networks (e.g., Capotosto et al., 2009; Sauseng et al., 2009; Thut et al., 2011) and are thought to reflect engagement of executive skills including attentional control and the regulation of working memory (e.g., Hanslmayr et al., 2007; Klimesch et al., 2007; Haegens et al., 2011; Handel et al., 2011). It was therefore hypothesized that engagement of ATT would yield increased changes in alpha and beta-band activity in frontoparietal regions. Second, we were interested to learn whether the effects of ATT would yield a different oscillatory signature to other known forms of attention modification. For example, although increased theta activity has been traditionally linked to short and long-term memory (e.g., Fell et al., 2003; Vertes, 2005), it has also been reliably observed to reflect a relaxed, drowsy state during mindfulness and meditation-based techniques (for reviews, see Cahn and Polich, 2006; Ivanovski and Malhi, 2007; Chiesa and Serretti, 2010; Travis and Shear, 2010). It was therefore hypothesized that in comparison to these findings, the effects of ATT would yield little or no change in theta-band activity.

In order to test our predictions, we designed a randomized controlled comparison in which participants were assigned to either an active (follow ATT instructions) or control (listen passively but do not follow ATT instructions) condition. Restingstate EEG data were recorded before and after exposure to the ATT and tonic power change was investigated in the three frequency bands of interest: alpha, beta, and theta. This allowed us to separate the presumed mechanistic effects of ATT (engaging in attentional control strategies) from simple exposure to a therapeutic listening task. Hence, in doing so, this design provided us with a structurally equivalent control condition that allowed EEG within and between-group changes to be attributed to manipulation of the IV (engaged vs. passive exposure to ATT). Furthermore, as this was one of the first known EEG studies to evaluate the effects of ATT, we recruited a non-clinical group of healthy subjects whom were naïve to the technique. Thus, participants were not socialized to the metacognitive model as would normally be expected in routine clinical practice. This helped us protect against possible measurement bias and placebo effects, and also prevented the investigated mechanism (engagement of ATT's attentional exercises) from being disturbed by the influence of medication and/or psychopathology. From an ethical point of view, it is also important to first establish nonclinical neurophysiological effects which future clinical samples can be compared against (thus avoiding unnecessary testing of the latter group).

# MATERIALS AND METHODS

fpsyg-09-01768 September 18, 2018 Time: 19:3 # 3

This study was conducted in accordance with the Declaration of Helsinki (World Medical Association, 1964) and was approved by the University of Manchester Ethics Committee (ref number: 13214).

# Participants

Thirty-six student volunteers (22 female, 24.33 ± 6.99) gave written informed consent to take part in the study. Participants were recruited from the University of Manchester via poster advertisement and received either course credits or monetary remuneration for taking part. All participants had normal or corrected vision, were right-handed, and had no current or historical neurological or psychiatric conditions.

Participants completed a number of validated self-report measures prior to the trial to ensure equivalence between independent groups on measures of attentional control, metacognition, and current mood: the Attentional Control Scale (ACS: Derryberry and Reed, 2002), the Metacognitions Questionnaire-30 (MCQ-30: Wells and Cartwright-Hatton, 2004) and the UWIST Mood Adjective Checklist (UMACL: Matthews et al., 1990). The UWIST was also measured post-ATT in order to establish whether any change in mood occurred as a result of the technique (see Results). Participants also completed a post-manipulation check immediately after the study. This measure consisted of two questions: (1) 'How much did you find yourself moving your attention around as instructed during the audio recording?' and (2) 'How much did you find yourself listening passively without moving your attention around during the audio recording?' Participants were required to record their responses on a 0–100% Visual Analogue Scale (VAS).

# Experimental Procedure

Participants were randomly assigned<sup>1</sup> to an Active Condition (AC; n = 18, 10 female) or a Control Condition (CC; n = 18, 12 female) and all listened to the ATT recording. Those in the AC were required to follow ATT instructions (participant instructions: 'Please listen to the audio recording. You are required to follow the instructions') and those in the CC were required to ignore the instructions (participant instructions: 'Please listen to the audio recording. You are required to listen passively without following the instructions'). Participants were required to complete the post-manipulation check and a measure of current mood (UWIST) following ATT. The duration of the experiment was approximately 24 min: pre-resting state (6 min), ATT (12 min), post-resting state (6 min). Participants were debriefed following the study. There were no differences between groups on any of the pre-trial measures.

# EEG Recording

Continuous EEG was recorded at rest before and after exposure to ATT. Each recording lasted approximately 6 min

<sup>1</sup>http://www.randomizer.org/

in duration, with 3 min eyes-open (EO), and 3 min eyesclosed (EC). The order of EO and EC was randomly assigned and then counterbalanced across participants. The experiment was conducted in a light- and sound-attenuated, electrical shielded room at ambient temperature. Participants were seated comfortably on a chair and were requested to minimize eyeblinks and physical movements during recording. Participants were monitored during recording to ensure they did not fall asleep. EEG data were recorded using a 64-electrode BioSemi ActiveTwo amplifier conforming to the international 10–20 system (Jasper, 1958). Electrodes were attached in standard formation (details of BioSemi referencing and grounding conventions<sup>2</sup> ). The signal was digitized at 512 Hz with an open passband from 0.01 to 100 Hz. Horizontal and vertical electrooculograms were recorded using separate electrodes placed above and below the right eye and at the outer canthi of both eyes.

# Spectral Analysis

Continuous EEG data were imported into BrainVision Analyser (Brain Products GmbH, 2015). Data were re-referenced to the common average of electrodes across the scalp. Independent Components Analysis (ICA) was used across all for recordings (12 min in total) to remove ocular artifacts. Data were then reconstructed and segmented into 1s epochs, and spectral analysis was conducted using Fast-Fourier transformation (FFT) within pre-defined bands: Theta (4–7 Hz), Alpha (8–12 Hz), Beta (13–30 Hz). This yielded FFT average power values for each EEG frequency band expressed in log units, 10<sup>∗</sup> log10(µV 2 /Hz), as a measure of frequency density (activity) in all four recordings (pre-resting state EO/EC, post-resting state EO/EC).

Three topographic regions of interest (ROIs) were calculated by averaging power values across the following electrode sites: Anterior (AF7, Fp1, Fpz, Fp2, AF8, AF3, AFz, AF4, F7, F5, F3, F1, Fz, F2, F4, F6, and F8), Midline (FT7, FC5, FC3, FC1, FCz, FC2, FC4, FC6, FT8, T7, C5, C3, C1, Cz, C2, C4, C6, T8, TP7, CP5, CP3, CP1, CPz, CP2, CP4, CP6, and TP8), Posterior (P7, P5, P3, P1, Pz, P2, P4, P6, P8, PO7, PO3, POz, PO4, PO8, O1, Oz, and O2). Prior to statistical analysis, all data were normalized using natural logarithm (In) transformation and then pre-to-post resting-state change indices were calculated for each condition (i.e., post-minus pre-baseline resting state values). These represented unitary values of tonic power change following exposure to ATT and were assumed to reflect the extent to which neuronal synchrony was increased or decreased. The use of unitary index values was also selected in order to reduce the error variance for statistical analysis.

# RESULTS

All analyses were performed using IBM SPSS v22 (IBM Corp, 2013). The initial phase of analysis evaluated whether any differences were observed between or within groups on the pre-selected measure of mood state (UWIST), and whether

<sup>2</sup>http://www.biosemi.com/faq/cms&drl.htm

any between-group differences were observed on the postmanipulation check (which was designed to assess compliance with the task instructions). This was followed by a planned evaluation of differences between groups on EEG tonic power changes across the frequency bands (alpha, beta, and theta). Here, the primary variable of interest in the EEG data was the effect of engagement with ATT (active condition) on spectral power in comparison to non-engagement/passive listening of ATT (control condition). Finally, an unplanned exploration of the correlation coefficients between band-power changes across both conditions was also conducted in order to learn more about whether ATT yielded a different oscillatory signature in comparison to that reported for other forms of attention modification (such as mindfulness and meditation). Given the pilot nature of these data, no specific corrections were employed for multiple comparisons during phase 2 and 3 of the analysis: this decision was taken to reduce the possibility of Type 2 errors given the relatively small sample size obtained. While we recognize that this limits the reliability of our findings, we felt that this was the most appropriate action to take given that the use of corrections may have obscured any possible effects.

Phase 1: In order to examine pre-to-post change in mood state, a 2 (condition) × 2 (time) mixed analysis of variance (ANOVA) was conducted on the four subscales comprising the UWIST – Tense Arousal (TA), Energetic Arousal (EA), Hedonic Tone (HT), and Anger Items (AI) – where condition was a between-subjects factor, and time was a within-subjects factor. No significant main or interaction effects were observed on any of the subscales (all p > 0.05) indicating that mood state did not differ between time points. In order to assess whether participants followed experimental instructions, oneway ANOVAs were conducted on the post-manipulation checks. A significant difference was observed between groups on Question 1 [F(1,35) = 300.32, p < 0.001], with those in the AC (83.33 ± 11.11) yielding higher scores than those in the CC (19.83 ± 10.87). In contrast, a significant difference was observed between groups on Question 2 [F(1,35) = 141.99, p < 0.001], with those in the AC (19.17 ± 11.66) yielding lower scores than those in the CC (77.94 ± 17.38). These differences suggest that participants in each condition followed the respective instructions.

# Spectral EEG

Phase 2: In order to evaluate differences between groups on EEG tonic power changes across the frequency bands, a series of 2 (Condition: AC and CC) × 3 (ROI) mixed ANOVA's were conducted on tonic change indices for each frequency band (Alpha, Beta, and Theta) during EO and EC – where condition was a between-subjects factor, and ROI was a within-subjects factor.

Alpha: a significant main effect of condition was observed [F(1,34) = 4.25, p = 0.04] indicating elevated change in global Alpha activity for the AC in comparison to CC during EO. Despite a insignificant interaction (p = 0.37), inspection of the between-group comparisons confirmed that this effect was most evident in the Midline ROI [F(1,34) = 4.66, p = 0.04, d = 0.80]. In addition, a significant condition by ROI interaction effect was observed [F(2,68) = 4.02, p = 0.02] for Alpha during EC. Univariate analysis confirmed that this was caused by a significant group difference in the Anterior ROI [F(1,34) = 4.74, p = 0.04, d = 0.76] indicating elevated change in Alpha activity for AC in comparison to CC. No differences were observed for Midline or Posterior ROIs (p's > 0.05) during both EO and EC. Beta: no significant main or interaction effects were observed for Beta-band activity during EC (p's > 0.05). However, a significant main effect of condition was observed [F(1,34) = 4.91, p = 0.034] indicating elevated change in global Beta activity for AC in comparison to CC during EO. Theta: no significant main or interaction effects were observed for Theta-band activity during EO or EC (p's > 0.05) – **Figure 1** displays topographic plots representing the significant between-group differences in tonic change for Alpha during EO and EC and Beta during EO. To help supplement further interpretation of the overall between group differences, the means, standard deviations, and between-group Cohen's d effect sizes and 95% confidence intervals (CI) were calculated (see **Table 1**). Inspection of the means indicated that in general, the AC yielded a positive (increase) change in spectral band power across a majority of the ROIs for both the EO and EC conditions. In contrast,



<sup>∗</sup>P < 0.05. Negative values indicate a decrease in power, and positive values indicate an increase in power. Between-group Cohen's d effect sizes and 95% CIs are presented to the right of mean values and standard deviations. ROI = Region of Interest.

the CC appeared to yield a negative (decrease) change in spectral band power across a majority of ROIs for EO and almost half the ROIs for EC. These data thus indicate that the direction and pattern of change largely differed according to group: for the AC, greater positive change was observed to occur in the Anterior, followed by the Midline, followed by the Posterior ROIs in both Alpha and Beta across EO and EC. Theta, on the other hand, demonstrated minimal change across ROIs for both EO and EC. In contrast, the CC showed less consistency between ROIs in Alpha and Beta during EO and EC, and demonstrated greater negative ROI change in Thetaband activity across EO and EC (with the Anterior ROI most pronounced).

Phase 3: In order to evaluate associations of band-power change within and across ROIs, a series of exploratory bivariate correlational analysis were conducted across EO and EC for both conditions. Positive frontoparietal associations were observed between Alpha and Beta during EO, but these were only found to be significant in the AC (r's = 0.83 and 0.50, for Anterior and Midline respectively). In addition, the AC yielded significant positive frontoparietal associations between Alpha and Beta during EC (r's = 0.58 and 0.77, for Anterior and Midline respectively), which were only observed in the Anterior ROI for the CC (r = 0.84). These data thus indicate that the frontoparietal changes in Alpha were generally accompanied by changes in Beta in the same ROI. However, these associations were largely significant in the AC rather than the CC. In addition, inspection of within group correlations between ROIs for both Alpha and Beta were investigated to determine level of oscillatory synchrony between frontoparietal areas. As suspected, significant positive associations were observed between Anterior and Midline ROIs for Alpha during EO and EC in the AC (r = 0.68 and 0.66, respectively) but not the CC (r's = 0.16 and 0.31, respectively). Similarly, significant positive associations were also observed between Anterior and Midline ROIs for Beta during EO and EC in the AC (r's = 0.7 and 0.8, respectively) but not the CC (r's = 0.02 and −0.25, respectively). These data indicate that frontoparietal changes in Alpha and Beta were highly correlated between Anterior and Midline ROIs, however these associations were only significant in the AC rather than the CC. Finally, level of asymmetry between alpha ROIs was investigated to determine whether enhancement of Anterior regions led to suppression over Posterior sites. Both the AC and the CC demonstrated Alpha asymmetry (negative correlation), but this effect was again only significant in the AC (r = −0.92).

# DISCUSSION

The present study is the first to demonstrate that a single dose of the Attention Training Technique enhances resting alpha and beta-oscillations in frontoparietal networks known to be implicated in top-down attention and executive control. As predicted, participants in the AC showed significant elevated change in frontoparietal alpha and beta-band activity. Furthermore, anterior and midline ROIs in both alpha and beta were significantly correlated in the AC indicting greater degrees

of neuronal synchrony. In contrast, limited theta-band activity was observed in both the AC and CC. This oscillatory signature distinguishes ATT from other forms of treatment that employ attention modification tasks. For example, studies evaluating the effects of autogenic relaxation training and mindfulnessbased techniques have shown increased theta-band activity in association with relaxed, drowsy states (e.g., Brown, 1974; Austin, 1999; Chan et al., 2011); a finding also commonly associated with various forms of meditation (e.g., Delmonte, 1984; Andresen, 2000; Travis and Shear, 2010). In addition, such studies also tend to report either little to no change in beta-band activity (Dunn et al., 1999; Cahn and Polich, 2006) and/or decreased frontoparietal beta-band activity (e.g., Ikemi, 1988; Jacobs et al., 1996).

The role of beta-band activity has received growing interest due to a wealth of animal and human studies indicating that beta-band enhancement reflects engagement of frontoparietal networks assumed to be involved in top-down attentional control (e.g., Bisley and Goldberg, 2003; Gross et al., 2004; Basile et al., 2007; Swann et al., 2009). In addition, alpha and betaband enhancements are observed to co-occur during tasks involving information retrieval and selective attention (Zanto and Gazzaley, 2009), and both are reported to strongly correlate in recent biologically plausible neural network models evaluating working memory abilities (Lundqvist et al., 2011). These findings have given rise to the hypothesis that both frequencies may serve similar neurocognitive functions (Waldhauser et al., 2012). Given that ATT is designed to improve top-down attentional control and flexibility over competing sources of information, the observed combination of enhanced alpha, and beta sits in agreement with these findings. From a conceptual point of view, these findings also provide support for the hypothesis that ATT's neuronal mechanism of change may lie in the training of frontoparietal areas associated with top-down executive control. Indeed, recent imaging studies evaluating ATT have also reported similar findings; Rosenbaum et al. (2018) interpreted their results as evidence of ATT increasing areas of the cognitive control network and dorsal attention network (they also go on to point out that aberrant functioning in both these areas are known to lead to negative emotional outcomes).

In addition to identifying ATT's oscillatory profile, the current findings also highlight the important implication of engaging with the ATT instructions. As predicted by S-REF theory, those who passively experienced ATT without engaging in the technique (CC) showed static or decreased change in anterior and midline ROIs for both alpha and beta. This may suggest that it is not exposure to ATT per se which yields neurocognitive change, but the degree to which individuals engage in the attentional tasks. This finding was further supported by significant alpha asymmetry observed in the AC in contrast to the CC. Evidence suggests that alpha enhancement of frontoparietal networks associated with sustained and directed attention correlates negatively with posterior amplitude (e.g., Corbetta and Shulman, 2002; Jensen and Mazaheri, 2010). Greater alpha asymmetry in the AC is therefore interpreted as reflecting greater levels of engagement in the attentional tasks. Furthermore, the presence of significant alpha asymmetry again separates ATT from other forms of attention modification, such as mindfulness and meditation, that tend to show aligned anterior-posterior alpha symmetry (e.g., Satyanarayana et al., 1992; Lagopoulos et al., 2009) and/or midline-posterior asymmetry (e.g., Ivanovski and Malhi, 2007; Chiesa and Serretti, 2010).

These findings also have an important clinical implication when considered in the context of reduced prefrontal functioning, which has been widely observed across multiple psychiatric conditions (e.g., MacDonald and Carter, 2003; Blumberg et al., 2004; Meyer et al., 2004; MacDonald et al., 2005). For example, prefrontal dysfunction characterized by diminished tonic alpha power has been reliably observed in schizophrenic patients (e.g., Sponheim et al., 1994, 2000) and in studies investigating the neurophysiology of depression and anxiety (e.g., Henriques and Davidson, 1990; Thibodeau et al., 2006). However, common psychological treatments such as cognitive remediation (for reviews, see Kurtz, 2003; Bellack, 2004) and computerized attention modification paradigms (e.g., Amir et al., 2009; Bar-Haim, 2010) regularly struggle to yield superiority above treatment as usual and often fail to explicitly link change in neurophysiology with the techniques being applied (Siegle et al., 2007). In contrast, ATT is a clinically reliable strategy aimed at enhancing global top-down attentional and executive control which has now been shown to enhance tonic alpha and beta power in frontopareital networks. Although the current results were not directly evaluated in association with clinical phenomena, it seems reasonable to assume that the neurophysiological effects of ATT may be implicated in the improvement of prefrontal functioning.

This study has some important limitations. First, as noted above, these results are unable to determine whether the observed neurophysiological changes are accompanied by symptom reductions in clinical populations. Assessing this prospect will involve repeated measurement of tonic alpha and beta-band change during a full course of ATT treatment with a clinical sample in comparison to a control. This will also help determine whether ATT yields a dose-response effect in parallel with increased symptom change. Second, although this study was able to control for trait measures of attentional control and flexibility, and a state measure of current mood, we did not employ an attention-related behavioral measure. Furthermore, despite efforts to ensure successful randomisation and counterbalancing, this study was unblinded to the experimenter. Thus, future replications will benefit from blinded replications with supplemented measures of top-down attentional control. Third, given the small sample size, we are unable to determine whether some of the negative findings are false negatives; the trends toward significance here may reach significance with larger sample sizes.

# CONCLUSION

To our knowledge this is the first EEG study to evaluate the neurophysiological effects of ATT. A single dose of the treatment was observed to yield significant tonic alpha and beta-band

enhancement in frontoparietal networks known to be implicated in top-down attentional and executive control. The specific effect of enhanced frontoparietal alpha and beta-band activity in combination with static theta-band activity suggests ATT yields a different oscillatory signature to other forms of intervention such as mindfulness and meditation-based strategies. There is growing clinical and analog evidence to suggest that ATT exerts strong therapeutic effects. These preliminary data suggest that the biological effects of ATT can be readily detected, may be equally promising and present an exciting opportunity for new lines of enquiry examining its neural substrates.

# REFERENCES


# AUTHOR CONTRIBUTIONS

AW conceived the study idea and supervised the study. MK and AW designed the study and analyzed the data. MK collected the data. Both authors contributed to writing the manuscript.

# ACKNOWLEDGMENTS

The authors are grateful to Prof. Wael El-Deredy and Caroline Lea-Carnall for advice and supervision in EEG data acquisition and analysis.




**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Knowles and Wells. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

fpsyg-10-00023 January 22, 2019 Time: 17:28 # 1

# Shifting Instead of Drifting – Improving Attentional Performance by Means of the Attention Training Technique

Vincent Barth<sup>1</sup>† , Ivo Heitland<sup>1</sup> \* † , Tillmann H. C. Kruger1,2, Kai G. Kahl<sup>1</sup> , Christopher Sinke1,2† and Lotta Winter<sup>1</sup>†

<sup>1</sup> Department of Psychiatry, Social Psychiatry and Psychotherapy, Hannover Medical School, Hanover, Germany, <sup>2</sup> Division of Clinical Psychology and Sexual Medicine, Hannover Medical School, Hanover, Germany

#### Edited by:

Gerald Matthews, University of Central Florida, United States

#### Reviewed by:

Sarah Velissaris, Calvary Health Care Bethlehem, Australia Peter Myhr, Independent Researcher, Stockholm, Sweden

#### \*Correspondence:

Ivo Heitland Heitland.Ivo-Aleksander@ mh-hannover.de

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 10 August 2018 Accepted: 07 January 2019 Published: 23 January 2019

#### Citation:

Barth V, Heitland I, Kruger THC, Kahl KG, Sinke C and Winter L (2019) Shifting Instead of Drifting – Improving Attentional Performance by Means of the Attention Training Technique. Front. Psychol. 10:23. doi: 10.3389/fpsyg.2019.00023 Background: The Attention Training Technique (ATT) as part of Metacognitive Therapy (MCT) has shown to be a promising treatment element for several psychiatric disorders such as depression and anxiety. ATT predicts improvements of the ability to shift attention away from internal and non-relevant stimuli (e.g., ruminative thoughts) toward the relevant stimuli and aims to increase attentional flexibility and control. The current study investigated the impact of the Attention Training Technique on attentional performance.

Methods: Eighty-five healthy participants (29 in two doses ATT, 28 in four doses ATT and 28 in the control group; 18–37 years of age) were administered a test battery for attentional performance before and after an intervention of two doses ATT (23 min duration) vs. four doses of ATT (46 min duration) vs. a control condition (non-intervention audio file via headphones. The test battery measured selective attention, inhibition, working memory, and attentional disengagement and comprised the following tasks: dichotic listening, attentional bias, attentional network, stroop, 2-back and a 3-back.

Results: After ATT (both two and four doses), reaction time during dichotic listening was significantly faster compared to the control condition. Furthermore, reaction time to neutral stimuli in the attentional bias task was faster after four-doses ATT compared to two doses ATT and the control condition. We found a trend toward a reduced stroop effect for both ATT conditions compared to control group. There were no effects of ATT with regard to the attentional network task, the 2-back or the 3-back task.

Conclusion: This first empirical evidence suggests that ATT promotes specific attentional flexibility in healthy participants. Based on the same mechanism, ATT may have beneficial effects on attentional performance in clinical populations and might be a promising tool in both healthy and clinical participants.

Keywords: metacognitive therapy, attentional training technique, attentional performance, MCT, ATT, metacognition, healthy participants

# INTRODUCTION

fpsyg-10-00023 January 22, 2019 Time: 17:28 # 2

Attention is a central function of neural processing. In 1890, James stated: "Everyone knows what attention is. It is taking possession by the mind, in clear and vivid form, of one out of what seem several simultaneously possible objects or trains of thought. Focalization, concentration of consciousness are of its essence. It implies withdrawal from some things in order to deal effectively with others" (James, 1890, pp. 403–404). Thus, attention enables the organism to prioritize some processes while ignoring less important ones. The Self-Regulatory Executive Function model (S-REF; Wells and Matthews, 1994) proposes that psychological disorders, e.g., depression or anxiety, develop when the person's style of thinking and coping leads to prolonged maladaptive emotional responses. These thinking and coping styles, e.g., rumination, worrying, threat monitoring etc., form the cognitive attentional syndrome (CAS; Wells, 2005; Fisher and Wells, 2009). Rumination as an active coping style leads to performance deficits and reduced flexibility of the cognitive system (Wells and Matthews, 1996). In addition, it is characterized by inflexible attention and the reduced ability to shift attention toward relevant stimuli (Whitmer and Banich, 2007).

Metacognitive Therapy (MCT) is a psychotherapeutic treatment that is based on the S-REF model of psychological disorder (Wells and Matthews, 1994, 1996). It aims to decrease the strength or remove the CAS entirely by changing metacognitive beliefs and re-establishing attentional flexibility (Wells, 2009).

The Attention Training Technique (ATT) is a component of the MCT manual with the aim to reduce the CAS by reorienting attention, i.e., shifting away from self-focus (Wells, 1990, 2007). The ATTs main focus is to improve attentional control and attentional flexibility by combining three auditory attentional exercises: selective attention, attentional switching and divided attention (Wells, 2007). The aim of the ATT is to strengthen the ability to focus on demand and improve the ability to focus on multiple stimuli at the same time. ATT leads to increased attentional control and reduced ruminative thoughts (Papageorgiou and Wells, 2000). A recent metaanalysis has demonstrated the efficacy of ATT as a standalone treatment for depression and anxiety, yielding greater treatment gains than comparison groups (autogenic training, progressive muscle relaxation, etc, Knowles et al., 2016). There is initial evidence showing that the ATT improves self-regulation in children (Murray et al., 2016). This gives rise to the question of whether ATT has the ability to not only relieve symptoms, but also to improve attentional performance by increasing control and flexibility of attention. The aim of the present study is to determine whether cognitive performance and selective attention increases after ATT training in healthy participants. Whereas therapeutic effects of ATT have been demonstrated by several studies, there is little knowledge about which specific neuropsychological domains are affected by the training. Therefore, the present study investigates which domains of attentional performance may be improved after ATT in healthy participants and the amount of training needed to gain effects.

# MATERIALS AND METHODS

The sample consisted of 85 healthy students recruited from a German university. Participants confirmed to be free of psychiatric diagnoses according to ICD-10 Criteria in the last 3 month. Data of four participants were discarded due to incomplete or invalid recordings, resulted in a total sample size of n = 81. The sample was between 18 and 37 years of age (mean age: 23.7, SD = 3.6). 64.2% of participants were female, 35.8% were males. All study procedures were approved by the local medical ethical committee. All participants provided written informed consent in accordance with the Declaration of Helsinki. Participants received monetary compensation for participation.

# Procedure

Participants were randomly assigned to one of three groups: two doses ATT (n = 27), four doses of ATT (n = 27) and the active control group sham training (n = 27, for a detailed description see sham training). The procedure took place on two consecutive days (see **Figure 1** for the procedural overview). The participants sat in front of a 19 inch LCD-Screen (Samsung Syncmaster 914n) with Sennheiser HD 558 over-ear headphones. Participants first received general information regarding the experiment and subsequently provided written informed consent. Participants then completed a German version of the attentional control questionnaire (ACS, Derryberry and Reed, 2002). Afterward, all groups performed a test battery to assess attentional performance. This completed the experimental procedures for the two doses ATT and the active control group on day 1. The group of four doses ATT completed two training session of ATT using an audio file (23 min., for a detailed description of the audio file see ATT) after the test battery. Experiment length on the first day was approximately 42 min for two doses ATT and the control group and 65 min for the 4 doses ATT condition. On

fpsyg-10-00023 January 22, 2019 Time: 17:28 # 3

day 2, the two doses and four doses ATT groups started with listening to two training sessions of ATT (23 min). The control group listened to the control treatment (22 min, see sham ATT). Then, every group performed the attentional performance test battery as described below directly after the training session. The experiment ended with a debriefing. The total experiment length on day 2 was approximately 55 min for all groups. All cognitive tasks and the delivering of the audio files were programmed using neurobehavioral systems presentation <sup>R</sup> software version 18.3 (Neurobehavioral Systems, Inc., Berkley, CA, United States).

# Measures

The attentional performance test battery comprised six wellvalidated tasks. Tasks were presented in the order of the following description with short breaks in between. Every test started with a short exercise block to ensure participants followed the instructions of the tasks. After the training run at the start of each task, the instructions were repeated to ensure that participants understand the task correctly.

# Dichotic Listening

The first task was the dichotic listening task as described in Asbjørnsen and Hugdahl (1995). We used the dichotic listening task to measure whether ATT improved selective attentional focusing in the domain of auditory processing. Six consonantvowel syllables (ba, da, ga, ka, pa, and ta) were presented at the same time via audio files over headphones on both ears (each ear one syllable), with a total of 36 different combinations. Participants were instructed to focus on the relevant target on the preferred sides or forced listening condition (only the left or only the right side) and ignore the other stimuli. Syllables were presented in identical (e.g., left: ba / right: ba) or different combinations (e.g., left: ta / right: da). Instructions were given to determine the syllables in three conditions: freely identifying only one of both presented stimuli (free choice), identifying stimuli on the left ear (only left) and identifying stimuli on the right ear (only right). Participants were required to indicate the correct stimuli by pressing the first letter of the relevant syllable (B, D, G, K, P, or T) on the computer keyboard as quickly as possible. 36 trials (every possible combination of the syllables) were presented in each condition, with randomized variable intertrial interval (varied systematically from 750 to 1125 ms, M = 1000 ms).

Due to incomplete or invalid recordings, group sizes in the analyses were: sham ATT (n = 26), two doses ATT (n = 27), four doses ATT (n = 27). The Outcome variable was the weighted mean of all left and right ear correct reaction times in milliseconds in the forced listening condition. The T2−T1 difference of these weighted means were subject to analyses.

# Attentional Network Task

The attentional network task was used to assess performance within three domains of attention: alerting, orienting an executive control (Fan et al., 2005). Consistent with Fan et al. (2005), we used three cue conditions (no cue, center cue, spatial cue) and two target conditions (congruent and incongruent). A fixation cross was displayed at the center of the screen during the whole trial against a gray background. The trial started with showing either no-cue (fixation cross remained unchanged) or middle cue (asterisk on the position of the fixation cross) or spatialcue (asterisk above or below the fixation cross, on the position were the target appears) for 200 ms. Then the cues disappeared and a jittered pause (300–1050 ms, M = 495 ms) with only showing the fixation cross followed. The target stimuli consisted of a row of five black arrows pointing either left or right, displayed either below or above the fixation cross and participants were instructed to indicate the direction of the middle arrow. This arrow in the middle was flanked on both sides by two arrows in the same direction (congruent condition) or in the opposite direction (incongruent condition). Target and flanker arrows were presented for 1600 ms. Participants had to identify the direction of the centrally presented arrow by pressing the identical arrow buttons (left or right) on the computer keyboard. After giving a response, the arrows disappeared and only the fixation cross was presented for the intertrial interval for 1600 ms (range 800–2000 ms). Participants had to shift spatial attention from the fixation point to the target stimulus in each trial in order to determine the proper response (Fan et al., 2005). Font size of the arrows was 60 and 55 for the asterisks. The three cue condition allowed to measure alerting and/or orienting benefits by giving no cue (baseline), middle cue (alerting, temporally informative) and spatial-cue (alerting plus orienting, temporally, and spatially informative). 120 trials were presented.

Due to incomplete or invalid recordings, group sizes in the analyses were: sham ATT (n = 23), two doses ATT (n = 23), four doses ATT (n = 21). The five outcome variables were mean reaction times of correct hits (for the no cue, the middle cue and the spatial cue conditions), and mean reaction times for congruent and incongruent stimuli, which were conducted for alerting, orienting and executive control as described by Fan et al. (2005). The T2–T1 difference of these weighted means were subject to analyses.

# Emotional Dot Probe Task

The emotional dot probe task was used to measure selective attentional control in the visual domain. Similar to Donaldson et al. (2007), a permanent central fixation cross on the computer screen was presented with a word pair (one word above, the other below the central fixation point) displayed for 1000 ms. This was followed by a fixation cross for 400 ms. The target (asterisk) presentation appeared in the position of one of the words for 2 s. in each trial, one word had a negative valence and the other was neutral. Words were taken from the ANGST-Database (Schmidtke et al., 2014). Neutral words had a valence between −0.2 and 0.2, emotional words had a valence of <−2. Participants were required to indicate the position of the asterisk as quickly as possible by pressing one of two response buttons (left for the word above and right for the word below) on the computer mouse. Font size of the words was 65 and 55 for the asterisks. The target was either presented on the position of the emotional word or the neutral word. Fifty trials were presented per condition. The intertrial interval was jittered around 750 ms (range 500–1000 ms). Two conditions were recorded: asterisk in the position of the neutral or emotional word, whereas the neutral word condition stands for the attentional disengagement from fpsyg-10-00023 January 22, 2019 Time: 17:28 # 4

the emotional word toward the asterisk in the position of the neutral word.

Due to incomplete or invalid recordings, group sizes in the analyses were: sham ATT (n = 23), two doses ATT (n = 24), four doses ATT (n = 23). Outcome variables were the mean reaction times for neutral and the emotional correct responses in milliseconds. Analyses in the emotional dot probe task were conducted by subtracting emotional reaction times minus neutral reaction times in order to reveal the costs of attentional disengagement. Analyzed were the T2−T1 difference of emotional minus neutral reaction times and neutral and emotional reaction times additionally.

# Stroop Task

The stroop task (Stroop, 1935) was used to measure selective attention and executive control as inhibition in the process of parallel distribution processing model (See MacLeod, 1991). Trials presented the capitalized color words RED, YELLOW, GREEN, and BLUE for 1 s against a black background. In congruent trials, words were presented in its matching hue [e.g., BLUE in blue, colors used in RGB space: red (255,0,0), yellow (255,255,0), green (0,255,0), blue (0,0,255)]. Incongruent trials showed color words in a mismatching hue of the other three colors (e.g., BLUE in green). Participants had to indicate the hue of the words and ignore the semantic meaning of the color words. Participants gave their answers by pressing four keys, colored in the four named colors, on the keyboard, on the position of the letters S (red), X (yellow), K (green) and M (blue). An intertrial interval which jittered around 1750 ms (range 1500–2000 ms) was set before the next trial started. Font size of the words was 80. A total of 100 trials were presented, equally distributed across conditions (i.e., 50 congruent and 50 incongruent trials).

Due to incomplete or invalid recordings, group sizes in the analyses were: sham ATT (n = 26), two doses ATT (n = 27), four doses ATT (n = 27). Outcome variables were the mean reaction times of congruent hits and mean reaction times of incongruent trials in milliseconds. To index inhibition, we subtracted incongruent stimuli reaction times from congruent stimuli reaction times in order to determine stroop effect costs. Additionally, T2−T1 difference of incongruent minus congruent reaction times and incongruent reaction times were analyzed.

# 2-Back / 3-Back

The N-back task was used to measure working memory (WM) performance as described in Braver et al. (1997). We used a sequential letter task in the version of 2-back and 3-back. Participants had to determine whether the current letter was identical to the previous letter two trials (2-back) or three trials (3-back) before (see Braver et al., 1997, p. 57 for detailed description). Each displayed letter was presented for 1500 ms, followed by a 500 ms pause before the next letter appeared. In each version participants had to identify a target letter and non-target letter by pressing two keys (X for targets, M for nontargets). All 26 alphabetical letters were used in a randomized order, with no more than two targets in a row. One 2-back exercise block containing 10 letters was presented before the 2-back and 3-back tasks started. In addition, an experimenter verbally instructed participants by giving examples for the 2 and 3-back tasks. This was done with the purpose of ensuring every participant had understood the task. A total of 150 letters in each n-back task was presented, with 50 targets and 100 non-targets.

Due to incomplete or invalid recordings, group sizes in the analyses were in 2-back: sham ATT (n = 25), two doses ATT (n = 25), four doses ATT (n = 27) and in 3-back: sham ATT (n = 26), two doses ATT (n = 26), four doses ATT (n = 26). Outcome variables were the means of hits of target and nontarget reaction times in milliseconds. The T2−T1 difference of these weighted means were subject to analyses.

# ATT

The attention training technique was presented using a standardized audio file as described in the MCT manual (Callinan et al., 2014; Fergus et al., 2014). A German version of the ATT was used (available at http://www.metakognitivetherapie. de/). The audio file follows the instructions provided by the MCT manual (Wells, 2009). As described above, each training session consisted of hearing the ATT audio file twice. The first sound file included explanations often upcoming training (1 min), where the participants were instructed to focus a visual fixation point and not to suppress or avoid internal events (e.g., thoughts, emotions) while listening to the auditory stimuli. The ATT comprises three auditory attentional exercises and lasts 12 min in total. In the audio file six different sounds (a clock, church bells, bird song, insects, traffic and running water) are presented and a male voice gives instructions on what to focus the attention on. ATT audio file starts with selective attention (5 min), where the participants perceive instructions to give intense attention to a specific individual sound (e.g., the ticking of a clock) while resisting distraction by others. Participants are instructed to focus on the voice of the instructor and the six different sounds as well as sounds in the room around the person successively. The next part of ATT was the rapid attention switching (5 min), in which participants have to switch attention between different sounds and spatial locations with increasing speed as this phase progresses. The last exercise practices divided attention (1 min), in which participants have to expand the width and depth of their attention and attempt to process multiple sounds and locations simultaneously (Wells, 2009). After finishing the first ATT session, the subject had the option for a short break and afterward the task continued with another ATT session identical with the first session but without the initial explanation of the instructor (double training).

# Sham ATT

The control / sham training group listened to a non-treatment audio file (11 min each file, for two sessions 22 min total), which comprised the same six different sounds identical in order, duration and intensity as in the ATT, but without any verbal instructions (audio file is available at http:// www.metakognitivetherapie.de/). As in the ATT conditions, participants had a short break after hearing the first round of the sham training.

# ACS

A German version of the Attentional Control Scale (ACS; Derryberry and Reed, 2002) is a self-report measure of attentional control and attentional shifting. It comprises 20 items rated on a 4-point Likert scale (almost never, sometimes, often, always). The questionnaire measures the general capacity for attentional control. High scores on the ACS represent a good capacity in effortful attentional control, with the subscales of focused attention, shifting attention and attention flexibility. The Outcome variable was the total sum score of the ACS. The ACS score was included in order to control for potential confounding effects from pre-test attentional control ability.

# Statistical Analyses

fpsyg-10-00023 January 22, 2019 Time: 17:28 # 5

Analyses were conducted with SPSS Statistics version 23.0 (IBM Corp., Armonk, NY, United States) using repeated measure General Linear Model (GLM) and repeated Analysis of Variance (ANOVAs) for each task. For all analyses reported hereafter, a p-value of <0.05 was considered significant. To investigate whether ATT vs. sham training affect the respective attentional performance domains, we conducted repeated measures ANOVAs using ATT (2 dosed and 4 doses combined) vs. sham training as a factor and the respective outcome parameters per test (T1 vs. T2) as dependent variables. Age, gender, and ACS total score were used as covariates for all further tests to correct for potential confounding effects.

# RESULTS

# Sample Characteristics

There was neither a difference in age (p = 0.86) nor in ACS total score (p = 0.50) between all ATT groups. In addition, sex was evenly distributed across all groups [x 2 (2) = 0.75, p = 0.68]. There were no differences between the ATT (two and four doses combined) and sham training groups in age: (p = 0.98), ACS total score (p = 0.25) and sex [x 2 (1) = 0.61, p = 0.41]. In addition, all following significant and non-significant effects remained the same by excluding the covariates (age, gender, and ACS total score) from analysis.

# Dichotic Listening

The dichotic listening task was used to measure selective attentional focusing in the domain of auditory processing. Participants that received ATT were significantly faster in correctly responding than the sham training group (T2–T1) [F(1,75) = 5.17, p = 0.026, η<sup>p</sup> <sup>2</sup> = 0.065; see **Figure 2**]. Further analyses showed no differences between four and two doses ATT in reaction times (p = 0.59).

# Emotional Dot Probe

The emotional dot probe task was used to measure selective attentional control in the visual domain. There was no difference between ATT and sham ATT (T2–T1) in emotional minus neutral reaction times (p = 0.89). The ATT group (two and four doses combined) did not differ from the sham training group in

(SEM). <sup>∗</sup>p < 0.05.

neutral reaction times (p = 0.19) or in reaction times of emotional reaction times (p = 0.19).

There was no significant difference between four doses of ATT when compared to the sham training group in emotional minus neutral reaction times (p = 0.59). However, participants in the four doses of ATT group responded significantly faster to neutral stimuli in comparison with the sham ATT group [F(1,42) = 4.97, p = 0.031, η<sup>p</sup> <sup>2</sup> = 0.106, see **Figure 2**]. Furthermore, there was a trend toward faster reaction times for the four doses of ATT group vs. sham ATT with regard to emotional stimuli [F(1,42) = 3.22, p = 0.08, η<sup>p</sup> <sup>2</sup> = 0.071]. Two and four doses of ATT did not differ significantly with regard to the neutral reaction times. However, there was a trend showing that the group that received four doses of ATT were slightly faster in reacting to corresponding stimuli than the two dose ATT group F(1,42) = 2.97, p = 0.092, η<sup>p</sup> <sup>2</sup> = 0.066).

# Stroop

The stroop task was used to measure selective attention and executive control as inhibition. There was a trend for faster reaction times (incongruent – congruent reaction times) in participants that received ATT compared to the sham ATT group [F(1,75) = 3.12, p = 0.081, η<sup>p</sup> <sup>2</sup> = 0.040]. Two and four doses of ATT did not differ from each other with regard to stroop task costs (p = 0.9). Analyzing the incongruent reaction times, ATT group and sham ATT group did not differ from each other (p = 0.246). Additional, there was no group difference in incongruent reaction times between two and four doses ATT (p = 0.534).

# ANT

The Attentional network task was analyzed to measure alerting, orienting and executive control. There was no difference between ATT groups and sham training with regard to alertness (p = 0.28), in orienting (p = 0.53) or in executive control (p = 0.92).

# 2-Back / 3-Back

fpsyg-10-00023 January 22, 2019 Time: 17:28 # 6

The N-back task was used to measure working memory performance. 2-back and 3-back were analyzed by the means of hits of targets reaction times in milliseconds. ATT group and sham training showed no difference in reaction times of hits (p = 0.77) nor number of correct hits for the 2-back task. There was no difference between ATT group and sham ATT in nontargets reaction times in the 2-back task (p = 0.3). Furthermore, there was no differences with regard to 3-back reaction times of hits between ATT group and sham ATT (p = 0.55) nor number of hits. There was no difference between ATT group and sham ATT in non-targets reaction times in the 3-back task (p = 0.48).

# DISCUSSION

This study examined the impact of the ATT, a standalone treatment of the MCT manual, on attentional performance in a randomized single blinded pre-post comparison of two doses ATT, four doses ATT and sham ATT. We showed that participants who received ATT were faster in auditory selective attention during dichotic listening. In addition, participants who received four doses of ATT performed faster in attentional disengagement in the emotional dot probe task compared to sham training. Furthermore, there was a trend toward faster reaction times in participants who received ATT compared to sham ATT in responding to stroop effect costs. There were no effects of ATT with regard to the ANT and the 2-back/3-back tasks.

The attention training technique showed a positive impact on auditory selective attention. Two and four doses of ATT yielded faster responses when identifying syllables on the right or left side while ignoring the irrelevant stimuli on the contralateral ear. This demonstrates that ATT induces near transfer effects in the trained domain (i.e., auditory attention). Our findings suggest that attentional flexibility through improving selective attentional control can be trained via only two doses of ATT. There was a significant improvement regarding attentional disengagement from emotional stimuli of participants with four doses ATT as measured with an emotional dot probe task. This provides evidence for a specific training effect of disengagement in the domain of the visual attention and a transfer effect from the auditory to the visual modality. Training with ATT leads to increased attentional flexibility in the form of faster disengagement of attention from irrelevant and/or emotional stimuli toward relevant stimuli. Consistent with initial evidence for attentional disengagement from negative stimuli (see Callinan et al., 2014), we found additional support for a growth of attentional flexibility. These findings are consistent with the theory of MCT and ATT, hypothesizing an improvement of attentional control (Wells, 2009).

As a further outlook, ATT may not only reduce clinical symptoms as shown by Knowles et al. (2016), but also improve attentional performance in healthy and clinical population. In fields aiming to improve attentional abilities (e.g., highperformance athletes) or reduce deficits (disorders related to the CAS) it may be useful to train attentional control via ATT. For example, as stated by Wells (2007) ATT could reduce auditory hallucinations, as evident from two case studies (Valmaggia et al., 2007; Levaux et al., 2011). Training attentional flexibility could allow less maintenance of auditory hallucinations or ruminative thoughts. The current study provides a first step toward understanding the mechanisms of ATT and its effect on healthy populations.

In the domain of selective attention and executive control, operationalized by the stroop task, only a trend of ATT was shown. Possibly for ATT, more training sessions or more statistical power would be necessary to determine an effect in the domain of subject's ability to deflect task irrelevant information. This could be in line with the identified dose effect in the emotional dot probe task, and its similar underlying mechanism of blending out irrelevant task information. As the stroop measures not only selective attention but also executive control (see MacLeod, 1991), ATT training of two or four doses could enhance selective attention but not executive control in the sufficient amount to determine group differences in the stroop task data. Taken together, training ATT may not improve cognitive abilities in general, but rather the specific capacity of attentional disengagement.

We found no evidence for an ATT effect in the domain of the attentional network based on alerting, orienting and executive control. The theoretical background of alerting is defined by Posner and Petersen (1990) as achieving and maintaining an alert state. This domain is not defined as the main aim of ATT and therefore possibly explains the lack of training impact. Whereas orienting (selection of information from sensory input) and executive control (resolving conflict among responses) could be reasonably defined as a potential aim of ATT, both networks might be too close to the level of automatic processing and therefore not be modified through ATT as based on the S-REF-model (see Wells and Matthews, 1994). Furthermore, in comparison with mean reaction times of the Fan et al. (2005) sample (n = 16) of adults ranging from 18 to 36 years (e.g., congruent reaction times hits: M = 717 ms, SD = 110), our sample has demonstrated faster pre-training reaction times (congruent reaction times hits: M = 440 ms, SD = 44 ms). Hence, this study sample might be different as it consists of a better trained student sample. Furthermore, the Fan et al. (2005) study recorded the ANT using 228 trials to determine alerting, orienting and executive control, whereas our study used 120 trials and approximately 10 min of duration. Implementation might be too brief to record potential training effects in the three domains of attention. Further, four doses of ATT might be too little training to detect potential effects of ATT on the three parameters of the ANT.

We found no evidence for an effect of ATT in the domain of WM measured by the 2-back and 3-back tasks. 2-back and 3-back tasks were included in this study in order to determine whether fpsyg-10-00023 January 22, 2019 Time: 17:28 # 7

ATT related training effects might be observed in the area of WM. Research regarding the domains of WM has shown that it consists, besides attentional control, of primary and secondary memory (Shipstead et al., 2014). Consistent with the idea that ATT improves attentional control rather than WM, our data suggests that WM performance does not seem to be trained significantly through ATT.

Here, we successfully demonstrated the training effect of ATT on healthy participants. However, the statistical power of the current study is limited by its sample size (n = 81). Future studies should incorporate larger samples to enhance statistical power. In addition, increasing the number of trials could result in more accurate estimation of the training effect. We omitted using longer tasks as to not overstrain the cognitive resources of the participants as the overall cognitive load was already quite high (23 min training/sham training + six different tasks). Due to multiple comparisons in statistical analysis there could be increased Type I errors. As this study was exploratory, providing an overview of those domains potentially affected by ATT, confirmatory studies are necessary to validate the present results.

We found preliminary evidence for a dose-dependent effect of ATT. Data suggests that four doses of ATT yielded greater training responses than two doses. Whereas previous clinical studies investigated doses ranging from 1 to 11, Knowles et al. (2016) conclude that one to two doses ATT could yield immediately measurable effects in symptom reduction. Symptoms reduced substantially after three doses of ATT and remained stable throughout the trial of additional ATT doses. With regard to the effect of ATT on attentional performance, we found evidence for an effect of only two doses of ATT. While more studies with a wider range of ATT doses are necessary to determine the optimal dose of ATT, the current findings suggest that more than two doses ATT should be applied. Our study investigated the direct effect of ATT with the performance test directly following the last training session. While a direct effect of ATT was demonstrated, we cannot assess at present whether there is only a temporary effect of ATT on attentional performance. Clinical case-studies using a follow-up design with a dose range of 6 to 11 ATT sessions suggest sustained effects of ATT on symptoms after 6 or 12 months, respectively

# REFERENCES


(Knowles et al., 2016). To our knowledge, no follow-up studies have been conducted regarding ATT induced improvements in the domain of auditory and visual attentional control. Future studies should also investigate the duration of the observed effects and the ideal dose of ATT.

It might be worthwhile to investigate the corresponding clinical and neurological correlates of the demonstrated ATTbased attentional performance effects. It would be interesting to evaluate if training effects are larger in a clinical population than in healthy participants. Such studies would allow to evaluate if attentional control and flexibility mediate the reduction of symptoms and where it is related to the mechanism behind observed reduced clinical symptoms after training ATT. Taken together, this study is the first to show that ATT has a positive impact on attentional performance using an elaborate sham control condition. This suggests ATT as a promising tool to improve attentional performance.

# ETHICS STATEMENT

This study was carried out in accordance with recommendations of the ethics committee of the Hannover Medical School with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the ethics committee of the Hannover Medical School.

# AUTHOR CONTRIBUTIONS

All authors designed the original concept. VB recruited and instructed the participants under IH supervision. CS designed the experiments. VB, IH, and CS were responsible for data collection and data analysis. All authors wrote the manuscript.

# ACKNOWLEDGMENTS

We wish to thank all participants for participating in this study.


fpsyg-10-00023 January 22, 2019 Time: 17:28 # 8


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Barth, Heitland, Kruger, Kahl, Sinke and Winter. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# One Step Ahead—Attention Control Capabilities at Baseline Are Associated With the Effectiveness of the Attention Training Technique

Ivo Heitland<sup>1</sup> \* † , Vincent Barth<sup>1</sup>† , Lotta Winter<sup>1</sup> , Niklas Jahn<sup>1</sup> , Alev Burak<sup>1</sup> , Christopher Sinke1,2, Tillmann H. C. Krüger1,2 and Kai G. Kahl<sup>1</sup>

<sup>1</sup> Department of Psychiatry, Social Psychiatry and Psychotherapy, Hannover Medical School, Hanover, Germany, <sup>2</sup> Division of Clinical Psychology and Sexual Medicine, Hannover Medical School, Hanover, Germany

#### Edited by:

Gerald Matthews, University of Central Florida, United States

#### Reviewed by:

Ava Schulz, University of Zurich, Switzerland Bartosz Zurowski, University Medical Center Schleswig-Holstein, Germany

#### \*Correspondence:

Ivo Heitland heitland.ivo-aleksander@ mh-hannover.de †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology

Received: 21 June 2019 Accepted: 21 February 2020 Published: 31 March 2020

#### Citation:

Heitland I, Barth V, Winter L, Jahn N, Burak A, Sinke C, Krüger THC and Kahl KG (2020) One Step Ahead—Attention Control Capabilities at Baseline Are Associated With the Effectiveness of the Attention Training Technique. Front. Psychol. 11:401. doi: 10.3389/fpsyg.2020.00401 Background: Attentional control has been observed to play an important role in affective disorders by impacting information processing, the ability to exert top– down control in response to distracting stimuli, and by affecting emotional regulation. Prior studies demonstrated an association between attentional control and response to psychotherapy, thereby identifying attentional control as an interesting prognostic pre-treatment factor. Improving attentional control and flexibility is a cornerstone in metacognitive therapy (MCT), which is trained by the use of the Attentional Training Technique (ATT). However, as of yet, it remains unclear if pre-treatment attentional control is related to the effect of ATT.

Methods: An aggregated sample of 139 healthy participants [study 1: 85 participants, mean age 23.7 years, previously published (Barth et al., 2019); study 2: 54 participants, mean age 33.7 years, not previously published] performed an attentional performance test battery before and after applying ATT. Before ATT was administered, attentional control was measured using a well-established self-report instrument, i.e., the Attentional Control Scale (ACS; Derryberry and Reed, 2002). ATT was given in 2, 4, or 15 doses and compared to sham ATT. The test battery comprised a selection of established neurocognitive tasks: emotional dot probe, Stroop, 2-back, and dichotic listening.

Results: Sham ATT showed no interaction with ACS score on performance outcome in all tests. At four doses of ATT, ACS score was associated with training response, i.e., subjects with high self-reported attentional control before training showed the largest improvements post-training (all P-values <0.05; see Figure 3). At 2 and 15 doses of ATT, the ACS score was unrelated to training response.

Conclusion: This is a first attempt in understanding the optimal dosage in which ATT should be administered dependent on the individual characteristics of each subject pre-training. The current data suggest self-reported attentional control pre-training as a marker to determine an optimal individual ATT training profile. Future studies should

**160**

investigate if other domains of metacognitions also interact with training outcome and evaluate the extent to which this relationship transfers to clinical samples. If successful, assessing attentional control prior to treatment in clinical samples could be of use regarding personalized therapy plans and treatment outcome.

Keywords: metacognitive therapy, Attentional Training Technique, dose-dependent effects, Attentional Control Scale (ACS), MCT, ATT, attentional performance

# INTRODUCTION

Attentional control (AC) is described as the general capacity to control attention in relation to positive or negative information (Derryberry and Reed, 2002). AC comprises focusing and shifting attention. Derryberry and Reed (2002) describe attentional shifting, also referred to as orientation, as a process of attentional disengagement from one target, moving attentional resources to a new target and subsequently engaging the new attentional target. Attentional focusing is to the ability to intentionally hold attention to desired stimuli and to avoid shifting attention to irrelevant or distracting stimuli (Derryberry and Rothbart, 1988).

Several studies demonstrated that anxious participants with good AC were better in disengaging from threatening information (Fox et al., 2001, 2002; Derryberry and Reed, 2002). Furthermore, others did report that attentional focusing can predict anxiety scores in healthy participants, while attention shifting abilities can predict depression scores in the healthy population (Ólafsson et al., 2011). Accordingly, AC allows anxious persons to limit the impact of threatening information, whereas those with poor AC are more likely to be preoccupied by threatening cues (Derryberry and Reed, 2002; Mathews and MacLeod, 2005). In contrast, participants with higher trait anxiety and worrisome thoughts take longer to switch attention from neutral information to emotional (Johnson, 2009). Of note, the relationship between anxiety and AC seems to be bidirectional. That means that not only does high AC function as a buffer for anxious pathologies, but also, anxiety itself can decrease AC by impairing efficient functioning of the goaldirected attentional system (Eysenck et al., 2007). Following that line of thought, Eysenck et al. (2007) stated that potential adverse effects of anxiety depend on AC involving the inhibition and shifting of attention. These processes of initially shifting attention toward threat cues and subsequently holding attention toward the threat is explained by a dual process view (Mathews and MacLeod, 2005). Bottom–up activation of threat representations within a salience network could explain the initial attention shift toward emotional cues (Öhman and Mineka, 2001). Attention to threat cues in anxiety is explained by top–down activation of competing representations related to other goals by an AC system (Matthews and Mackintosh, 1998).

In addition to findings regarding anxious traits, AC seems to play an important role in a number of affective disorders like anxiety and depression (Gotlib et al., 2004; Eysenck et al., 2007; Buckman et al., 2019). Poor AC is associated with impaired emotion regulation in depression (Joormann and D'Avanzato, 2010; Koster et al., 2011; Joormann and Michael Vanderlind, 2014; DeJong et al., 2019). Similar to anxiety, impaired attentional disengagement from negative self-referent information is linked to depressive symptoms like rumination (Koster et al., 2011). Buckman et al. (2019) showed that self-reported AC pretreatment does predict the level of depressive symptoms posttreatment as well as the risk of relapse to depression. Koster et al. (2011), continuing that line of thought, suggest improving AC first in order to change one's habitual style of thinking in depression, while only verbal interventions might not aim directly at impaired AC. In conclusion, this suggests AC as an interesting prognostic pre-treatment factor regarding anxious and depressive pathologies.

One model describing the connection between affective disorders and (impaired) AC is the Self-Regulatory Executive Function model (S-REF; Wells and Matthews, 1994). The S-REF comprises three interacting levels: a level of automatic and reflexively driven processing units, a level of attentional demanding and voluntary processing, and a level of stored knowledge or self-beliefs (Wells and Matthews, 1994). Self-regulation is processed in a limited capacity at the voluntary processing level and relies on voluntary attention for execution (Wells and Matthews, 1994). Operations processed by the controlled processing system are guided by self-knowledge or self-beliefs (Wells and Matthews, 1994). In the S-REF model, attentional biases are a consequence of threat monitoring strategies in anxiety maintained by dysfunctional metacognitive beliefs. In patients that focus on channels associated with threat, demanding resources of voluntary attention can lead to impaired AC. AC strategies, and behind these, dysfunctional beliefs, might be a stress coping strategy. Furthermore, the style of thinking and coping is able to cause prolonged maladaptive emotional responses (Wells and Matthews, 1996).

Improving AC and flexibility is a cornerstone in metacognitive therapy (MCT), which is trained by the use of the Attentional Training Technique (ATT; Wells, 1990, 2007). The ATT is based on the S-REF model and aims to improve attentional flexibility by training selective attention, attentional switching, and divided attention. The ATT has been proven as an efficient standalone treatment for depression and anxiety (see Knowles et al., 2016). In a previous study (Barth et al., 2019), we demonstrated that two and four doses of ATT improve attention performance regarding auditory information (dichotic listening task) and attentional disengagement (emotional dot probe) in comparison to an active control group. Of note, a recent study demonstrated that only a single ATT session could already improve AC measured by the Stroop task (Fernie et al., 2019).

Derryberry and Reed (2002) developed a self-report questionnaire to measure AC as the general ability to deliberately

control, focus, and shift attention. This Attentional Control Scale (ACS) is used in this study to investigate the potential predictive power of pre-treatment attentional abilities on performance outcomes. Note that in the theoretical framework of the S-REF model, a self-report measurement such as the ACS potentially not only measures the self-evaluation of one's own attention abilities but also may measure metacognitive beliefs of participants about their ability to control, focus, and shift attention.

This study aims to investigate if differences in pre-treatment attentional capabilities will affect outcome differences depending on different doses of ATT. Therefore attentional and executive functioning in healthy controls was tested before and after different doses of attentional training. We hypothesize that the better the self-rated attention control, the higher the improvement through attentional training.

# MATERIALS AND METHODS

All study procedures were approved by the local ethical committee of Hannover Medical School. Written informed consent in accordance with the Declaration of Helsinki was provided by all subjects. All subjects received monetary compensation for participation. The current study comprises an aggregated sample derived from two independent studies performed in our lab, i.e., study 1 (Barth et al., 2019) and study 2 (Jahn et al., 2020). In total, the aggregated sample consists of 139 healthy participants.

# Procedure

Both studies were designed as randomized placebo-controlled trials. The procedures for both studies were largely similar. For an overview of the design of both studies, see **Figure 1**. Before the experiments started, participants were reported to be free of psychiatric diagnoses according to International Statistical Classification of Diseases and Related Health Problems (ICD-10) criteria in the last 3 months. In study 1, this was assessed using a short clinical interview with a clinician. In study 2, the German version of the Structured Clinical Interview for Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) (SCID) screening was used. In both studies, subjects first filled in the ACS questionnaire and performed a neurocognitive test battery at baseline on a computer. Participants were then subjected to either ATT or sham ATT in the lab. Subjects were trained with ATT/sham ATT on two consecutive days (study one) or on eight consecutive days (study 2). On the last day after the ATT training session, the neurocognitive test battery was performed again.

# Sample 1

The first sample consists of 85 healthy students recruited from a German university (for details, see Barth et al., 2019). Participants were between 18 and 37 years of age (mean age: 23.7, SD = 3.6). Data of four subjects were discarded due to incomplete or invalid recordings. For a detailed description of all experimental procedures for the first sample, see Barth et al. (2019) and **Figure 1**. The experiment took place on two consecutive days (see **Figure 1**). In this sample, the ATT/sham ATT manipulation comprised groups of two doses of ATT, four doses of ATT, and sham ATT with two doses of sham training. The four-dose ATT group started the training on the first day with two sessions of training after finishing the test battery. The two-dose ATT and sham ATT groups only performed the test battery on day 1. On day 2, all groups started with two sessions of training or sham training and completed the task battery afterward (see **Figure 1**).

# Sample 2

The second sample consisted of 54 healthy participants ranging from 25 to 50 years of age (mean age: 33.7, SD = 7.7). Of the subjects, 64.8% were female; 35.2% were male. Data of four subjects were discarded due to an incidental white matter lesion finding on MRI (N = 1 in the sham ATT group), depressive symptoms in the SCID screening at baseline (N = 1 in the sham ATT group), misunderstanding of task instructions (N = 1 in the sham ATT group), and falling asleep during functional magnetic resonance imaging (fMRI) measurement (N = 1 in the ATT group). In comparison with study 1, the main objective in this sample was to evaluate the neurobiological effects of ATT; therefore, a part of the neurocognitive test battery was conducted in an fMRI scanner. The fMRI data are currently being processed and will be presented in a separate report (Jahn et al., 2020). The first training and the last ATT training were performed in the lab, comprised two doses of ATT each, and were done 8 days apart (see **Figure 1**). In between, subjects were instructed to perform two doses of ATT daily at home (see **Figure 1**). Participants provided written documentation of these ATT trainings at home. The average amount of completed trainings was M = 14.8 SD = 2.1 for the sham ATT group and M = 14.9, SD = 2.2 for the ATT group.

# ATT and Sham ATT

The ATT was presented using a standardized German audio file as described in the MCT manual (Wells, 2009). The ATT's main focus is to improve AC and attentional flexibility (Wells, 2009). The ATT comprises three auditory attentional exercises: selective attention, attention switching, and divided attention. Each training session in lab and at home consisted of hearing the ATT audio file twice, in which only audio file 1 contained explanations of the upcoming training (for detailed description, see Barth et al., 2019). One session of ATT lasts 12 min in total, with instructions (1 min), selective attention exercise (5 min), rapid attention switching (5 min), and divided attention (1 min). The sham training group listened to a non-treatment audio file, which comprised the same sounds, duration, and intensity as in the ATT but without verbal instructions. In this report, four groups were investigated in total, i.e., 2, 4, and 15 doses of ATT and sham ATT (2 and 15 doses combined).

# Attentional Control Scale (ACS)

The ACS (Derryberry and Reed, 2002) is a self-report measure of AC, attentional focusing, and attentional shifting. It consists of 20 items rated on a four-point Likert scale (almost never, sometimes, often, always). The questionnaire was developed as an instrument to measure the general capacity for AC, with high sum scores indicating good AC. The ACS comprises two subscales measuring the capability to focus attention (ACS focus) and to shift attention dynamically (ACS shifting). The ACS questionnaire was completed on the first day before the test battery was performed.

# Neurocognitive Test Battery

The neurocognitive test battery comprised a number of wellvalidated tasks to assess attentional performance. In both samples, these were a dichotic listening task, an emotional dot probe task, a Stroop task, and a 2-back task. Additionally, a 3-back task and the attentional network task were included in sample 1. In sample 2, these tasks were excluded to account for the longer duration of the experimental procedures due to the fMRI measurement, and as the data from study 1 did not warrant further use. All tasks started with written instructions and a short exercise block to ensure participants followed the instructions.

# Dichotic Listening

The dichotic listening task was used as described in Asbjørnsen and Hugdahl (1995). The task was used to test whether ATT improved selective attentional focusing in the domain of auditory processing. Participants had to focus on one ear (first trial, left ear; second trial, right ear) while listening to different consonant– vowel syllables. These were presented simultaneously on both ears via headphones. For a detailed description of the task, see Barth et al. (2019). As described there, the outcome variable was the weighted mean of all left and right ear correct reaction times in milliseconds in the forced listening condition. The T2 minus T1 difference of these weighted means was subject to analyses. Due to incomplete or invalid recordings, group sizes in the analyses were: sham ATT, n = 51; 2 doses of ATT, n = 27; 4 doses of ATT, n = 27; and 15 doses, n = 25.

# Emotional Dot Probe

The emotional dot probe was utilized to measure selective AC in the visual domain. For detailed description of the task procedure and details, see Barth et al. (2019). The test procedure was similar to Donaldson et al. (2007). A word pair, with one above a central fixation point and one below, was displayed for 1,000 ms. In study 2, the word pairs and targets were presented left and right of the fixation cross in order to better match the used response buttons located at the left and right index finger. Due to a prolonged inter-stimulus-interval (ISI) for fMRI analyses, only 90 trials were presented (45 congruent and 45 incongruent) in the fMRI version of this task. For both versions, in each trial, one word had a negative valence, and the other was neutral. After the words

disappeared, participants had to react to a target (asterisk), which appeared either in the position of the emotional word or in the position of the neutral word for 2 s. Fifty trials were presented per condition. As there is a bias in humans to allocate attentional resources toward salient and emotional stimuli (Macleod et al., 1986), the condition in which the asterisk appears at the location of the emotional word is typically referred to as congruent, as attention is already allocated at the target location. In contrast, the condition in which the asterisk appears in the location of the neutral word is typically referred to as incongruent and requires attentional disengagement, as attention is allocated at the opposite location, leading to longer reaction times compared to the congruent condition. In study 1, subjects completed the emotional dot probe while sitting in front of a computer. In study 2, this task was conducted while participants were lying in the MRI scanner. Subjects had to press two buttons with a computer mouse (study 1) or two input devices for each hand with two buttons on each (study 2). The stimuli were presented on a 32 inch display from Neuro-Nordic-Lab (NNL) at the end of the scanner; participants were able to see the screen through a mirror right above their head. Outcome variables were the mean reaction times in milliseconds. As an index of task improvement, the T2 minus T1 difference for the reaction times was analyzed. Due to incomplete or invalid recordings, group sizes in the analyses were: sham ATT, n = 47; 2 doses of ATT, n = 23; 4 doses of ATT, n = 24; and 15 doses, n = 25.

# Stroop Task

The Stroop task (Stroop, 1935) was used to measure selective attention and executive control as inhibition described in the parallel distribution processing model (see MacLeod, 1991). Stroop task presented capitalized color words (RED, YELLOW, GREEN, and BLUE) against a black background. Two conditions were conducted: in congruent trials, words were presented in their matching color (e.g., the word BLUE in blue). In incongruent trials, words were presented in a mismatching hue of the other three colors (e.g., BLUE in red). Participants had to indicate the hue of the words and ignore the semantic meaning of the color words. One hundred trials were presented, which were equally distributed across conditions (50 congruent and 50 incongruent trials). For a detailed description, see Barth et al. (2019). In study 1, the Stroop task was performed in the lab while subjects sat in front of a computer. In study 2, the Stroop task was conducted while participants were lying in the MRI scanner. Participants had to press two buttons with the thumb and index finger of their left hand (red and yellow) and two buttons with the thumb and index finger of their right hand (blue and green). To ensure full understanding of the task, color–button correspondences were displayed at both sides of the screen on paper. The primary outcome variable was the mean reaction times of congruent hits and mean reaction times of incongruent hits in milliseconds. As an index of Stroop task improvements, the corresponding T2 minus T1 differences were analyzed. Due to incomplete or invalid recordings, group sizes in the analyses were: sham ATT, n = 50; 2 doses of ATT, n = 27; 4 doses of ATT, n = 27; and 15 doses, n = 25.

# 2-Back

The N-back task measures working memory (WM) performance as described in Braver et al. (1997). We used a sequential letter task in this version of 2-back, in which participants had to determine if the current letter was identical to the letter two trials before (see Braver et al., 1997, p. 57, for detailed description). Each displayed letter was presented for 1,500 ms, followed by a 500 ms pause before the next letter appeared. Participants had to respond to every letter and identify if the current letter was a target (identical with the letter two trials before) or a non-target by pressing two keyboard buttons. All 26 alphabetical letters were used in a randomized order, with no more than two targets in a row (for detailed description, see Barth et al., 2019). In total, 150 letters were presented, with 50 targets and 100 non-targets. Outcome variables were the means of hits of target reaction times in milliseconds. The T2 minus T1 difference of these means was subject to analyses. Due to incomplete or invalid recordings, group sizes in the analyses were sham ATT, n = 49; 2 doses of ATT, n = 25; 4 doses of ATT, n = 27; and 15 doses, n = 25.

# Statistical Analyses

All statistical analyses were conducted with SPSS Statistics version 23.0 (IBM Corp., Armonk, NY, United States). An alpha of 0.05 was used. To investigate if ACS score at baseline modulated ATT-dependent performance improvements, nonparametric correlations between ACS score and performance improvements (T2 - T1) were computed per task (dichotic listening, emotional dot probe, Stroop, and 2-back) and dosage (sham, 2 × ATT, 4 × ATT, 15 × ATT). Of note, the outcome (significant vs. non-significant) of all correlational analyses presented in the following did not depend on the choice of parametric (Pearson's r) vs. non-parametric correlations (Spearman's rho). That means all correlations reported in the following that were significant for Spearman's rho were significant when analyzed using Pearson's r. Furthermore, all non-significant results with regard to Spearman's rho remained non-significant when Pearson's r was computed.

# RESULTS

# Sample Characteristics

There were no differences in the ACS total score at baseline between all ATT groups (p = 0.23; sham ATT: M = 59.0, SD = 7.32; 2 doses of ATT: M = 56.22, SD = 6.25; 4 doses of ATT: M = 56.74, SD = 8.17; and 15 doses: M = 60.08, SD = 7.02). In addition, sex was evenly distributed across all groups [x 2 (4) = 0.895, p = 0.93]. As expected from the different inclusion criteria per study, the sample used for study 2 was significantly older than in study 1 (p < 0.01).

# Manipulation Check: Sham-Controlled ATT Effects Across Samples

As reported earlier (Barth et al., 2019), improvements across tasks were larger for the experimental groups that performed ATT than for the sham ATT groups. A brief overview of these

results is presented here; for a more detailed description, please see (Barth et al., 2019).

In sample 1, participants who received two doses of ATT and four doses of ATT showed larger improvements (T2 - T1) in the dichotic listening task [F(1,75) = 5.17, p = 0.026, η 2 <sup>p</sup> = 0.065], in the emotional dot probe task [only four doses: F(1,42) = 4.97, p = 0.031, η 2 <sup>p</sup> = 0.106], and, as a trend, in the Stroop task [F(1,75) = 3.12, p = 0.081, η 2 <sup>p</sup> = 0.040] when compared to sham ATT. There were no significant ATT vs. sham ATT effects with regard to the 2-back task (p = 0.77).

Detailed analyses of ATT vs. sham ATT data including fMRI will be presented in another report (Jahn et al., 2020). In brief, we replicated the ATT vs. sham ATT effects reported in sample 1. That means that subjects who received ATT showed larger improvements (T2 − T1) with regard to dichotic listening [F(1,45) = 4.158, p = 0.047, η 2 <sup>p</sup> = 0.085] and the emotional dot probe task [attentional disengagement: F(1,44) = 8.265, p = 0.006, η 2 <sup>p</sup> = 0.158] than the sham-control group. There were no effects with regard to the Stroop task (p's > 0.102) and the 2-back task (p = 0.457). An overview of the ATT vs. sham ATT effects from both samples is presented in **Figure 2**.

# ACS as a Factor in ATT-Dependent Performance Improvements per Dose

In subjects who performed sham ATT, there were no associations between ACS score at baseline and neurocognitive performance improvements (all p-values > 0.147). Furthermore, there were no associations between ACS score at baseline and neurocognitive performance improvements in the 2 × ATT group (all p-values > 0.149) or the 15 × ATT group (p > 0.421).

In subjects who performed ATT four times, however, a high ACS score at baseline was associated with larger performance improvements in the emotional dot probe task [rs(24) = −0.451, p = 0.027], the Stroop task [rs(27) = −0.479, p = 0.009], and the 2-back task [rs(27) = −0.684, p < 0.001]. ACS total score was not associated with improvements of dichotic listening reaction times in the four-dose ATT group. An overview of these results is displayed in **Figure 3**.

# DISCUSSION

The present study investigated if and to what extent individual differences in self-reported AC at baseline are associated with neurocognitive performance improvements after having performed ATT or sham ATT. For that purpose, two independent samples completed a baseline assessment of AC followed by a neurocognitive test battery and were then subjected to various doses of ATT (2, 4, 15, and a sham group). One day (sample 1) or 1 week (sample 2) later, they returned to the lab to complete the neurocognitive test battery again. In both samples, subjects showed larger improvements in the neurocognitive assessments after ATT than after sham ATT. Of note, this effect was unrelated to ATT dosage, meaning ATT-dependent improvements were not larger at 15 doses of ATT than at 4 doses of ATT. This might be attributable to a ceiling effect. As healthy subjects typically report higher AC than patients and do not suffer from a cognitive attentional syndrome (CAS), four doses of ATT might be all that's needed in improving attentional performance in that sample, with no additional benefits of more training. There were no ATT-dependent improvements in the 2-back task. As previously discussed in Barth et al. (2019), ATT does seem to train attention processes rather than basic WM performance, which is the process measured during the 2-back task. Hence, the absence of an ATT × 2-back improvement is consistent with that line of thought and previous findings (see Owen et al., 2005; Schmiedek et al., 2014).

Interestingly, the ATT-dependent improvement of neurocognitive performance was modulated by AC at baseline. Subjects who reported high AC pre-training showed larger neurocognitive performance improvements after only four doses of ATT, while no effects of pre-training AC were observed at 2 or 15 doses of ATT or after sham ATT. To our knowledge, this is the first reported link between pre-training AC and benefits of the ATT training, and underlines the importance of assessing pre-training individual differences in AC when ATT is applied.

Several mechanisms might be responsible for this effect. The ACS is a self-report instrument assessing AC capabilities, i.e., to focus and to switch attention. These are processes that ATT specifically aims to improve. As such, having a solid foundation of AC before being subjected to ATT might allow for an easier integration and application of ATT. Like with many other training programs, getting familiar with the program and getting used to the structure of the training is essential to integrate the learning experience. High pre-training levels of AC might allow for a faster switch from "getting used to" to training attentional flexibility. Therefore participants with high levels of AC might profit faster from training ATT. While ATT might be most beneficial for subjects with low baseline AC on the long-term, this group might simply need more training to achieve similar effects than an average- or high-AC group.

Of note, several different questionnaires have been studied to assess self-reported AC and metacognitive beliefs regarding attentional capabilities. The ACS stems from research on attentional biases and threat monitoring, which is most prominently found in anxiety disorders (Derryberry and Reed, 2002). Traditionally, the ACS is viewed as a measurement for AC capabilities rather than the corresponding (meta)cognitive beliefs. Recent studies (e.g., Quigley et al., 2017) have raised questions regarding that view by demonstrating a dissociation between the ACS and corresponding behavioral measurement for AC. Thereby, they made the suggestion that the ACS might be more closely related to perceptions and beliefs regarding AC than actual AC capabilities. This fits with the observation that the most consistent associations with the ACS have been reported regarding anxiety and depression (Ólafsson et al., 2011; Reinholdt-Dunne et al., 2013, 2019; Judah et al., 2014). In line with these findings, studies have shown that anxious and depressed individuals display negatively biased beliefs about themselves and their abilities, including AC (Beck et al., 1979; Chambless and Gillis, 1993; Spada et al., 2010; DeVito et al., 2019). Another questionnaire, the Meta-Cognition Questionnaire (MCQ; Cartwright-Hatton and Wells, 1997), was developed for a broader range of psychopathologies

and aims to measure beliefs about worry, threat monitoring, and the controllability of thoughts. Future research has to clarify if AC as measured by questionnaires is more related to measureable attentional capabilities or, rather, one's confidence and opinion regarding AC. If metacognitions are indeed measured with the ACS, our findings are in line with the concept that AC and flexibility are influenced by the metacognitive beliefs a person has. Those subjects who were more confident regarding their AC benefited sooner than those who had poorer beliefs about their AC.

Mechanistically, the strongest effects of pre-training ACdependent ATT change were found in the 2-back task, even though no overall differences between the ATT and sham ATT groups were found. Attentional performance, i.e., focusing attention on relevant tasks while processing previous stimuli in the WM in the current case, seems to be associated with AC abilities at baseline. A similar effect was found in the emotional dot probe task. Participants with good AC at baseline were faster in focusing and reacting to targets with emotional valence after completing ATT, while there was no ATT effect on attentional disengagement. This might be due to larger voluntary attention resources in participants with good AC, which might allow them to benefit even more from training with four doses of ATT. This is in line with the theoretical underlying mechanisms of the S-REF model (Wells and Matthews, 1996) stating impaired AC as a consequence of demanding voluntary attention resources by inflexible attention and a heightened threat bias.

Consistent with these statements, ATT-dependent improvements regarding attentional disengagement from irrelevant stimuli in the incongruent condition in the Stroop task were largest in high-ACS subjects. It seems that participants with good AC benefit more from training with four doses of ATT, which is

shown in faster disengagement from irrelevant stimuli. As Fernie et al. (2019) stated, AC is not necessarily bound to emotional stimuli but rather more generally to disengaging from irrelevant stimuli. In the dichotic listening task, there was no modulation of ACS score at baseline on ATT-dependent improvements. The absence of an ACS effect in this task might be due to the modality overlap, meaning that training in the auditory modality as done in ATT and subsequently performing an auditory task might be significantly easier. Following that line of thought, ATT-based training effects might already be rather high regardless of poor AC.

With regard to dosage effects, we found no general advantage of 15 doses of ATT > 4 doses of ATT, as all room for improvement seemed to be covered by four doses of ATT already. Hence, the absence of an ACS modulation at 15 doses suggests a potential ceiling effect. Using that amount of training, pretraining differences might have evened out and no longer play a crucial role in ATT-based improvements. Typically, a variety of treatment effects follow an inverted u-shape dose response curve. This phenomenon was first described by Yerkes and Dodson (1908) regarding arousal and performance and has since been translated to, amongst others, behavioral pharmacology [see Calabrese (2008) for an overview], the neurobiology of human learning (Baldi and Bucherelli, 2005), and optimal patient–therapist relationships during psychotherapy (Dinger et al., 2009). In all these examples, the "sweet spot" for optimal treatment benefits lies in the middle of the distribution, with the medium intensity, duration, or dosage of treatment having the highest relative benefits. In the current study, a link for the optimal training benefits was already found at four doses of ATT, with no benefits of 11 additional doses using a healthy sample. Of note, there were no disadvantages in additional ATT sessions, as effects of 4 doses of ATT and 15 doses of ATT were comparable.

Hence, four doses of ATT was shown to be the optimal dosage for a healthy sample with relatively normal AC capabilities at baseline. In a clinical sample with potentially lower pretreatment AC and greater problems regarding attentional flexibility, the optimal ATT dosage might be much higher. Following that line of thought, in a clinical setting, it might be worthwhile to account for baseline differences in AC when planning the dosage or when handling a patient's expectations. This idea is in line with findings from a recent clinical study (Buckman et al., 2019). In a small cohort of depressed patients, baseline ACS predicted treatment response as well as residual depressive symptoms post-treatment and relapse rate, independent of symptom severity at the beginning. Moreover, clinical improvements were accompanied by an increase in ACS score from pre- to post-treatment, further underlining the importance of AC.

Certain limitations should be taken into account when interpreting the results of the current study. First, while the sample size of the aggregated samples used here is considerable (N = 135), larger follow-up studies are needed to fully elucidate the relationship between pre-training AC and ATT effects. Second, we only used four different ATT dosages, i.e., 2, 4, and 15 doses of ATT plus a sham-control group. While that approach allowed for finding a dose-dependent effect of ATT when pre-training AC was taken into account, a more elaborate design could allow for a more complete understanding and to potentially discover "sweet spots" for individual training profiles based on pre-training AC. This might be done using either a sham-controlled within-subject design or more ATT dosage groups (e.g., 1 dose of ATT to 10 doses of ATT). Third, the duration between preand post-assessment was either 1 day or 1 week. This does not allow for conclusions regarding longer as well as inbetween time spans, which remains an interesting target for future studies. Fourth, both samples were significantly different in age, which limits the comparison of the 2- and 4-dose (sample 1) with the 15-dose (sample 2) group. Moreover, measurements for sample 2 took partly place in the MRI, leading to experimental changes and slightly different reaction times at baseline. This was accounted for by using reaction time improvements from T1 to T2 as an outcome measurement for all tasks in question. Fifth, the current study is limited to healthy participants. Translation of our findings to a clinical sample remains a very important task for the future. This also seems essential for using individual pre-treatment AC characteristics as potential biomarkers in determining individual ATT profiles in clinical practice. Sixth, note that this study combines data from two samples, with one previously published (Barth et al., 2019) and the other one being measured in the fMRI scanner. Due to the exploratory nature of this study, we did not correct for multiple comparisons, which would have slightly impacted the results for the pre-training AC × ATT findings. One out of three significant findings would narrowly exceed the alpha threshold (P = 0.027), while the other two survive Bonferroni correction (P = 0.009 and P < 0.001). As always advised regarding reports of novel associations, replication is preferred before stronger conclusions can be drawn.

For a long period of time, clinical practice has used a "one size fits all" mentality regarding various treatments and training techniques. In the last decades, numerous studies have demonstrated the importance of individual differences pre-treatment and their effects on treatment outcome (e.g., Haby et al., 2006; Lambert, 2017). This has led to a great spur in studies aimed at establishing biomarkers and usable heuristics for clinical practice, with great promise but, so far, limited success. We therefore believe that it is of utmost importance to continue the quest for personalized treatment plans in order to be able to offer optimal treatment guidelines and opportunities for patients. MCT and ATT in particular may be good targets for such an approach, as they are evidence-based and controllable psychotherapy methods with a clear definition.

Taken together, we here provide preliminary evidence suggesting pre-training AC as a factor in dose-dependent neurocognitive improvements following ATT. This suggests that self-reported AC pre-treatment might be used as a marker to determine an optimal individual ATT training profile. Future studies should replicate the current effects and investigate if other domains of metacognitions also interact with training outcome. Also, it remains crucial to evaluate the extent to which this relationship transfers to clinical samples. If successful, assessing AC prior to treatment in clinical samples could be of use regarding personalized therapy plans and evaluating treatment outcome.

# DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

# ETHICS STATEMENT

fpsyg-11-00401 March 28, 2020 Time: 18:58 # 9

All study procedures were reviewed and approved by the local ethical committee of Hannover medical school. Written

# REFERENCES


informed consent in accordance with the Declaration of Helsinki was provided by all subjects prior to participation.

# AUTHOR CONTRIBUTIONS

CS, IH, and LW designed the experiments. NJ and AB recruited the subjects and collected the data under IH's and VB's supervision. VB, IH, NJ, and CS were responsible for data processing. VB, IH, and CS analyzed the data. All authors wrote the manuscript.

# ACKNOWLEDGMENTS

We wish to thank all subjects for participating in this study and addisca GmbH for funding.



**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Heitland, Barth, Winter, Jahn, Burak, Sinke, Krüger and Kahl. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Efficacy of Metacognitive Therapy: A Systematic Review and Meta-Analysis

Nicoline Normann<sup>1</sup> \* and Nexhmedin Morina<sup>2</sup>

<sup>1</sup> Department of Psychology, University of Copenhagen, Copenhagen, Denmark, <sup>2</sup> Department of Clinical Psychology and Psychotherapy, Institute of Psychology, University of Münster, Münster, Germany

Background: Metacognitive therapy (MCT) continues to gain increased ground as a treatment for psychological complaints. During the last years, several clinical trials on the efficacy of MCT have been published. The aim of the current study was to provide an updated meta-analytic review of the effect of MCT for psychological complaints.

Methods: We conducted a systematic search of trials on MCT for young and adult patients with psychological complaints published until January 2018, using PsycINFO, PubMed, the Cochrane Library, and Google Scholar. Trials with a minimum of 10 participants in the MCT condition were included.

#### Edited by:

Adrian Wells, University of Manchester, United Kingdom

#### Reviewed by:

Asle Hoffart, Modum Bad Psychiatric Center, Norway Peter Fisher, University of Liverpool, United Kingdom

\*Correspondence: Nicoline Normann Nicoline.normann@psy.ku.dk

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 06 July 2018 Accepted: 25 October 2018 Published: 14 November 2018

#### Citation:

Normann N and Morina N (2018) The Efficacy of Metacognitive Therapy: A Systematic Review and Meta-Analysis. Front. Psychol. 9:2211. doi: 10.3389/fpsyg.2018.02211 Results: A total of 25 studies that examined a variety of psychological complaints met our inclusion criteria, of which 15 were randomized controlled trials. We identified only one trial that was conducted with children and adolescents. In trials with adult patients, large uncontrolled effect size estimates from pre- to post-treatment and follow-up suggest that MCT is effective at reducing symptoms of the targeted primary complaints, anxiety, depression, and dysfunctional metacognitions. The comparison with waitlist control conditions also resulted in a large effect (Hedges' g = 2.06). The comparison of MCT to cognitive and behavioral interventions at post-treatment and at follow-up showed pooled effect sizes (Hedges' g) of 0.69 and 0.37 at post-treatment (k = 8) and follow-up (k = 7), respectively.

Conclusions: Our findings indicate that MCT is an effective treatment for a range of psychological complaints. To date, strongest evidence exists for anxiety and depression. Current results suggest that MCT may be superior to other psychotherapies, including cognitive behavioral interventions. However, more trials with larger number of participants are needed in order to draw firm conclusions.

Keywords: metacognitive therapy, meta-analysis, psychotherapy, anxiety, depression, psychopathology, mental disorders

Metacognitive therapy (MCT; Wells, 2009) continues to gain ground as a treatment for psychological complaints. MCT is theoretically grounded in the self-regulatory executive function model (Wells and Matthews, 1994, 1996), which states that psychopathology arises as a result of a perseverative thinking style called the cognitive attentional syndrome (CAS). The CAS consists of dysfunctional coping strategies that a person employs as an attempt to manage distressful thoughts and feelings. These include worry, rumination, threat monitoring, thought control strategies,

**170**

avoidance, and reassurance seeking (Wells, 2009). The model proposes that negative thoughts and feelings are temporary in nature, however, when a person responds to these with CAS activity, this may cause extended psychological distress and may inadvertently exacerbate and prolong negative affect. The model further suggests that the CAS arises from a person's positive and negative metacognitive beliefs, i.e., beliefs about cognition. Positive metacognitions are beliefs about the need to engage in CAS activities, e.g., "Worry helps me stay prepared," whereas negative metacognitions are beliefs about the uncontrollability and dangerousness of thoughts and feelings, e.g., "I have no control over my worry/rumination" and "Feeling like this means I am losing my mind" (Wells, 2009).

In MCT, metacognitive beliefs and processes related to the CAS are identified and modified during treatment. The treatment is manualized, as outlined by Wells (2009). However, flexible application of the manuals is advocated to fit the specific patient's needs. Although MCT targets transdiagnostic processes, the exact case formulation model as well as combination of techniques vary depending on the disorder in question. The first step in therapy is to conceptualize an idiosyncratic case formulation together with the patient, and to socialize the patient to the maintaining processes, including the impact of worry and rumination and the ineffectiveness of current coping strategies. Next, metacognitive beliefs are verbally challenged in Socratic dialogues, and behavioral experiments are used to test and generate change in the person's metacognitive predictions or beliefs about CAS strategies. Main emphasis is laid on challenging the negative beliefs before moving on to challenging the positive metacognitive beliefs. The patient is instructed to postpone worry and rumination processes. The aim is for patient to experience that worry and rumination are processes that can be postponed by disengaging from further processing, that they are harmless, and have no advantages. Specifically designed therapeutic techniques, such as the attention training technique or detached mindfulness (Wells, 2009), are used. The attention training technique (Wells, 1990) is an auditory task that requires the patient to engage in selective attention, divided attention, and attention switching. It is designed to increase the patient's executive control and regain attentional flexibility. In detached mindfulness the patient is instructed to become aware of internal trigger thoughts and detach from them by taking a step back and disengaging any further coping or perseverative processing in reaction to them. The patient practices these new ways of reacting to trigger thoughts in therapy as well as between sessions, and their implementation is proposed to strengthen the patient's ability to disengage from worry and rumination processes. The techniques furthermore challenge the patient's belief that worry and rumination are uncontrollable. Toward the end of therapy focus is on reversing any residual CAS activity. Altogether, MCT aims at increasing the person's experience of attentional control, reducing self-focused attention, and fostering the development of adaptive beliefs and coping strategies.

Several clinical trials have examined the efficacy of MCT. Normann et al. (2014) meta-analytically summarized relevant trials on MCT that were published until early 2014. The authors incorporated 16 trials with patients with anxiety and depression and concluded that MCT is very effective in these populations. It must be noted, however, that only nine of the trials in this meta-analysis were controlled trials and most trials were based on rather small samples. Very recently, Rochat et al. (2018) assessed the efficacy of single-case studies on MCT in a meta-analytic review and also reported that these studies support treatment efficacy of MCT for anxiety, depression, and other psychopathological symptoms. Since the meta-analysis by Normann et al. (2014), several clinical trials on the effect of MCT have been published. Furthermore, the meta-analysis by Normann et al. (2014) focused on depression and anxiety disorders only. To address these limitations, the current study aimed at providing an updated review and meta-analysis on the effect of MCT. The main objective was to investigate whether MCT improves symptoms of psychological complaints on primary and secondary outcome variables in comparison to control conditions. For this purpose, we focused on both uncontrolled as well as controlled trials. With regard to the secondary outcomes, we aimed at assessing whether treatment has an impact on comorbid anxiety or depression as well as metacognitions.

# METHODS

The aims and methods of this meta-analysis have been registered with the International Prospective Register for Systematic Reviews, with ID number CRD42018084507 (available from https://www.crd.york.ac.uk/PROSPERO). The meta-analysis was conducted using the guidelines and checklist outlined by the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) Group (Moher et al., 2009). Accordingly, our main research question describing the Population, Intervention, Comparison, Outcome, and Study design (PICOS) was: In individuals with psychological complaints (P), does metacognitive therapy (I), compared to control conditions (C), improve symptoms of psychopathology (O) in randomized controlled trials (S)? However, due to the limited number of controlled trials meeting our criteria, we decided to also include uncontrolled trials and thus first examine the efficacy of treatment with respect to within-group effect sizes (i.e., change from pre- to post-treatment or follow-up). We first examined the efficacy of treatment from pre- to post-assessment for the primary outcome of all included studies. However, since uncontrolled effect sizes do not account for the impact of time on symptoms, we view controlled effect sizes as more reliable when it comes to assessing treatment efficacy. We also calculated effect sizes for secondary outcome measures of anxiety, depression and metacognitions to the extent these were available. With respect to between-group analyses, we calculated effect sizes for the primary outcome on studies comparing MCT with waitlist control and active treatment control conditions, respectively.

# Eligibility Criteria

The criteria for inclusion of a trial in the meta-analysis were: The study had to (1) evaluate MCT as developed by Adrian Wells, and (2) have a sample size of at least 10 patients with psychological complaints in the MCT condition. In order to include as many trials as possible, we did not place any a priori restrictions on study design, comparison conditions, age of participants, publication type, or statistical presentation of results. We excluded studies that examined specific MCT techniques in isolation (e.g., attention training) as opposed to the treatment as a whole, and studies that combined MCT techniques with other types of therapy, e.g., cognitive therapy. Studies had to be written in English, Danish, Norwegian, Swedish, Dutch, or German in order to be included, as these were the languages at least one of the authors is proficient in.

# Literature Search

As this study is an update of a previous meta-analysis (Normann et al., 2014), trials that were included in the previous metaanalysis were also included in this meta-analysis if they fulfilled our current inclusion criteria. We identified additional studies by searching the databases PsycINFO, PubMed, and the Cochrane Library for the period between January 2014 and January 2018, using the search string "(metacognitive or meta-cognitive) AND (therapy OR trial OR treatment OR psychotherap\* OR intervention)." We also conducted a backward search of the reference lists from articles that met the inclusion criteria. Further, trial registries (www.clinicaltrials.gov; www.isrctn.com) were searched for potential completed trials on MCT. Google Scholar was included as an additional information source. In Google Scholar the search string was limited to articles that used the term "metacognitive therapy" in their title, as the full search string yielded more than 1,000 hits, which is more than the database was able to display. The last search was conducted January 11th, 2018.

# Study Selection and Data Extraction

After removing duplicates, the titles and abstracts of all search hits were screened, and those that did not fulfill our inclusion criteria were excluded. The full text versions of the remaining records were retrieved and assessed for eligibility for inclusion in the meta-analysis. The final list of studies was jointly discussed by both authors.

For each included publication, a list of study characteristics was extracted: type of psychological complaint treated, comparison conditions, sample size, attrition rates at the end of therapy, gender distribution, mean age, comorbidity rates, number of therapy sessions, intervention format (individual or group), follow-up period(s) from the end of therapy, statistical analyses (completer or intent-to-treat), and treatment fidelity checks. We also extracted information for the effect size calculation, namely means and standard deviations for the primary and secondary outcome measures at pretreatment, post-treatment and the longest reported follow-up period available. Data from intent-to-treat samples was used to the extent possible. The primary outcome measure was chosen based on which measure had the most specific relevance for the psychological complaint in question. If a study did not provide sufficient data for performing the meta-analysis, or if central study characteristics were lacking, the information was requested from the authors of the study in question via e-mail. Study selection and extraction of study characteristics was performed by the first author, in consultation with the second author.

# Risk of Bias Assessment

We assessed the quality of reporting of each included study with the Risk of Bias tool developed by the Cochrane Collaboration (Higgins et al., 2011). The risk of bias of the individual studies was examined across six domains: random sequence generation, allocation concealment, blinding of outcome assessment, incomplete outcome data, selective reporting, and other sources of bias. Each domain was assigned a judgement of low, unclear, or high risk of bias. An "unclear" judgement was given if the reporting of what happened in the study was not described in suffice detail to allow judgement of either low or high risk of bias. We did not assess the blinding of participants and personnel (performance bias), as it is not feasible to blind therapists and clients to a psychotherapeutic intervention. In order to ensure consistency in the judgements across all studies, we chose to reassess the risk of bias for the studies from the previous meta-analysis, as those judgements were based on three raters. Four studies were consensuscoded by both authors, and interrater reliability was established for the remaining studies, which were all double-coded. The intraclass correlation coefficient using a two-way random effects model (absolute agreement; single measurement) was 0.84, 95% CI [0.78–0.89], indicating very good reliability. Subsequently, discrepancies in the codes were handled through discussion, until consensus was reached.

# Statistical Analyses Plan

The meta-analysis was carried out using the software program Comprehensive Meta-Analysis (version 3.3; Borenstein et al., 2009). Due to the heterogeneity of the included studies, we expected there to be natural variations in the distribution of the true effect sizes, and therefore a random effects model was used (Borenstein et al., 2009). Hedges' g was chosen as the effect size metric throughout the meta-analysis. Similarly to Cohen's d, it is based on the standardized mean difference, but it applies a correction factor to obtain an unbiased estimate in small samples (Borenstein et al., 2009). Values of 0.2, 0.5, and 0.8 can be conservatively interpreted as small, medium, and large magnitudes of effect, respectively (Cohen, 1988). The correlation between the measures at pre- and post-treatment and pretreatment and follow-up was needed for the effect size calculation, and this was not provided in the studies. As recommended by Morris and DeShon (2002), we retrieved these correlations from authors in a subset of the studies, and conservatively estimated the correlation to r = 0.50, which corresponded to the upper limits of the confidence intervals of the aggregate correlations in the subset of studies.

As mentioned above, we first computed within-group effect sizes (i.e., change from pre- to post-treatment or follow-up) for the primary outcome of all included studies at post-treatment and follow-up. We also calculated effect sizes for secondary outcome measures of anxiety, depression and metacognitions, to the extent these were available. With regard to between-group analyses, we calculated effect sizes for the primary outcome on studies comparing MCT with waitlist control and active treatment control conditions, respectively.

In order to measure variability in the study outcomes, we used the I 2 statistic, which describes the percentage variation across studies that is due to heterogeneity rather than chance. It has been suggested that I 2 values of 25, 50, and 75% may be interpreted as referring to low, moderate, and high levels of heterogeneity (Higgins et al., 2003). We further performed subgroup analyses for subgroups of at least four trials, as has been recommended (Fu et al., 2011). Due to the low number of trials included in these analyses, the p-value for statistical significance was set to 0.1.

We also assessed potential publication bias using visual inspection of funnel plots of the primary outcome measures. In accordance with recommendations from Sterne et al. (2011), publication bias was assessed if there was a minimum of 10 studies available. Particularly, we were interested in examining whether there was an asymmetry in the plot with smaller studies having larger effect sizes, which is indicative of publication bias (Sterne et al., 2011). The trim-and-fill procedure by Duval and Tweedie (2000) was used to calculate the likely number of missing studies and estimate an effect size that corrects for publication bias.

# RESULTS

# Search Results and Study Selection

**Figure 1** displays a PRISMA (Moher et al., 2009) flow diagram of the study selection process. A total of 1536 records were identified, and 25 trials were eligible for inclusion in the final meta-analysis. Eleven of the eligible studies were from the 2014 search (of which nine were included in the analyses), whereas the additional 14 studies were identified through the updated search. Two of these (Wells et al., 2015; Nordahl et al., 2018) were peer-reviewed publications of trials that were included in the first meta-analysis, which at that time only were available as dissertations. With the exception of one study (Esbjørn et al., 2018), all other studies that fulfilled our inclusion criteria were conducted with adult populations. Our results therefore focus on the efficacy of MCT for adults only.

# Study Characteristics

**Table 1** provides an overview of the studies included in the metaanalysis, along with their study characteristics. We included 25 trials based on 26 records. Of these, 10 studies compared MCT with active control conditions, 9 compared MCT with waitlist control conditions, and 10 were uncontrolled trials. Eight of the active control conditions were cognitive and/or behavioral interventions. These included generic and disorder-specific cognitive behavioral therapy (k = 5), behavioral activation

(k = 1), applied relaxation (k = 1), and prolonged exposure (k = 1). Other interventions included mindfulness-based stress reduction (k = 1) and Masters-Johnson sex therapy (k = 1).

The trials were conducted in the United Kingdom (k = 10), Norway (k = 6), Iran (k = 4), the Netherlands (k = 3), Australia (k = 1) and New Zealand (k = 1). One study was a PhD dissertation (Wenn, 2017), one was an unpublished manuscript (Shareh and Dolatshahi, 2012), and the remaining 24 records were articles published in peer-reviewed journals.

# Patient Characteristics

A large proportion of the identified studies treated patients suffering from anxiety and depression (see **Table 1**). There were eight trials on depressive disorders. Of these, seven were on major depressive disorder, whereas one study also included a small proportion of patients with bipolar II and bipolar not-otherwise-specified (Jordan et al., 2014). Five trials were conducted on generalized anxiety disorder, three were conducted on post-traumatic stress disorder, and three were conducted on transdiagnostic samples with anxiety and/or depression. The remaining six trials were on cancer distress, schizophrenia spectrum disorders, body dysmorphic disorder, hyposexual desire disorder, obsessive-compulsive disorder, and grief.

The majority of studies included participants that fulfilled criteria for a psychological disorder according to either DSM-IV-TR (American Psychiatric Association, 2000) or ICD-10 (World Health Organization, 1992) criteria. Two studies (Fisher et al., 2015; Capobianco et al., 2018) did not use structured psychiatric interviews, but rather included patients with elevated levels of anxiety and/or depression based on a cut-off score from a selfreport. One study (Wenn, 2017) assessed diagnostic criteria, but also included participants who did not meet the criteria in question. With few exceptions, the reported comorbidity rates were high. Four studies did not report comorbidity rates, and one study (Ramezani et al., 2017) excluded patients with comorbid disorders. For the 17 studies that reported comorbidity rates in percentages, the mean for the MCT conditions was 65% (standard deviation 24, range 0–100%). These were primarily Axis I disorders consisting of anxiety and depressive disorders. Few studies also reported relatively low rates of substance abuse and eating disorders. Further, inclusion of patients with certain Axis II disorders was reported in seven studies, with reported rates ranging between 8.3 and 50%. Seven studies specified that they worked with refractory cases (ranging from 25 to 100% of participants) that had not previously responded to other forms of psychotherapy. One study was conducted on an inpatient group (Johnson et al., 2017), whereas the remaining studies were conducted in outpatient settings. All studies used adult samples, with the exception of one study that also included adolescents from age 16 and up (Rabiei et al., 2012).

Altogether, 780 patients were included in the meta-analysis. Of these, 468 were offered MCT and meta-analyzed at posttest. In the post-test comparisons with waitlist controls, 208 patients were in the MCT condition and 125 were in the control condition. Data from control patients that received treatment after their waiting period was included and thus meta-analyzed twice in separate groups. In the post-test comparison with active treatment controls, 234 patients were in the MCT condition and 232 were in the control condition. The mean number of participants included in each trial was 31.2 (standard deviation 27.3, range 10–126).

# Metacognitive Therapy

Individual therapy was applied in 18 of the trials, whereas a group format was applied in seven studies. The vast majority of studies (n = 18) followed a published disorder-specific treatment manual for the primary disorder, whereas four studies followed the generic model of intervention as presented by Wells (2009). Three studies (Rabiei et al., 2012; Morrison et al., 2014; Ramezani et al., 2017) investigated a psychological disorder for which no formal manual had yet been developed. In these cases, the authors had adapted a treatment manual for another disorder to the disorder in question. Number of mean therapy sessions ranged from 6 to 14, with an overall mean of 9.5 (standard deviation 2.3) across all 25 studies. Group sessions tended to last between 90 and 120 min, whereas individual sessions usually were between 45 and 60 min. Typically, MCT was conducted weekly. However, some trials chose to intensify treatment in the beginning of therapy, and others chose to prolong treatment as to allow for incorporation of techniques into everyday life. For example, the study on schizophrenia delivered 12 sessions over approximately 9 months (Morrison et al., 2014). With regard to treatment fidelity, the majority (n = 19) of studies reported that continued supervision was provided from experts in order to ensure adherence to the treatment protocols. However, only six studies had assessed treatment fidelity in a formal manner, i.e., with checklists and video or audio recordings, and conclude that therapists adequately adhered to the protocols.

# Outcome Measures

The primary outcome measure of each study is presented in **Table 1**. With regard to the secondary outcome measures, in the majority of cases the Beck Anxiety Inventory (Beck et al., 1988) and Beck Depression Inventory(-II) (Beck et al., 1961, 1996) were used as measures of anxiety and depression symptoms, respectively. With respect to measures of metacognition, the Metacognitions Questionnaire(-30) (Cartwright-Hatton and Wells, 1997; Wells and Cartwright-Hatton, 2004), the Positive Beliefs about Rumination Scale (Papageorgiou and Wells, 2001) and the Negative Beliefs about Rumination Scale (Papageorgiou et al., 2003) were mostly used. Thirteen of the publications reported on the efficacy of treatment on positive and negative metacognitions separately, whereas five publications reported on the efficacy of MCT on metacognitions in general, without distinguishing between positive and negative metacognitions. We conducted the analyses accordingly. The measures included in each of the secondary analyses are listed in **Table 3**.

# Follow-Up

Out of the 25 studies, 22 had follow-up data that was included in our analyses. The mean length of the included follow-up periods was 8.2 months from post-treatment (standard deviation 5.9, range 3–24 months). As displayed in **Figure 1**, we excluded one publication (van der Heiden and Melchior, 2014), which was a

#### TABLE 1 | Study characteristics.


(Continued)


Percent attrition is at post-treatment. Follow-up months indicates the longest follow-up period from post-treatment, and parenthesis indicates that the follow-up was not used in the analyses. Means are given for number of therapy sessions, and if means are not available, the maximum number of sessions allowed is stated. N analyzed refers to number of participants that data was available for. <sup>a</sup>Follow-up analyses did not use ITT. <sup>b</sup>Refers to the total sample, as data was not available for each group. <sup>c</sup>Comorbid anxiety disorders. <sup>d</sup>Median number of sessions. <sup>e</sup>8 analyzed for primary outcome, 10 for secondary outcomes. <sup>f</sup> MCQ data was based on completers. AR, applied relaxation; BA, behavioral activation; BAI, Beck Anxiety Inventory; BDD-YBOCS, Yale-Brown Obsessive-Compulsive Scale Modified for Body Dysmorphic Disorder; BDI, Beck Depression Inventory; CBT, cognitive behavior therapy; Compl, completer analysis; FSFI, Female Sexual Function Index; HADS, Hospital Anxiety and Depression Scale; IES, Impact of Events Scale; ITT, intention-to-treat analysis; (ITT), no attrition, thus equivalent to intention-to-treat analysis; IUT, intolerance-of-uncertainty therapy; MBSR, mindfulness based stress reduction; MDD, major depressive disorder; MJST , Masters-Johnson Sex Therapy; NI, No information; PANSS, Positive and Negative Syndromes Scale; PDS, Post-traumatic Stress Diagnostic Scale; PE, prolonged exposure; PG13, Prolonged Grief Disorder Scale; PSW Q, Penn State Worry Questionnaire; QUIDS16-C, Quick Inventory of Depressive Symptomatology-Clinician rating; STAI-T, State-Trait Anxiety Inventory—Trait scale; WL, waitlist; Y-BOCS, Yale-Brown Obsessive-Compulsive Scale.

30-month follow-up of an included trial (van der Heiden et al., 2012), as the publication reported on 34 out of the original 126 participants. We further chose not to include the follow-up data in another study (van der Heiden et al., 2013), as the authors had not included the data in their primary analysis due to a large dropout rate.

# Risk of Bias

**Table 2** presents the risk of bias of each included study. Overall, the most prevalent rating given was low risk of bias. However, unclear risks of bias were present with regard to allocation concealment, as only 5 out of the 15 controlled trials had described in adequate detail how the randomization schedule was concealed, so that participants and assessors could not foresee which treatment they were allocated to. Separating the randomization from the recruitment process is essential for ensuring that researchers or assessors do not influence assignment of potential participants to treatment arms. Of the 25 studies, 18 had an unclear risk of selective reporting, as they did not report whether they had published a study protocol for the study. We found high risks of attrition bias in five studies, where intent-to-treat analyses were not applied. Furthermore, we found high risks of detection bias in two studies, as they had not blinded the outcome assessor for the primary outcome measure at posttreatment. Altogether, the risk of bias was rated as low in 73% of cases, unclear in 23% of the cases, and high in 4% of the cases. Furthermore, the trials did not differ substantially on risk of bias and thus this variable could not be included in subanalyses.

# Treatment Effects

### Within-Group Effect Sizes

**Figure 2** displays a forest plot of the effect sizes from preto post-treatment on the primary outcome measures across all 25 included studies. The pooled pre- to post-treatment effect size was large, g = 1.72, 95% CI [1.44–2.00], p < 0.001, and this effect was maintained over time, as evidenced by the large pretreatment to follow-up effect size, g = 1.57, 95% CI [1.26–1.87], p < 0.001, k = 22. Subanalyses revealed that MCT also resulted in large and significant reductions of secondary outcome measures that included anxiety, depression, and dysfunctional metacognitions (see **Table 3**). The pooled effect sizes on the primary outcome measures and for measures of anxiety, depression, and metacognitions are displayed in **Table 3**.

### Between-Group Effect Sizes

**Figures 3A,B** display the pre- to post-treatment effect sizes and forest plots for MCT compared with waitlist and active control conditions for the primary outcome measures. A large pre- to post-treatment effect size was found for the studies comparing MCT to waitlist controls on primary outcome measures, g = 2.06, 95% CI [1.52–2.60], k = 9. Only two studies assessed the efficacy of MCT as compared to the waitlist at follow-up, therefore no meta-analytic synthesis was conducted.

Comparison of MCT with active control conditions revealed a medium to large effect size in favor of MCT, g = 0.68, 95% CI [0.41–0.95], k = 10. This comparison at follow-up revealed a small to medium effect size favoring MCT, g = 0.39, 95% CI [0.15–0.63], k = 9. Given that eight out of ten active control conditions were cognitive and behavioral interventions (see **Table 1**), we also focused on the comparison of MCT to these interventions. Compared to behavioral activation and cognitive behavior therapy (CBT), a medium to large effect size was found favoring MCT, g = 0.69, 95% CI [0.36–1.03], k = 8. Compared to behavioral activation and CBT at follow-up, a small to medium effect size was found favoring MCT, g = 0.37, 95% CI [0.07–0.66], k = 7.

# Heterogeneity

For the pre- to post-treatment within-group effect size on primary outcome measures I <sup>2</sup> was 69.74%, Q = 79.84, p = < 0.001, indicating a high degree of variability in the study outcomes. Similarly, high heterogeneity was observed at followup (I <sup>2</sup> = 74.63, Q = 82.79, p = < 0.001). For the comparison of MCT with waitlist and cognitive behavioral interventions, heterogeneity values were also large, I <sup>2</sup> = 71.84%, Q = 28.41, p = <0.001 and I <sup>2</sup> = 59.09%, Q = 17.11, p = < 0.001, respectively. We explored the possible sources of heterogeneity by undertaking subgroup analyses, given that at least four trials could be included in the category of interest.

# Subgroup Analyses

The subgroup analyses were undertaken using within-group effect sizes, as no relevant subanalyses could be conducted on between group effect sizes. This was the result of a low number of trials in the categories of interest. For example, none of the disorders were investigated in four or more randomized controlled trials that compared the efficacy of MCT to waitlist or an active control condition.

One of the trials produced an effect size of g = 6.14 from pre- to post-treatment (Dammen et al., 2015), which may be considered as an outlier from the pooled mean effect size of g = 1.72. When this study was excluded, the pre- to posttreatment and pretreatment to follow-up effect sizes did not change substantially (g = 1.66, 95% CI [1.40–1.91] and g = 1.51, 95% CI [1.22–1.80], respectively).

With respect to the efficacy of MCT for specific psychological complaints, only two disorders were investigated in four or more trials and thus enabled subanalyses. For trials with patients with depression, a large within-group effect size was obtained, g = 2.68, 95% CI [1.85–3.51], k = 8. A large effect size was also produced when only the trials with patients with GAD were analyzed at post-treatment, g = 1.61, 95% CI [1.23–1.98], k = 5.

Because 13 of the trials were co-authored by the originator of MCT (Adrian Wells), the possibility of allegiance bias was examined by comparing the results of these studies with the remaining studies. Both groups of publications revealed large effect sizes, with g =1.98, 95% CI [1.52–2.44] for studies by Wells and colleagues and g = 1.49, 95% CI [1.17–1.81] for studies by independent authors. The results indicated that the studies conducted by Wells and colleagues produced significantly higher effect sizes (p = 0.09). However, when the above mentioned potential outlier (Dammen et al., 2015) was removed, the effect size of the trials conducted by Wells and colleagues was reduced

#### TABLE 2 | Risk of bias.


to g = 1.84, 95% CI [1.43–2.24], k = 12, and the difference between the groups was no longer significant (p = 0.19).

Studies that had applied intent-to-treat analyses, including those without dropouts displayed a significantly lower effect (g = 1.60, 95% CI [1.32–1.89], k = 20) as opposed to those that based their results on completer analyses (g = 2.50, 95% CI [1.52-3.49], k = 5), p = 0.08.

With regard to treatment format, we found that studies that had applied an individual treatment format had a significantly lower effect size (g = 1.57, 95% CI [1.30-1.84], k = 18) than the trials with a group format (g = 2.45, 95% CI [1.59– 3.30], k = 7), p = 0.06. However, when the above mentioned potential outlier (Dammen et al., 2015) was removed, the effect size of the trials applying a group format was reduced to g = 2.09, 95% CI [1.34–2.84], k = 6, and the difference between the groups was no longer significant (p = 0.20). Finally, meta-regressions indicated that pre- to post-treatment changes in positive or negative metacognitions did not significantly explain heterogeneity (Q = 0.91, p = 0.34 for positive metacognitions and Q = 0.59, p = 0.44 for negative metacognitions).

# Publication Bias

Inspection of the funnel plot depicting the within-group preto post-treatment effect sizes for the primary outcome measures revealed an asymmetry indicative of potential publication bias, as the direction of the effect of the smaller trials was toward the right side of the plot, i.e., toward higher effect sizes. Duval and Tweedie's (2000) trim-and-fill procedure identified six studies to be missing, and the produced imputed point estimate resulting from the analysis was g = 1.49, 95% CI [1.19–1.79]. Accordingly, the effect size was still large. A pattern of asymmetry was also observed when trials comparing MCT to active control conditions at pre- to post-treatment were examined. Here, Duval and Tweedie's (2000) trim-and-fill procedure identified three studies to be missing, and the produced imputed point estimate resulting from the analysis was g = 0.53, 95% CI [0.24; 0.82].

TABLE 3 | Pre- to post-treatment and pretreatment to follow-up effect sizes.


dysmorphic disorder; GAD, generalized anxiety disorder; OCD, obsessive-compulsive disorder; PTSD, post-traumatic stress disorder.

k, number of studies included in the analysis; n.a., not applicable, as number of trials too small to conduct analysis; MC, metacognitions. Effect sizes were based on the following questionnaires: For anxiety: Beck Anxiety Inventory; Depression Anxiety Stress Scales-Anxiety; Hospital Anxiety and Depression Scale-Anxiety; State-Trait Anxiety Inventory-Trait. For depression: Beck Depression Inventory; Beck Depression Inventory-II; Depression Anxiety Stress Scales-Depression; Hospital Anxiety and Depression Scale-Depression. For positive metacognitions: Cognitive Attentional Syndrome-1 (positive belief items); Metacognitive Questionnaire (MCQ) or MCQ-30 (positive beliefs subscale); Positive Beliefs about Rumination Scale. For negative metacognitions: Cognitive Attentional Syndrome-1 (negative belief items); Metacognitive Questionnaire (MCQ) or MCQ-30 (negative beliefs about uncontrollability and danger subscale); Negative Beliefs about Rumination Scale. For general metacognitions: Anxious Thoughts Inventory-Meta Worry; Thought Control Questionnaire-Worry; Thought Fusion Inventory.

Given that fewer than 10 trials compared MCT to a waitlist, publication bias could not be assessed in this regard.

# DISCUSSION

In this meta-analysis, we set out to investigate whether MCT improves symptoms of psychological complaints on primary and secondary outcome variables in comparison to control conditions. We were able to assess the efficacy of 25 trials on MCT for a variety of psychological complaints, altogether examining 780 adult patients. Due to the relatively low number of studies, we computed both within- and between-group effect sizes. Our results indicate that MCT is effective in alleviating psychological symptomatology as well as maladaptive metacognitions. The results further suggest that MCT is superior to waitlist and active treatment control conditions.

We were able to include 16 trials that were not part of the first meta-analysis on the efficacy of MCT (Normann et al., 2014). In contrast to the first meta-analysis, we included studies on a variety of other psychological complaints, rather than on anxiety and depression only. Despite the stricter inclusion criteria for trials (i.e., a minimum of 10 participants instead of five), the effect sizes found in this meta-analytic update were largely comparable to those found previously (Normann et al., 2014). The withingroup analyses yielded overall large effects from pre- to posttreatment across the included trials (g = 1.72), and these were maintained at follow-up (g = 1.57). Similarly, in nine trials MCT was compared to waitlist control conditions, and large effects

were found in favor of MCT (g = 2.06). Compared to the last meta-analysis, these results have the advantage of being based on a larger number of studies in each of these groups.

In 10 trials, the efficacy of MCT was compared to a range of other psychotherapeutic interventions. One strength of the included control conditions is that they were evidence-based treatments for the respective disorders. We found that MCT resulted in significantly higher symptom reduction on the primary outcome measures as compared to other therapies, with a medium to large effect size at post-treatment and a small to medium effect size at follow-up. Eight out of 10 of the comparison conditions were forms of cognitive and/or behavioral interventions. When comparing the CBT conditions with MCT, MCT also outperformed CBT at post-treatment and follow-up with a medium to large (g = 0.69) and a small to medium (g = 0.37) effect size, respectively. This is a slightly lower difference in effect than that previously reported, which was based on five trials and resulted in a large pre- to posttreatment effect size in favor of MCT (g = 0.97) (Normann et al., 2014). Although our results indicate that the effect of MCT was significantly higher than in the active control conditions, this is a finding that needs to be interpreted with caution. The number of studies included in these analyses was low and there were variations in the findings across the studies. Furthermore, the difference between MCT and other types of therapy was not as large at follow-up as at post-treatment. This is reflected in the lower bounds of the 95% confidence intervals, which were close to zero in the follow-up comparison. Thus, additional randomized controlled trials with larger sample sizes are needed in order to draw firm conclusions on whether there are differences in treatment effects between MCT and CBT interventions. Furthermore, future research should investigate whether MCT and CBT work differently for different groups of patients with psychological complaints.

MCT was applied to a large variety of psychological complaints. The vast majority of trials, however, targeted anxiety or depression, including post-traumatic stress disorder, as their primary outcome. Accordingly, our results primarily indicate that MCT is effective for alleviating anxiety and depression. The effect of MCT for other psychological complaints, including grief, schizophrenia, body dysmorphic disorder, hyposexual desire disorder, and obsessive compulsive disorder, was only examined in one trial each. With respect to the comparison with CBT, the examined studies exclusively targeted anxiety and depression symptomatology, and therefore they only generalize to this patient group.

We found that MCT not only produced large effects on symptoms related to the targeted problem, but also alleviated secondary, more general symptoms of anxiety and depression. This indicates that MCT also effectively targets comorbid problems of anxiety and depression, which is in line with the theory that MCT targets transdiagnostic processes related to psychopathology (Wells, 2009). This notion relates directly to the finding that MCT produced large changes in metacognitive beliefs and processes at post-treatment and follow-up. In MCT, metacognitions are conceptualized as transdiagnostic beliefs and processes that relate to the development and maintenance of psychological complaints. Visual inspection of the effects for negative metacognitive beliefs indicates that they were larger than for positive metacognitive beliefs. This finding corresponds with the fact that the primary focus in therapy is to challenge negative metacognitive beliefs, as positive metacognitive beliefs about worry and rumination are also prevalent in the general population and less specific to psychopathology (Wells, 2009; Sun et al., 2017). Altogether, these results support the notion that MCT can be effectively applied as a transdiagnostic approach for patients with different psychological disorders. Provided that future empirical data corroborate current results, this would entail great benefits for clinical practice. Effective transdiagnostic approaches enable therapists to more easily conceptualize the common maintaining processes across clinically relevant issues by delivering treatment strategies within the one protocol. This increases not only the efficacy but also the efficiency of treatment as well as the ease of implementation.

Although the pre- to post-treatment effect sizes on the primary outcome measure were within the large range for all trials, the effect produced by the individual studies varied, as indicated by the high degree of heterogeneity. We were able to explore some of the potential reasons for this. One explanation for the heterogeneity is found in the type of statistical analysis used in the studies. As perhaps expected, studies that used completer analyses produced significantly higher effect sizes than those that used intent-to-treat analyses. There was indication that the studies conducted by the originator produced higher effect sizes. It should be noted, however, that the trials conducted by other groups also produced a very large effect size (g = 1.49). More importantly, when the study by Dammen et al. (2015) was removed, there was no significant difference between the groups. We found no difference in effect based on treatment format, when the outlying study was removed, suggesting that MCT is equally effective in individual and group formats. Due to the relatively low number of trials, it was not possible to explore other potential reasons for the heterogeneity.

This meta-analysis has both strengths and limitations. One strength compared to the previous meta-analysis on MCT, which had included three trials on depressive disorders, is that we were able to include eight trials on depressive disorders. This enables us to draw stronger conclusions that MCT is an effective therapy for this group of patients. We were also able to examine the effect of MCT for a larger range of disorders than anxiety and depression, which is of relevance, given that meta-analytic findings suggest that dysfunctional metacognitive beliefs and processes are found across psychological disorders (Sun et al., 2017). Furthermore, we were able to more accurately examine long-term treatment effects of MCT. In the first meta-analysis, only eight out of 16 studies provided sufficient information in order to be included in the follow-up analyses, whereas in this meta-analysis 22 out of 25 studies provided follow-up data. Another strength relates to the samples of the included studies. These were samples that were representative of a clinical population, with high rates of comorbidity and previous treatment attempts. One limitation to the current meta-analysis is that we were not able to conduct secondary analyses with use of controlled effect sizes, due to the low number of studies included. This poses the risk of over-estimating the effect sizes, as withingroup analyses do not account for changes in symptomatology over time that are not related to the intervention. Notably, however, with regard to the primary outcome measure, we did not find indications that the within-group effect size was overestimated, as the pooled effect size for MCT compared to waitlist was also large. This highlights the relevance of incorporating open trials, as they continue to provide valuable information on the efficacy of MCT. A further limitation is that risk of bias was unclear or high in almost one third of the cases, and it remains unknown how this may have affected the meta-analytic results. Lastly, we had limited options in assessing allegiance bias. Although the subanalyses of studies conducted by the originator vs. those by other researchers did not show clear indications of allegiance bias, the author groups may still have favored MCT. One noticeable exception was the study by Nordahl et al. (2018), which had a balanced author group with regard to allegiance, as the originators of both the CBT protocol and MCT protocol took part in the study.

Based on the results of this systematic review and metaanalysis, we encourage that future trials on MCT apply randomized control designs with evidence-based comparison conditions, particularly when investigating anxiety or depression, in order to strengthen conclusions on the efficacy of MCT. Based on our assessment of risk of bias in the studies currently available, we recommend that future studies improve the quality of reporting by clarifying how the allocated treatment was concealed from the participant and investigator up until treatment start, and that they publish study protocols prior to running the trials, in order to minimize the risk of bias. Furthermore, results from this meta-analysis underscore the importance of reporting intentto-treat analyses, in order to not overestimate the treatment effects. Finally, future research needs to examine the efficacy of MCT applied as a transdiagnostic treatment for different clinical populations.

In conclusion, the results of this meta-analysis indicate that MCT is highly effective in reducing symptoms of a range of primary targeted psychological complaints along with symptoms of anxiety, depression, and maladaptive metacognitions. There are preliminary indications that MCT may be more effective than other therapeutic interventions, including cognitive behavioral therapies. However, more studies are needed in order to investigate the accuracy of these preliminary findings.

# AUTHOR CONTRIBUTIONS

NN and NM conceived the study. NN conducted the systematic literature search, screened studies for eligibility, and extracted data from the relevant publications. NM conducted the statistical analyses. NN wrote the first draft of the Introduction, Methods, and Discussion sections of the manuscript. NM wrote the first draft of the Results section and contributed to revisions and modifications of the manuscript. Both authors approved the final version.

# REFERENCES


and meta-analyses: the PRISMA statement. PLoS Med. 6:e1000097. doi: 10.1371/journal.pmed.1000097


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Normann and Morina. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

<sup>∗</sup>References marked with an asterisk indicate studies included in the metaanalysis.

# Metacognitive Therapy of Early Traumatized Patients With Borderline Personality Disorder: A Phase-II Baseline Controlled Trial

Hans M. Nordahl1,2 \* and Adrian Wells3,4

<sup>1</sup> Department of Mental Health, Norwegian University of Science and Technology, Trondheim, Norway, <sup>2</sup> Division of Psychiatry, Nidaros DPS, St. Olav's University Hospital, Trondheim, Norway, <sup>3</sup> School of Psychological Sciences, The University of Manchester, Manchester, United Kingdom, <sup>4</sup> Greater Manchester Mental Health NHS Foundation Trust, Prestwich, United Kingdom

Metacognitive therapy (MCT) is proving to be an effective and brief treatment for

#### Edited by:

Francisco J. Ruiz, Fundación Universitaria Konrad Lorenz, Colombia

#### Reviewed by:

Costas Papageorgiou, Priory Hospital Altrincham, United Kingdom Javier Fernández-Álvarez, Catholic University of the Sacred Heart, Italy

> \*Correspondence: Hans M. Nordahl hans.nordahl@ntnu.no

#### Specialty section:

This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology

> Received: 14 April 2019 Accepted: 04 July 2019 Published: 30 July 2019

#### Citation:

Nordahl HM and Wells A (2019) Metacognitive Therapy of Early Traumatized Patients With Borderline Personality Disorder: A Phase-II Baseline Controlled Trial. Front. Psychol. 10:1694. doi: 10.3389/fpsyg.2019.01694 anxiety disorders and depression, but there are no investigations of its feasibility and effect on primary personality disorders. We conducted a baseline controlled phase II trial of MCT on a group of patients with Borderline personality disorder all reporting early trauma history with sexual or physical abuse. All had been referred to our study after hospitalization and subsequently treated at the university outpatient clinic at NTNU. Twelve patients referred for severe long-term trauma and emotional instability were offered participation in the program. All gave their consent and were included in the trial. We aimed to examine retention over treatment and followup, if the treatment can be delivered in a standardized way across complex and heterogeneous patients and any evidence associated with treatment effects on a range of measures to inform subsequent trials. We measured change in mood, borderlinerelated symptoms, interpersonal problems, trauma symptoms, suicidal thoughts and self-harming behaviors across pre- post-treatment and by 1- and 2-year follow-up. Treatment appeared feasible with all patients completing the course and 11 out of 12 completing all follow-up assessments. All outcome measures showed a high retention rate and no drop-outs from the treatment. Large improvements over time and treatment gains were maintained at 2 years. There was significant reduction of borderline symptom severity, interpersonal problems and trauma symptoms from pre to 2-year follow-up. The results indicate that MCT may be applied to Borderline personality disorder and that future more definitive trials are warranted.

Keywords: borderline personality disorder, early childhood abuse, metacognitive therapy, rumination, selfharming behavior

# INTRODUCTION

Patients with Borderline personality disorder (BPD) may be characterized with instability in affect, behavior and self-esteem. They struggle typically with self-destructive forms of impulsivity and typically report a pattern of life-long unstable and dysfunctional relationships, volatile negative affect consisting of anger and depression with self-harming behaviors and suicidal ideation

(Oldham, 2006). These problems may occur as acute exacerbations leading to injuries and premature death (Black et al., 2004). Borderline personality disorder is associated with many comorbid disorders, typically traumatic stress, drug abuse, dysphoria, or recurrent depression (Zanarini et al., 2004). This group of patients are in need of targeted interventions to deal with intense dysphoric mood, dysfunctional behaviors, and risk.

The current comprehensive approaches to Borderline personality disorder (BPD) are Dialectical Behavior Therapy (DBT; Linehan, 1993), Schema Therapy (ST; Young et al., 2003) and Transference Focused Psychotherapy (TFP; Kernberg et al., 2002), and more recently Mentalization Based Therapy (MBT; Bateman and Fonagy, 2006). A supplementary or adjunct treatment program called STEPPS is also recently in use (Blum et al., 2008). STEPPS consists of psychoeducation and cognitive behavioral approaches in a package consisting of both individual and group based interventions. Most of these treatments may be categorized as integrative, as they use a broad range of strategies that encompass a wide variety of techniques drawn from different approaches (Lieb et al., 2004). There are indications of beneficial effects of comprehensive psychotherapies as well as non-comprehensive psychotherapeutic interventions for BPD, however, the treatments are often long-term and resource demanding with a high relapse rate (Cristea et al., 2017).

There are several common features shared by the widely used comprehensive therapies. All of them emphasize the therapeutic relationship and validation, structure and directedness, with focus on interpersonal difficulties, management of emotional distress and associated self-harming behaviors (SHB) or suicidal risks (Stoffers et al., 2012). There is therefore no surprise that these treatments have generally equal outcomes at post-treatment and by 12 months follow-up on a variety of borderline-relevant domains (Clarkin et al., 2007; Bateman et al., 2015). Based on the similarity in content and equal effect sizes, there is currently no single treatment of choice for BPD.

Metacognitive therapy (MCT; Wells, 2009) is proving to be an effective treatment for anxiety and depression disorders, with emerging evidence it could be more effective than cognitive behavioral approaches (Nordahl et al., 2018; Normann and Morina, 2018). This raises the possibility that it might also be useful in borderline personality disorder patients, who show long term difficulties regulating anxiety and mood.

So far Metacognitive therapy has not been systematically applied with primary personality disorder, but it targets core transdiagnostic processes in psychopathology that should be evaluated in a treatment trial. We therefore adapted the principles of the self-regulatory executive function model (Wells and Matthews, 1996) and metacognitive therapy for anxiety and depression (Wells, 2009) to develop a treatment protocol of BPD.

The resulting protocol offers a brief targeted treatment for patients with BPD or borderline personality spectrum disorder. It consists of several components. First, the preparation and formulation of contracts, shaping of the patient's expectation of therapy and planning of collaboration and the main tasks. Second, a metacognitive-focused modification of self-defeating beliefs and strategies, third targeting executive functions, including de-centered responding to negative thoughts; fourth, involvement of the community psychiatry service in general psychiatric management by the end of therapy and in the following time after the therapy. In addition, where applicable the family and caregivers are involved in order to facilitate family life and support the patient so they will be encouraged and committed to attend the program. In this study we planned to deliver sessions for a maximum of 12 months and the patients were asked to sign a contract that that had been informed and consented to this requirement before entering into the treatment. All patients were offered continuation in general health care management at community health care centers after the 12 months treatment phase was completed.

The goal of the current study was to explore the feasibility, tolerability and preliminary evidence of treatment associated effects of the protocol, also called a phase-II trial. In the current phase-II baseline-controlled trial each patient acted as their own control and we conducted an exploratory assessment of outcomes, which were measured prior to therapy, at posttreatment, at 1- and 2-years after treatment. In this study we were able to examine the feasibility, symptom change at various stages and the long-term effects. The drop-out rate and level of attendance was used as the primary indication of feasibility and tolerability.

Patients with borderline personality, especially those requiring hospitalization are often more complex with diverse and multiple pathologies. As this was our target group we selected a range of measures of outcome, we wanted to see if changes could be observed across more specific but also general measures especially those that assess risk, trauma symptoms, and quality of life.

# MATERIALS AND METHODS

# Design

Patients with early experiences of sexual or violent traumas and BPD were recruited and treated in this open trial. They were all referred to our outpatient clinic for treatment after being hospitalized. All patients received comprehensive treatment according to the protocol and each acted as their own control. To optimize the design of the study we applied a systematic replication design with measures at baseline (4 times over 6 weeks), in which each patient acted as their own control at pretreatment. We assessed again at post-treatment and at follow up by 1 and 2 years. Baseline measures were ratings of depression and anxiety symptoms (BDI-II and BAI). We measured the change in borderline-related symptoms (structured interview of the DSM-IV criteria), interpersonal problems (IIP-64), symptom severity (SCL-90-R, BDI, and BAI), post-traumatic symptom criteria (PDS), the metacognitive beliefs and cognitions (ERIS) and quality of life (WHO-5) once at pre-treatment and at 1- and 2-year follow-up.

# Subjects

Twelve inpatients from local psychiatric hospitals were subsequently referred after hospitalization and treated at

the university outpatient clinic 2007–2012. Of the 12 patients 10 were females (83%) and 2 were males. Mean age was 32.08 (range 19–51). In the group 50% were single, 25% were married or co-habitant and 25% were divorced/separated. Two-thirds of the sample were unemployed or students/trainees of work placement. The rest were in part-time jobs or on disability pension. The mean years of previous treatment including both outpatient treatment and being hospitalized were 7.2 years, and 83% (10/12) were currently treated with psychopharmacology. The distribution of medication was: Neuroleptics (33%), antidepressants (58%), antiepileptic (25%), and benzodiazepines (25%). The patients taking drugs had been stabilized on medication after discharge from the hospital, and they agreed to carry on with the same dosage during the treatment, unless the GP or psychiatrist suggested otherwise.

To be included they had to read and sign a form declaring the rules and time frame of the program. The inclusion criteria were; (1) Borderline personality disorder as primary disorder, (2) age 18 or older, (3) they should consent to the time frame of the treatment program of maximum 12 months and be committed to follow the outpatient treatment on a regular basis. The only exclusion criteria were having somatic illness that needed continuous medical attention, having an active psychosis or substance addiction (alcohol or drugs).

All referred patients accepted these terms after a thorough briefing and signed the contract of participation in the program. None of the referred patients were excluded. Self-harming behaviors and suicidality was assessed by an independent assessor at all time-points of evaluation. This was also monitored through self-report questionnaires during the treatment. This monitoring was a safeguard and an outcome to decide if the patient at any time was self-harming or planning suicide or changed their inclination to act on those thoughts.

The mean number of additional ADIS-IV diagnoses found in this sample was above 4 (M = 4.7). Five patients fulfilled criteria for 6 disorders, and 3 patients fulfilled 5 disorders. Two patients had only 2 disorders, and 2 patients had 3 and 4 disorders, respectively.

The most frequent additional disorders were 12 patients (100%) with anxiety disorder (Social anxiety disorder or Panic disorder/Generalized anxiety disorder), 10 patients (83%) fulfilled the criteria of chronic PTSD, 8 had recurrent depression (67%), and 4 had substance abuse (25%). See **Table 1**.

# Primary and Secondary Measures

The primary outcomes were the drop-out and attendance rates for patients across treatment. The secondary outcomes were specific (borderline-related symptom criteria; SCID-II criteria SCID-II; First et al., 1997/2004) and general symptom severity, impact on processes of worry, rumination and metacognitions, quality of life and risk.

# Measures

The inventory of interpersonal problems (IIP-64), the 64 item version was used to measure various problem areas in interpersonal dysfunction with a five-point Likert-type scale. High scores for the total scale and for its 8 subscales indicate an increased level of interpersonal problems and distress. The internal consistency coefficient (Cronbach's Alpha) and testretest reliability for the original inventory were 0.93 and 0.78, respectively (Horowitz et al., 1988).

The post-traumatic stress diagnostic scale (PDS; Foa, 1996; Foa et al., 1997) was developed and validated to provide a brief but reliable self-report measure of post-traumatic stress disorder (PTSD) for use in both clinical and research settings. The scale is intended to screen for the presence of PTSD in patients who have identified themselves as victims of a traumatic event or to assess symptom severity and functioning in patients already identified as suffering from PTSD. The test is self-administered and requires a reading age of >12 years.

Emotional and relationship instability scale (ERIS; Nordahl and Wells, 2009b), is a rating scale developed to measure borderline-relevant symptoms and beliefs, in particular the patients psychological distress related to abandonment and rejection. Also ERIS measures maladaptive coping behaviors (CAS) and metacognitive beliefs associated with maladaptive coping. It is one of the few self-report measures designed for patients with borderline personality and it uses a Likert scale from 0 to 8 (0 = None of the time, 8 = Every time). In a study of 133 patients with borderline spectrum personality, we found that ERIS possesses a reliability (internal consistency) of ICC = 0.91, with a three-factor latent structure that explained 48% of the

TABLE 1 | Demographic characteristics of the sample (N = 12).


<sup>∗</sup>Other than BPD. GAD, Generalized Anxiety Disorder; MDD, Major Depressive Disorder; PD, Personality Disorder.

scale variance. The psychometric properties of the ERIS were satisfactory (Bruset Ludviksen, 2014).

WHO-5 well-being index, is a brief questionnaire, which consists of 5 questions tapping the subjective well-being of the respondents. The scale is derived from other rating scales, and each of the 5 items is scored from 5 (all of the time) to 0 (none of the time), so the range is from 0 to 25, indicating maximal wellbeing. The WHO-5 is widely used and has adequate predictive validity both as a screening tool and an outcome measure in clinical trials (Topp et al., 2015).

# Procedure

All patients were assessed with ADIS-IV (Di Nardo et al., 1994) and SCID-II (First et al., 1997/2004) by independent clinicians at the outpatient clinic. As well as diagnosis, the severity of their borderline disorder, the inclusion and exclusion criteria were assessed for each patient by independent assessors at the university clinic. The patients were given both oral and written information about the study and implications of participating. All patients gave written consent in order to participate in the study. They then completed four baseline data assessments of anxiety and depression before the treatment was provided, and at the pretreatment stage the full set of measures were administered to each of the subjects. This set of measures was administered again at post-treatment, and by 1- and 2-year follow-up.

# Treatment

The first phase in the protocol was to negotiate a contract and shape the patient's expectation about his/her and the therapist's role in the program. In addition, there was some planning of the collaboration and availability of the therapist and early involvement of the community service. The second and the third phase focused on self-defeating beliefs and the self-regulatory executive functions (Wells and Matthews, 1994) of the patient:

The following steps were implemented:


The formulation (1) was shared with the patient, and the therapist socialized to the treatment. The socialization is necessary to help the patient to understand their distress in a metacognitive framework. The aim of the formulation and socialization is to develop a common understanding of the problems and the basis for the interventions. The selfdefeating factors (2) are conceptualized as a set of unhelpful metacognitive beliefs about control ("I cannot control my mind;" "I cannot stop ruminating"), or change ("my mind is broken" or "the problem is in my genes"). They also include beliefs about the value of negative self-referential thinking ("I need to put myself down in order to feel safe" or "only by punishing myself I can feel okay"). These self-defeating metacognitions can work against any treatment engagement and recovery.

Self-harming behaviors and suicidal threats are also selfdefeating factors and should be addressed specifically in these sessions. We helped the patient be explicit about them, even if it is subjectively shameful and normally avoided. In this early work the therapist helped the patient to develop a sense of responsibility for his/her actions; as this is something the patient chooses to do to handle distress and negative thoughts (i.e., labeled as an unhelpful coping strategy).

An advantage/disadvantage analysis was run in collaboration with the patient, and the therapist explored if there might be more beneficial ways to deal with life that might involve learning how to reduce worry and self-punishment. The therapist and patient made an agreement to stop acting on self-harming beliefs and behaviors and to test alternatives. A new plan and alternative strategies was developed, monitored and followed-up in sessions until the patient had modified the beliefs and behaviors that interfere with the goals of treatment.

Many borderline patients worry about rejection and abandonment in social relationships, and they ruminate about past events (e.g., being ignored) and losses (3) Rumination is a strategy, which involves dwelling on past failures, abandonments, and criticisms (Nolen-Hoeksema, 1991). Rumination has different forms, and the most prominent for borderline patients is depressive and angry rumination. Angry rumination consists of repetitive thoughts about the unfairness of life, where the patient experiences that they are scapegoated, unjustly blamed or not "understood" by others. This creates frustration and anger and feelings of being criticized, attacked or alone. Depressive rumination is about past failures, abandonments and losses and creates a self-critical, self-blaming, and depressed mood. Worry, in contrast consists of anticipating rejection, abandonment or loss of credibility, and this leads to anxiety and fear and the tendency to avoid or sacrifice relationships. Rumination and worry are seen as central maintenance processes in the metacognitive model and thus, an important intervention in MCT is to help the patients to reduce the level of worry and ruminations, as these contribute to anxiety, dysphoria or depressive mood.

Cognitive flexibility (4) is the ability to selectively control the focus of attention and to respond to worry and rumination by moving attention away from inner or external threatening stimuli. In order to be cognitively flexible, the patient must work on postponing responses, and develop greater awareness of choice in whether or not to respond to internal (thoughts, feelings) or external events (e.g., being ignored). The treatment protocol applies the Attention training technique (ATT; Wells, 2009) as this is designed to change the metacognitions that regulate thinking and facilitate emotional processing by interrupting excessive self-focused attention and brooding. Detached Mindfulness (DM; Wells, 2005), is also used as this technique helps the patient to explore and discover their executive control of thinking (5). The essential idea in DM is to leave alone any cognition even when triggered by some distressing

thoughts or feelings. In this way the patient can refrain from perseverative thinking such as rumination, worry or threat monitoring, and can be encouraged to choose to postpone these processes.

Improving the patient's functioning in work and relationships are the primary targets in setting new personal goals (6). The therapist worked with the patient in developing concrete goals in these areas. Even though the acute-phase treatment program finished after 12 months, we took a 2-year perspective of working toward a better life context within these areas. The therapist discussed the following with patients: "Where do you want to be in your life in 2 years from now in terms of: education? substance abuse? relationship to parents? dealing with your mood? contact with friends?" and other relevant domains. The advantages of working toward these goals were explored and a step by step concrete plan was formulated. The implementation of these goals was a recurrent issue during the program, and both the patient and the therapists monitored the progress toward these concrete and within reach goals.

The last phase of the protocol comprised transferring the patient to general community psychiatric management. This is a team consisting of a family therapist, or psychiatric nurse and general practitioners responsible for the support and follow-up of the patients. The main therapists continued to support the team in a role as a clinical supervisor. The main task of the community management team was to follow up the goals and adaptive strategies from treatment and to help and support the patient in the job or school situation. The general psychiatric service continues to follow-up the patient under monthly supervision of the therapist. The patients do not attend these meetings, but are informed by the psychiatric nurse. Normally the general psychiatric management follows the patient from 1 to 2 years after treatment, but this is based on individual needs.

# Therapists and Treatment Integrity

The treatment was conducted by two experienced therapists each of whom have 20 years of clinical practice and training in metacognitive therapy. Overall supervision in MCT was provided by AW and site supervision by HMN. Treatment followed a draft protocol by the authors and adherence and the level of competency was monitored by the Metacognitive therapy competency scale (MCT-CS; Nordahl and Wells, 2009a).

# Statistical Analysis

Primary outcome and secondary outcome measures were subjected to repeated measures ANOVA, with time (pretreatment, post-treatment, 1 and 2-year follow-up) as the repeated measures factor.

We used Hedge's g to estimate the effect sizes between preand post-treatment and between pre-treatment to 2-year followup (Hedge, 1981). Hedge's **g** is attained by subtracting the posttreatment means from the post-waitlist means and dividing this by the pooled standard deviation and correcting for the sample size (N < 20). Missing data was imputed by using unit imputation substituting the missing value by the mean of the observed values for that variable.

# RESULTS

All patients (N = 12) attended 12 months of therapy consisting of up to 40 sessions (range 20–45) and we had a 100% completion rate. There were no drop-outs during the acute treatment phase. No suicidal attempts were reported, although suicidal thoughts and self-injury occurred during treatment but showed a decline in severity (see **Figure 2**). All patients completed the 1 year followup measures, but 11 of 12 filled in the measures at 2-year followup. One patient was lost to 2-year follow-up and did not fill in the questionnaires as we were unable to get in contact with her.

# Feasibility and Retention

The patients reported in interview that they experienced the treatment to be helpful and meaningful to them reporting that the rational for the treatment made sense. The mean number of treatment sessions were M = 26.6 sessions (SD = 6.15) and the time frame was between 9 and 14 months (M = 11.5). There were no dropouts from pre- to post-treatment and there was a high retention rate where all attended between 70 and 90% of the sessions offered. The transition to community psychiatry management seemed to work well in most patients, and 8 of the 12 patients made use of this service after treatment and by 2-years follow-up 4 patients were still in contact with the community management service on a regular basis.

# Baseline Change

A repeated measures ANOVA was used to examine any changes in symptom level during baseline. Neither anxiety nor depression changes were significant during the baseline period [BAI; F(3,15) = 1.140, p = 0.354; BDI-II; F(3,15) = 1.584, p = 0.357], indicating that there were no major or systematic changes in anxiety and mood occurring during the pre-treatment period (see **Figure 1**).

# Outcome of Borderline Related Symptoms

One-way repeated measures ANOVAs were run for each of the outcome variables across pre-treatment, post-treatment and 1 and 2-year follow-up (N = 12). Mauchley's test of spherity was not significant and thus not violated, so adjustments were not needed. However, due to multiple post hoc analyses we used Bonferroni correction in the analysis of the main measures. The results indicated a significant time effect for borderline-related symptom criteria, Wilks' Lambda = 0.085, F(3,30) = 27.991, p < 0.001, η 2 <sup>p</sup> = 0.737, showing a significant reduction. The pairwise comparisons between pre and post-treatment showed a significant reduction (p = 0.001), and from pre to 1-year followup (p < 0.001) and from pre to 2-year follow-up (p = 0.003).

For the changes in depression across time the effects were as follows: Wilks' Lambda = 0.24, F(3,27) = 15.676, p < 0.001, η 2 <sup>p</sup> = 0.635. The pairwise comparisons here were from pre- to post-treatment (p = 0.007), from pre-treatment to 1-year follow-up (p = 0.013), from pre-treatment to 2 year follow-up (p = 0.009). For anxiety the results showed large reductions, Wilks' Lambda = 0.123, F(3,27) = 28.376,

FIGURE 1 | Levels depression and anxiety from baseline to 2-year follow-up.

p < 0.001, η 2 <sup>p</sup> = 0.759. The pairwise comparisons were; pre- to post-treatment (p < 0.001), pre-treatment to 1-year follow-up (p < 0.001), and pre-treatment to 2-year follow-up (p = 0.004).

Significant reductions across time were also observed for other symptom domains such as, interpersonal dysfunction, Wilks' Lambda = 0.104, F(3,27) = 30.247, p < 0.001, η 2 <sup>p</sup> = 0.771.

The pairwise comparisons were; pre-treatment to post-treatment (p < 0.001), pre-treatment to 1-year follow-up (p = 0.001), and pre-treatment to 2-year follow-up (p < 0.001).

Post-traumatic symptoms, Wilks' Lambda = 0.203, F(3,30) = 13.643, (p = 0.004), η 2 <sup>p</sup> = 0.577.

The pairwise comparisons were; pre-treatment to posttreatment (p = 0.001), pre-treatment to 1-year follow-up (p = 0.003), and pre-treatment to 2-year follow-up (p = 0.018).

Level of worry/rumination, Wilks' Lambda = 0.046, F(3,30) = 54.187, p < 0.001, η 2 <sup>p</sup> = 0.844.

The pairwise comparisons were; pre-treatment to posttreatment (p < 0.001), pre-treatment to 1-year follow-up (p < 0.001), and pre-treatment to 2-year follow-up (p < 0.001).

Quality of life (WHO-5), Wilks Lambda = 0.196, F(3,30) = 11.022, (p = 0.001), η 2 <sup>p</sup> = 0.542. The pairwise comparisons were; pre-treatment to post-treatment (p = 0.016), pre-treatment to 1-year follow-up (p = 0.003), and pre-treatment to 2-year follow-up (p = 0.032). The means and standard deviation for the outcome measures across time are presented in **Table 2**.

As the sample size was <20, we used corrected effect size (g) using Hedge's formula (Hedge, 1981). **Table 2** shows that the effect sizes were large, and comparable to other comprehensive treatments for patients with Borderline personality disorder (Giesen-Bloo et al., 2006; Clarkin et al., 2007). For both of the outcome measures the effects sizes (g) were between 1.0 and 1.8, whereas the PTSD symptoms showed moderate change from prepost (g = 0.724) and increased from pre- to 2-year follow-up (g = 1.09). The levels of self-reported worry/rumination about abandonment/rejection had a significant drop and indicates a major change in the thinking styles in all patients. Also, the QoL well-being index showed that patients overall were more satisfied by the end of treatment (g = 1.455) and at follow-up (1.136).

# Levels of Suicidal Thoughts and Self-Harm

The severity of suicidal thoughts/impulses and self-harming behaviors (SHB) was evaluated during the course of treatment by the therapists. The level of reported suicidal thoughts significantly decreased and this improvement held up at 2 year follow-up, Wilks' Lambda = 0.134, F(3,30) = 20.212, p < 0.001, η 2 <sup>p</sup> = 0.669. The pairwise comparisons were; from pre-treatment to post-treatment there was no significant reduction (p = 0.115), but from pre-treatment to 1-year follow-up the reduction was significant (p = 0.001), and from pre-treatment to 2-year follow-up (p = 0.003). The level of self-harming behaviors was also decreasing significantly across time, Wilks' Lambda = 0.380, F(3,30) = 8.625, p = 0.007, η 2 <sup>p</sup> = 0.463. However, the pairwise comparisons showed no significant reduction from pre-treatment to post-treatment (p = 0.173), but significant reductions from pre-treatment to 1-year follow-up (p < 0.040), and from pre-treatment to 2-year follow-up (p = 0.076). See **Figure 2**.

# Psychopharmacology

The patients taking drugs (n = 10) were stabilized on medication after discharge from the hospital, but the users of antidepressant medication and benzodiazepines were tapering their drugs toward the end of the 12 months' treatment program (n = 7).

# DISCUSSION

Feasibility trials are an important first step before applying an intervention to new patient groups as they provide information about tolerability and acceptability of new treatments and indicate whether it should be used or further tested against existing treatment. The results of the current study suggest that this protocol has some important qualities. No patients dropped out in the pre- and post-phase, and only 1 patient was missing at 2-year follow-up. The session attendance rate varied between 70 and 90%, which is highly satisfactory compared to other relevant studies (Landes et al., 2016). Overall most patients were significantly less symptomatic after treatment and upheld the gains during the 1 to 2-year follow-up. The outcome shows large effects sizes and most patients had clinical or subclinical levels of symptoms and functioning at post-treatment. In addition, the interpersonal problems and trauma symptoms and overall well-being showed significant changes, including areas that were not targeted in the intervention. This indicates that metacognitive therapy may be feasible and useful for patients with BPD and early trauma. The protocol applies a coherent and understandable theoretical rational for the treatment, which may be crucial for outcome and for lower attrition or retention (Verheul and Herbrink, 2007).

The preliminary results of our study compare well with other comprehensive treatments in terms of pre- and post-treatment effects and duration (Linehan et al., 2006; Bateman and Fonagy, 2009; Nadort et al., 2009; Doering et al., 2010).

The main target in Transference-focused psychotherapy is the patients' interpersonal dynamics which is manifested in the transference (misattribution of emotional reactions). Doering et al. (2010) conducted an 18 months' study of Transferencefocused psychotherapy for BPD and used numbers of drop-outs and suicide attempts as the main measures. In this study they found that the Transference-focused treatment was significantly better than treatment conducted by experienced community psychotherapists, but on measures of anxiety and depression levels there were no differences. Typically, two sessions per week are delivered. In the study the dropout rates were high (53%) and the assessments of follow-up had a high proportion missing (38%), which lowers reliability of the findings.

In Schema focused therapy the main target is the healing of early maladaptive schemas and modes (schema clusters). Nadort et al. (2009) conducted Schema Focused Therapy (SFT) in a group of patients with BPD in regular mental health care with an addition of a therapist telephone availability (TTA) all day. They treated the patient over the time span of 18 months, and had a recovery rate of 40%, and effect sizes of 1.5. The treatment effects and drop-out were 22% but no effect of the TTA component was found. These results are comparable to our own study in terms of being brief and well controlled with high effect sizes. However, Schema therapy is designed for 18– 36 months of treatment, and involves both individual and group sessions. There are no follow-up studies beyond 12 months, thus the long-term outcome is not known.

In DBT a main emphasis is on the patients' skills acquisition and behavioral shaping. This is conducted in a context


TABLE 2 | Means and standard deviation for the sample (N = 12) and changes from pre-treatment to 2-year follow-up across time!

<sup>∗</sup>N = 11. Hedge's g pre-post (T1-T2); Hedge'g Pre-2 years FU (T1-T4). BPD, Borderline personality disorder; BDI, Beck Depression Inventory; BAI, Beck Anxiety Inventory; IIP-64, Inventory of Interpersonal Problems; PDS, Post-Traumatic Diagnostic Scale; ERIS, Emotional and Relationship Instability Scale; QoL, Quality of Life.

of the dialectic of validation and problem solving. DBT integrates many techniques and is designed to be adapted to a variety of treatment settings and patients. Linehan et al. (2006) compared 12 months DBT with treatment as usual conducted by specialist therapists in BPD and suicidal behaviors, and reported significantly better outcomes of DBT compared to TAU on borderline related symptoms and behaviors. The attrition rate was 25% and 10% lost to 1-year follow-up in the DBT condition. The treatment of DBT is more comprehensive and includes weekly individual therapy with group skills training session, out-of-session paging, and consultation team for the therapist. Thus, DBT has the most intensive and structured scheme of treatment of all the comprehensive therapies.

Mentalization based therapy (MBT) is rooted in attachment theory, theory of mind and psychodynamic principles. The main target is to increase the patient's capacity to mentalize thoughts and emotions under stress, in order to stabilize cognition in settings of social interactions and emotional distress. It is proposed that the problems of mentalization in patients with BPD may be linked to dysfunctional early attachment (Bateman and Fonagy, 2006). In a study of patients with BPD in an outpatient setting it was reported that an 18 months' treatment program combining individual and group sessions showed a large improvement in selfinjurious behavior, suicidal behavior and hospitalization in the MBT group, and significantly better than in the clinical management comparison, which had an emphasis on social problem-solving (Bateman and Fonagy, 2009). The results were good for borderline-related behaviors, reduction of symptomatic distress, reduced use of medications and improved social functions. Approximately 75% completed the trial, but the data on retention rate (attending sessions) was not available (Bateman and Fonagy, 2009).

The main target in MCT is to improve self-regulatory executive functioning and reduce self-defeating processes. This intervention is different from other approaches used in the treatment of BPD. To work more systematically and directly on the attentional processes and executive functions, but also reducing the level of perseverative thinking, such as angry rumination, is unique to the MCT approach. Furthermore, targeting the cognitive attentional syndrome (CAS) and modifying the self-defeating metacognitive beliefs is at the core of MCT in order to achieve more adaptive self-regulation and cognitive flexibility.

The results in the current trial suggest that MCT was associated with good clinical response from pre- to posttreatment on borderline-related symptoms, mood and interpersonal problems. Also we observed an effect on trauma symptoms, which were not directly targeted in the treatment. The gains seem to be maintained at 2-year followup. This relatively brief intervention seems to be feasible for outpatient treatment, and appears to compare favorably with comprehensive treatments of longer duration. For a more comprehensive overview of the outpatient treatments conducted (see **Supplementary Table S1**).

There are some important limitations in the current study. First, the sample size is small, therefore any direct comparison of the results with larger comprehensive studies must be interpreted with caution. Low N trials have a higher risk of imprecise estimates and inflated effect sizes; the results may not therefore be reliable. Second, even though the participants were well monitored during the trial, by 2-year follow-up, one patient was not possible to locate. Thus, the data on that particular patient is missing and was not included in the analysis at 2-year follow-up. Third, there was no standardized format of care after treatment, as this was adapted to the individual client within the general community setting. The resources put into community care management could be different from one site to another, as this was due to local resources and availability of health care workers. Thus, we cannot estimate the degree of influence on the results this variability might have had at 1 and 2-year follow-up. Fourth, the competency level of the therapists was likely increasing during the trial, as it does in most trials, and

this indicates that the adherence and the competency level in the treatment protocol was probably not consistent across the course of the study.

# CONCLUSION

The MCT protocol evaluated here seems to be feasible and well tolerated by patients with BPD and early trauma. It was associated with significant improvement in a 9– 14 months' program and the symptom reductions were maintained over 2 years after treatment. The treatment effects across various domains indicate a trans-symptomatic improvement, which should be explored further. A metacognitive approach in combination with an adapted community service seems very promising and further testing allowing for variability in the length of the program are warranted using larger samples and comparative randomized controlled trial methodology.

# DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

# ETHICS STATEMENT

Ethical approval from the Regional Committees for Medical Research Ethics in Central Norway (REK No. 2011/2528) was obtained for this study.

# REFERENCES


# AUTHOR CONTRIBUTIONS

HN initiated and conducted the trial, did most of the treatments, or supervised the co-therapist, carried out the data analysis, and wrote the main draft of the manuscript. AW acted as a supervisor, advised on design and analysis, and contributed to writing the manuscript. Both authors approved the final version of the manuscript.

# FUNDING

The baseline controlled trial was sponsored by the Norwegian University of Science and Technology (Grant No. ISAK 2010/0236) and the Nidaros DPS, St. Olavs Hospital, Trondheim (Grant No. 09/0123).

# ACKNOWLEDGMENTS

We thank Kristoffer Bruset Ludvigsen, Else Karin Bragstad, Geir Svebakken, Kjell Røsdal, Stian Solem, Britt Buran, Gunnar Morken, and the team members at the Regional Trauma Clinic (REFT) at the Nidaros DPS for their contribution to various parts of the trial. Thanks to Tor Erik Nysæter, Ph.D. for the contribution to **Supplementary Table S1**.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg. 2019.01694/full#supplementary-material

systematic review and meta-analysis. JAMA Psychiatry 74, 319–328. doi: 10. 1001/jamapsychiatry.2016.4287


clinical applications. J. Cons. Clin. Psychol. 56, 885–892. doi: 10.1037/0022- 006X.56.6.885


**Conflict of Interest Statement:** HN and AW have received fees for teaching in MCT and CBT and royalties for books within this subject area.

Copyright © 2019 Nordahl and Wells. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Metacognitive Therapy for Posttraumatic Stress Disorder in Youth: A Feasibility Study

#### Michael Simons\* and Anna-Lena Kursawe

Department of Child and Adolescent Psychiatry, Psychosomatics and Psychotherapy, RWTH Aachen University, Aachen, Germany

Metacognitive therapy (MCT) is an effective treatment for posttraumatic stress disorders (PTSD) in adults. However, there is no evidence for the feasibility, acceptability, and efficacy of MCT for PTSD in youth so far. This study is the first to utilize MCT for children and adolescents with PTSD. Twenty-one children and adolescents (aged 8– 19 years) who were consecutively referred to the outpatient trauma clinic were treated with MCT. In all patients, treatment was well accepted and regularly attended. At posttreatment, MCT was associated with significant and large reductions in posttraumatic stress symptoms. Depending on the outcome measure, 95 or 85% of the patients were classified as recovered after treatment. Eighteen patients were included in the calculation of the overall outcome. Effect sizes on primary PTSD measures were large (Cohen's d = 3.42 and d = 1.92) and more than comparable to well-established treatments. Only six patients were available at follow-up, but their improvements were found to be stable. Despite the limitations of this uncontrolled study, the results suggest that MCT may be a feasible and promising treatment for traumatized children and adolescents and they justify a controlled trial evaluating the efficacy of MCT versus an already well-established intervention.

#### Keywords: adolescents, children, feasibility study, metacognitive therapy, posttraumatic stress disorder

# INTRODUCTION

Many children and adolescents experience traumatic events with the potential to impact their lives substantially. About 16% of youths subjected to a traumatic event develop posttraumatic stress disorder (PTSD) (Alisic et al., 2014). A recent meta-analysis shows that PTSD prevalence reduces by approximately 50% over the first 6 months after a traumatic event, while there is little evidence of further change in prevalence or symptom severity after 6 months. This suggests that it is increasingly unlikely for a child to lose a PTSD diagnosis without intervention beyond this point (Hiller et al., 2016). In the Diagnostic and Statistical Manual of Mental Disorders (4th ed., DSM-IV-TR; American Psychiatric Association [APA], 2000), PTSD is conceptualized as symptom clusters of intrusive re-experiencing, pervasive avoidance, and hyperarousal in the aftermath of at least one traumatic event. In DSM-5 (American Psychiatric Association [APA], 2013), a fourth symptom cluster has incorporated negative alterations in cognitions and mood.

A recent meta-analysis (Morina et al., 2016) and a recent review (Dorsey et al., 2017) showed that trauma-focused treatment, especially trauma-focused cognitive-behavioral therapy (Tf-CBT), was most effective when treating PTSD in children and adolescents. In comparison to a waitlist condition, Tf-CBT was superior (Hedges's g = 1.44). Tf-CBT usually includes approximately 10–12 parallel, mostly separate child and parent sessions. A recent German multicenter study found a modest effect size (Cohen's d = 0.50) of Tf-CBT (12 weekly 90-min parallel or conjoint sessions with

#### Edited by:

Hans M. Nordahl, Norwegian University of Science and Technology, Norway

# Reviewed by:

Odin Hjemdal, Norwegian University of Science and Technology, Norway Stian Solem, Norwegian University of Science and Technology, Norway Costas Papageorgiou, Priory Hospital Altrincham, United Kingdom

# \*Correspondence:

Michael Simons msimons@ukaachen.de

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 19 September 2018 Accepted: 28 January 2019 Published: 19 February 2019

#### Citation:

Simons M and Kursawe A-L (2019) Metacognitive Therapy for Posttraumatic Stress Disorder in Youth: A Feasibility Study. Front. Psychol. 10:264. doi: 10.3389/fpsyg.2019.00264

**194**

patients and caregivers) against a waitlist condition and a large within-group effect size of Tf-CBT (d = 1.51) (Goldbeck et al., 2016). The core element of this treatment is imaginal exposure, i.e., helping the patient recall traumatic events in detail and re-experiencing it. This kind of exposure is aimed at counteracting the patient's avoidance of distressing thoughts or trauma reminders to reduce anxiety by habituation. Further, as conceptualized by cognitive theorists (e.g., Ehlers and Clark, 2000), this strategy aims to integrate fragmented trauma memories into autobiographical memory. Surprisingly, explicit exposure does not moderate outcomes for posttraumatic stress or depressive symptoms (Dorsey et al., 2017).

Moreover, and in line with the new conceptualization of PTSD in DSM-5, cognitive changes are important targets in CBT. Thus, besides reducing cognitive and behavioral avoidance, Tf-CBT aims to correct dysfunctional cognitions about the self, the world, and the future that patients have developed after the traumatic event. In a recent meta-analysis, Diehle et al. (2014) show that Tf-CBT leads to a larger reduction in posttraumatic stress symptoms and trauma-related cognitions than non-active and active control conditions. Exposure therapy seems to be more efficacious than cognitive interventions without exposure, while cognitive restructuring has small advantages over treatments without cognitive restructuring.

Although current practice parameters recommend traumafocused treatments, a recent study by Wells et al. (2015) challenges this recommendation. The authors found a new, non-trauma-focused treatment approach, namely metacognitive therapy (MCT), to be superior to established trauma-focused treatments (i.e., prolonged exposure, PE) when treating PTSD in adults (Hedges's g = 4.52 for MCT vs. g = 1.53 for PE). According to the originator of MCT (Wells, 2009), cognitions (e.g., memory structure), and general beliefs are less crucial for the development of PTSD than cognitive processes such as thought suppression, rumination, worrying, and gap filling (i.e., trying to fill in gaps in the memory). Together with dysfunctional attentional and avoidant coping strategies, these thinking processes make up the so-called cognitive attentional syndrome (CAS). These maladaptive processes are motivated by metacognitive beliefs, i.e., beliefs about thinking. Positive metacognitive beliefs motivate these processes, e.g., "I have to get rid of these thoughts in order to stop me from going mad," "Worrying keeps me safe," "I have to think about the event in order to find out what I could have done to prevent it from happening," or "In order to cope with the event I have to remember it in every detail." Negative metacognitive beliefs refer to the uncontrollability and dangerousness of thinking, e.g., "I cannot stop worrying" or "I will go crazy if I cannot stop thinking about the event." Consistent with this model, Fergus and Bardeen (2017) found evidence that metacognitive beliefs, not cognitive beliefs, maintain posttraumatic stress. Further, Bennett and Wells (2010) found evidence that metacognitive beliefs about the trauma memory (e.g., the belief that gaps in the memory mean I am not normal), but not memory disorganization within the trauma narrative, positively predicted significant variance in posttraumatic stress symptoms. Further, Bardeen and Fergus (2018) found that deficits in executive control strengthened the positive association between metacognitive beliefs and posttraumatic stress symptoms. Wells (2009) emphasizes that the processes of the CAS are maladaptive in that they increase and maintain threat perceptions and block emotional processing. Thus, MCT aims to reduce the CAS and to modify the metacognitive beliefs which maintain it.

There is some evidence that MCT is applicable and might be efficacious in youths with obsessive-compulsive disorder (Simons et al., 2006) and with generalized anxiety disorder (Esbjørn et al., 2018). The present study describes the first attempt to treat traumatized minors with MCT with the aim to test the applicability and the feasibility of MCT for this age group. The main research question was to determine whether the established MCT manual which was developed for adults with the diagnosis of PTSD could be applied to traumatized youths. Therefore, we aimed to determine the number of patients who completed therapy regularly, the number of sessions required, and the magnitude of symptom reduction as an indication of the possible efficacy of this treatment.

# MATERIALS AND METHODS

# Participants

Twenty-one children and adolescents (age 8–19 years), who presented consecutively in the outpatient trauma clinic and who met criteria for PTSD according to DSM-IV/ICD-10 (World Health Organization [WHO], 1992; American Psychiatric Association [APA], 2000), were included in this study (see **Table 1**). Traumatic events comprised violent or sexual assaults, robbery, suicide of a relative, house fire, or car accident. The interval between trauma and commencement of treatment ranged from one to more than 48 months. Two of the three cases with greater than a 48-month-interval were female adolescents with long-term experiences of repeated sexual abuse and assault. One girl suffered from the aftermath of a sexual abuse event 11 years prior. In every case, the first diagnostic appointment followed no later than one week after the families' request. Therapy began shortly after the completion of the initial assessment. The wide interval between the traumatic incident and the beginning of treatment was due to the families' decision of when to access the outpatient clinic.

The diagnosis of PTSD was confirmed by means of a well-established structured interview (see below). Comorbid diagnoses/problems included attention deficit hyperactivity disorder (n = 2), depression (n = 4), obesity (n = 1), selfharming behavior (n = 1), and generalized anxiety disorder (n = 1). None of the patients included had comorbid problems of alcohol/drug dependency and none had previously obtained any cognitive-behavioral treatment or received pharmacotherapy. All participants provided informed consent and the study was approved by the ethics committee at the RWTH Aachen Faculty of Medicine (EK 240/18).

# Measures

The Clinician-Administered PTSD Scale for Children and Adolescents (CAPS-CA; German version: Steil and Füchsel, 2006)


TABLE 1 | Age, sex, type of trauma, duration (i.e., time between the traumatic event and the first appointment in the outpatient trauma clinic), and comorbidity.

F, Female; M, Male; ADHD, Attention Deficit/Hyperactivity Disorder; GAD, Generalized Anxiety Disorder.

is a semi-structured clinical interview designed to assess PTSD symptoms according to DSM-IV and ICD-10 and associated symptoms in children and adolescents. It consists of 36 questions based on a specific event the child identifies as most distressing. The diagnosis also incorporates a clinical judgment regarding the type of trauma and impact on functioning. The CAPS-CA was administered only before treatment to confirm the diagnosis of PTSD.

Primary outcome is feasibility, that is the proportion of patients who were offered treatment who completed and the number of sessions attended. Secondary outcome was the change in posttraumatic symptoms, self-rated by the patients. These were measured at pre- and post-treatments, as well as at a followup 3 to 5 months after the completion of therapy with the following measures:

The Revised Child Impact of Events Scale (CRIES-13) is a 13 item scale measuring posttraumatic intrusion, avoidance, and hyperarousal. Items are answered on a four-point Likert-scale (0 = not at all, 1 = rarely, 3 = sometimes, 5 = often). The total score ranges between 0 and 75 with a cut-off score at 30 suggesting a probable diagnosis of PTSD (Perrin et al., 2005).

The Child PTSD Symptom Scale (CPSS) consists of 17 symptom items that are answered on a four-point scale from 0 (not at all) to 3 (5 or more times a week); thus, the total score ranges between 0 and 51 with a cut-off score at 11 indicating more than mild posttraumatic stress and a score of 19.1 indicating moderate posttraumatic stress (Foa et al., 2001). Note that other studies found different cut-off scores to be the optimal cut-point for the highest specificity, e.g., 16 (Nixon et al., 2013) and 21.5 (Hukkelberg et al., 2014). Seven further items assessing impairment in functioning were not analyzed in this study. We decided to apply both measures because of their respective advantages: the CRIES is a well-established and an easy comprehensible measure with the disadvantage that it does not refer to the DSM-IV. The CPSS might be a little less comprehensible but refers explicitly to the DSM-IV which was the relevant classification system at the time the therapies were conducted.

In the first 11 patients, measures were administered only before and after treatment. Beginning with the twelfth patient, we planned a further follow-up administration. All outcome measures show good retestreliability: CRIES rtt = 0.85 (Verlinden et al., 2014), CPSS rtt = 0.84 (Foa et al., 2001).

# Intervention

The treatment was conducted by a clinical psychologist with extensive training in MCT (the first author in this study) and followed the manual developed and published by Wells and Sembi (2004) and Wells (2009) with only slight adaptions for the younger patients. It comprised of up to 14 sessions, each of about 40 to 50 min duration. An involvement of the parents in the treatment was not planned and only done if deemed necessary. Treatment was terminated when the patient and the therapist agreed that all the treatment goals (i.e., significant reduction of posttraumatic stress symptoms and resulting functional impairment) were achieved. The treatment started with a joint case formulation and becoming acquainted with the metacognitive model. Patients were introduced to the idea that the processing of a traumatic event is largely automatic, like the healing of a wound. However, the healing of a wound can be painful (itchy) and distressing and some people tend to scratch the wound which hampers the healing process. Likewise, some traumatized people utilize what they see as healing strategies like thought suppression, worrying, rumination, gap filling, threat monitoring, avoidance, and other behavioral strategies. Thus, treatment aimed to reduce and undo these unhelpful coping strategies. Thought suppression experiments (like: "Please, try not to think about a pink rabbit sitting on my head!") were conducted to demonstrate its paradoxical effect: When an individual tries to suppress specific thoughts, the frequency of these thoughts increases and becomes more accessible than before ("rebound effect," Wegner et al., 1987). Patients were introduced to new strategies in dealing with intrusive thoughts/memories. First, they learned to leave the thoughts alone ("detached mindfulness"). This was explained using analogies like the telephone metaphor: "You cannot decide when the phone rings, but you can learn to let it ring without picking up. Further, if the caller left a message on the answering machine/mailbox, you can deal

with it later. Similarly, it is not your decision when these thoughts pop into your mind, but you can learn to leave these thoughts alone and deal with them later." Likewise, patients learned to postpone worrying and rumination to a fixed time in the early evening which should not exceed 10 min. These postponement experiments aim to weaken metacognitive beliefs about the uncontrollability of worrying and rumination. Experiments and verbal strategies were used to challenge further negative and positive metacognitive beliefs about worrying and rumination. The treatment continued with attention modification experiments to reduce threat monitoring, oftentimes combined with being in social situations. For example patients were asked to enter situations they had avoided since the traumatic event while focusing their attention on the safe aspects of the situation. Treatment was terminated after discussing relapse prevention strategies. Cognitive behavioral strategies, like imaginal reliving or challenging of thoughts and beliefs about trauma, or repeated exposure in vivo with the aim of habituation, were not conducted. Patients were invited to talk about the traumatic event if they so wished, but in fact no one made use of this offer.

# Data Analyses

Analyses were performed using the Statistical Package for Social Sciences (SPSS, version 23.0). Raw scores were used for all analyses and the critical alpha level was set at 0.05. Due to the different numbers of questionnaires available at posttreatment and follow-up, sample sizes differed for each measure and time-point of measurement administration. Thus, analyses were performed separately for each questionnaire and timepoint. The main analyses were comprised of paired-sample t-tests on data of the CRIES and CPSS for pre-post-treatment comparisons, while the non-parametric Wilcoxon Signed-Ranks test was used on comparisons of post-treatment data with followup data on the CRIES and CPSS because of the small sample sizes. Corrected effect sizes of Cohen's d for significant effects of the treatment effects analyses (Hedges and Olkin, 1985; Cohen, 1988; Cumming and Finch, 2005) and reliable change indices (RCI) were documented for each participant to discover clinically meaningful changes in the individual score beyond measurement error<sup>1</sup> .

To calculate the effect size, we used the formula of Cohen's d = M1–M2/SDPooled. To calculate the RCI, we subtracted the post-treatment score from the pre-treatment score and divided the result by the standard error of the difference between the two scores, which was calculated using standard deviations of the current sample and reliability coefficients of the test instrument [RCI = (posttest – pretest)/SEM)] An RCI value greater than 1.96 is considered a clinically significant improvement ( RCI > 1.96: "improved," −1.96 < RCI < 1.96: "unchanged;" RCI < −1.96: "impaired;" Jacobson and Truax, 1991). As a further criterion of clinically significant change, we investigated if the scores of the CRIES and the CPSS fell below the cut-off scores. Using the RCI and the cut-off points, each patient could be classified as recovered (passed both criteria), improved (passed only the RCI criterion in the positive direction), unchanged (did not pass the RCI criterion), or impaired (passed the RCI criterion in the negative direction).

# RESULTS

All patients entering the study completed treatment; one patient who did not benefit from treatment was referred to inpatient therapy after completion. Treatment was rather short with an average of 7 sessions (range 3–14). Even the youngest patient, an 8 year old girl with chronic PTSD after house fire, completed therapy successfully after only five sessions. Treatment was shorter when specific processes, especially gap filling, or positive metacognitive beliefs could not be identified and thus were not in need of change. Although we did not collect data on parent's involvement, we can state that they were involved very rarely. As can be seen in **Table 2**, we were able to obtain pre to post treatment data on at least one outcome measure for all of the 21 patients.

Because of incomplete data sets due to administrative error (i.e., measures not given), three patients had to be excluded from pre-post calculations on CRIES and CPSS. Thus, analysis on CRIES and CPSS contained 18 patients (14 females, mean age 14.67, range 10–19 years) Further, because we started to collect follow-up data relatively late in the course of the study, data exists of only 6 patients (all female, mean age 14.33, range 13–17 years) for the CRIES and CPSS.

**Table 3** presents the means and standard deviations of each outcome measure at pre-treatment and post-treatment, and also the effect size (Cohen's d) from pre- to post-treatment.

# Pre-post Treatment Effects

t-Tests indicate that patients' symptoms improved significantly from pre- to post-treatment (CRIES: t(17) = 14.32, p < 0.001, d = −3.42; CPSS: t(17) = 8.23, p < 0.001, d = −1.92).

# Clinical Significance

At post-treatment, all but one of the 21 patients scored below the cut-off and no longer met the diagnostic criteria for PTSD. In regard to the individual RCI (see **Table 2**), clinically significant improvement was found in 18 out of 19 patients (CRIES), and in 17 out of 20 patients (CPSS). Thus, depending on the outcome measure, 95% (CRIES) and 85% (CPSS) met criteria for recovery, whereas one patient (subject 8) was found to be unchanged (CRIES) and impaired (CPSS).

# Follow-Up Treatment Effects

For both outcome measures of the CRIES and CPSS, Wilcoxon Signed-Ranks tests did not reveal significant differences (CRIES: MdnPost = 4.5, MdnFU = 5.0, Z = −0.14, p = 0.89; MdnPost = 3.5, MdnFU = 3.5, CPSS: Z = −0.37, p = 0.72), indicating that improvement in PTSD symptoms was maintained from

<sup>1</sup>The retest-reliability for each test to compute the RCI scores have been previously reported (Foa et al., 2001; Verlinden et al., 2014).


CRIES, Children's Revised Impact of Events Scale; CPSS, Child PTSD Symptom Scale; RCI, Reliable Change Index; Pre, Pre-treatment; Post, Post-treatment; FU, follow-up. <sup>∗</sup>Scores below the cut-off point.

TABLE 3 | Means and standard deviations for outcome measures at pre-treatment, post-treatment, and follow-up, Cohen's d for pre- to post-treatment.


CRIES, Children's Revised Impact of Events Scale; CPSS, Child PTSD Symptom Scale; M, Mean; SD, Standard Deviation; Pre, Pre-treatment; Post, Post-treatment; FU, Follow-up.

post-treatment to the follow-up at least in the 6 cases measured over this time frame.

# DISCUSSION

This is the first study that aimed to test the feasibility, acceptability, and effects associated with MCT in the treatment of PTSD in youths. The results show that treatment was feasible as indexed by all patients completing the course. In addition, the duration of treatment was within the range recommended for adults with MCT. Treatment was associated with clinically significant effects in posttraumatic stress symptoms in almost all participants. Effect sizes were large (Cohen's d = 3.42 in CRIES, d = 1.92 in CPSS) and seemingly higher than effect sizes reported in a study of Tf-CBT (Goldbeck et al., 2016). Depending on the outcome measure, 85 or 95% of patients were found to have recovered. In all 6 patients available for the follow-up, the improvement was maintained.

It appears that MCT treatment can be delivered to traumatized youths in a small number of sessions (mean = 7, range 3– 14) of 40–50 min duration each and it is associated with large symptom improvements. This compares favorably with Tf-CBT which is usually conducted over 10–12 sessions each lasting 90 min. Thus, MCT might be more time effective than Tf-CBT, but this remains to be directly tested. The results demonstrate the feasibility and possible efficacy of MCT and add to the recent literature evaluating MCT for PTSD in adults (Wells and Colbear, 2012; Wells et al., 2015). Further, if replicated, the results may have some important implications regarding the mediators of psychotherapy. First, an efficacious psychotherapy for PTSD may not have to be trauma-focused (i.e., imaginal reliving). To reduce posttraumatic intrusions, it may be sufficient to stop the efforts to suppress these thoughts as is practiced in detached mindfulness, and to reduce extended thinking processes.

The limitations of this study are obvious; it is an uncontrolled study with a single therapist. Furthermore, at follow-up, only six patients were available. Measures of metacognition, anxiety, and depression were not included and neither were parent reported outcomes. Because of the small sample size, moderators of treatment efficacy like comorbidity or type of traumatic event were not examined. Further, stable pre-treatment baselines were

not established. Thus, improvement could also be attributed to spontaneous remission. However, twelve patients (57%) suffered more than 6 months from PTSD which makes spontaneous remission rather improbable (Hiller et al., 2016). However, we cannot partial the effects of treatment from other possible influences on symptom change.

Despite these major limitations, the results show that a course of MCT treatment could be implemented with children and adolescents suffering from PTSD over a course consistent with adult treatment. The results signal the need for a better controlled study (i.e., randomization, blind assessors, different therapists, etc.,), testing MCT against a

# REFERENCES


well-established treatment of PTSD, like Tf-CBT or Eye Movement Desensitization and Reprocessing (EMDR). An investigation of the importance of changes in cognitions and metacognitions in the efficacy of treatment would also be of further interest.

# AUTHOR CONTRIBUTIONS

MS initiated the project and conducted assessment and therapy. A-LK analyzed the data and wrote the passages of data analyses and results.

health clinics. Psychother. Psychosom. 85, 159–170. doi: 10.1159/0004 42824


Wells, A., and Sembi, S. (2004). Metacognitive therapy for PTSD: a preliminary investigation of a new brief treatment. J. Behav. Ther. Exp. Psychiatry 35, 307–318. doi: 10.1016/j.jbtep. 2004.07.001

Wells, A., Walton, D., Lovell, K., and Proctor, D. (2015). Metacognitive therapy versus prolonged exposure in adults with chronic post-traumatic stress disorder: a parallel randomized controlled trial. Cogn. Ther. Res. 39, 70–80. doi: 10.1007/s10608-014-9636-6

World Health Organization [WHO] (1992). International Classification of Diseases, 10th Edn. Geneva: World Health Organization.

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Simons and Kursawe. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

fpsyg-10-02741 January 14, 2020 Time: 17:20 # 1

# Metacognitive Therapy for Individuals at High Risk of Developing Psychosis: A Pilot Study

Sophie Kate Parker1,2,3 \*, Lee D. Mulligan<sup>1</sup> , Philip Milner<sup>1</sup> , Samantha Bowe<sup>1</sup> and Jasper E. Palmier-Claus4,5

<sup>1</sup> Psychosis Research Unit, Greater Manchester Mental Health NHS Foundation Trust, Manchester, United Kingdom, <sup>2</sup> Youth Mental Health Research Unit, Greater Manchester Mental Health NHS Foundation Trust, Manchester, United Kingdom, <sup>3</sup> Division of Psychology and Mental Health, University of Manchester, Manchester, United Kingdom, <sup>4</sup> Spectrum Centre for Mental Health Research, Lancaster University, Lancaster, United Kingdom, <sup>5</sup> Lancashire & South Cumbria NHS Foundation Trust, Lancashire, United Kingdom

#### Edited by:

Hans M. Nordahl, Norwegian University of Science and Technology, Norway

#### Reviewed by:

Stian Solem, Norwegian University of Science and Technology, Norway Lotta Winter, Hannover Medical School, Germany

> \*Correspondence: Sophie Kate Parker sophie.parker@gmmh.nhs.uk

#### Specialty section:

This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology

Received: 07 February 2019 Accepted: 20 November 2019 Published: 17 January 2020

#### Citation:

Parker SK, Mulligan LD, Milner P, Bowe S and Palmier-Claus JE (2020) Metacognitive Therapy for Individuals at High Risk of Developing Psychosis: A Pilot Study. Front. Psychol. 10:2741. doi: 10.3389/fpsyg.2019.02741 Developing effective interventions for preventing first episode psychosis have been an important research focus in the last decade. Cognitive behavioral therapy is a currently indicated treatment for people at ultra-high risk of psychosis, however, access and resource issues limit its delivery within the NHS. Treatments which partial out potential active ingredients and are aimed at a range of psychological difficulties seen within this population have the potential to be more efficacious and efficient. We conducted a single-arm exploratory pilot trial, designed to investigate the feasibility and acceptability of Metacognitive therapy for individuals at ultra-high risk (UHR) of developing psychosis. Trial uptake was good, with 11 out of 12 referred individuals meeting for an eligibility assessment (one individual was excluded prior to the assessment). Of these, 10 individuals were eligible and included in the trial. Retention to treatment was high with 80% treatment adherence gained and an overall average of 8 sessions completed. All participants were offered follow-up assessments immediately post-treatment and at 6 months, which comprised measures of psychotic like experiences, anxiety and depression, and metacognitive processes implicated in the model. Retention to the post-treatment (12-week) follow-up was good, with 80% completion; however retention to the 6-month follow-up was lower at 60%. Clinically significant results were observed in psychotic like experiences, anxiety, depression and functioning with medium to large effect sizes. Measures related to beliefs and processes targeted within MCT showed clinically significant change with medium to large effect sizes. Our results suggest that MCT based upon a specific metacognitive model for individuals meeting ARMS criteria may be an important treatment target and warrants further attention. Limitations and possible focuses for future research are discussed.

### Registration: ISRCTN53190465 http://www.isrctn.com/ISRCTN53190465.

Keywords: metacognition, metacognitive therapy, at risk mental states, psychosis, cognitive attentional syndrome

# INTRODUCTION

fpsyg-10-02741 January 14, 2020 Time: 17:20 # 2

Given the cost to individuals, families and services of psychosis, it is unsurprising that there has been great emphasis in research on the prevention of the development of a first episode. There are reliable and valid criteria available to identify help-seeking individuals in diverse settings who are at high risk of developing psychosis. Yung et al. (1998) developed operational criteria to identify three subgroups possessing an "at risk mental state" (ARMS) for psychosis. Two subgroups specify state risk factors, defined by the presence of either transient psychotic symptoms, called Brief Limited Intermittent Psychotic Symptoms (BLIPS) or attenuated (subclinical) psychotic symptoms (AS). The other subgroup comprises trait-plus-state risk factors, operationally defined by the presence of diminished functioning plus a firstdegree relative with a history of psychosis. All subgroups are within a specified age range known to be at greatest risk for the onset of psychosis, and all participants in studies of ARMS to date have been help-seeking.

In addition to identification, developing effective interventions to prevent or delay transition to psychosis have been an important research focus, given the potential benefits for symptoms, recovery and other outcomes. To date, there have been eight randomized controlled trials, each using similar operational definitions of ARMS, that have investigated antipsychotic medication, omega-3 polyunsaturated fatty acids and/or psychological interventions. The studies were conducted in Australia (McGorry et al., 2002; Yung et al., 2011), North America (McGlashan et al., 2006; Addington et al., 2011), the United Kingdom (Morrison et al., 2004a, 2006, 2012), the Netherlands (van der Gaag et al., 2012; Ising et al., 2016) and Austria (Amminger et al., 2010). Significant benefits at 12 months post-intervention were found for both cognitive behavioral therapy (Morrison et al., 2004a, 2006; van der Gaag et al., 2012) and omega-3 polyunsaturated fatty acids (Amminger et al., 2010). Therefore, at present, the recommended psychological treatment for young people at high risk of developing psychosis is cognitive behavioral therapy. The treatment duration indicated by the current evidence base (see Stafford et al., 2013) may be up to 26 sessions if following an appropriate manual for cognitive behavioral therapy for young people at risk of psychosis (e.g., French and Morrison, 2004). However, only a small number of young people meeting the at-risk of psychosis criteria are offered such indicated interventions in the NHS. In 2016 an audit found that only 41% of clients under EIS nationwide are offered CBTp within 6 months of acceptance into EIT (Health Quality Improvement Partnership and the Royal College of Psychiatrists, 2016). The audit operationalized an offer of CBTp as an offer of 16 sessions, delivered by appropriately trained and supervised therapists.

The United Kingdom access and waiting time standards (Nhs England, the National Collaborating Centre for Mental Health, and the National Institute for Health, and Care Excellence, 2016), which came into force from 1st April 2016, ensures that Early Interventions Services offer assessments for ARMS. However, the ability for NHS services to offer indicated interventions in line with the research protocols for people meeting ARMS criteria is limited given their stretched resources. If we can find out more about the active ingredients in psychological interventions, we may be able to make our interventions more efficacious and efficient. There has also been a call for more intervention trials aimed at the range of psychopathology observed in those at high risk of developing psychosis, to inform best care practices (Carpenter, 2018).

The cognitive model developed by Morrison (2001) implicates both cognitive and metacognitive processes in the development of psychosis. It is possible that using a specifically metacognitive approach could be a more efficient, quicker treatment compared with a mixed model. Studies have demonstrated a role of metacognitive beliefs and processes in the development and maintenance of psychosis (Smári et al., 1994; Morrison et al., 2002, 2004b, 2005, 2007; Morrison and Wells, 2003, 2007; Sellers et al., 2018). Metacognitive processes are also prevalent and important in those meeting criteria for ARMS (e.g., Morrison et al., 2007; Brett et al., 2009; Debbané et al., 2009; Barkus et al., 2010; Debbané et al., 2012; Palmier-Claus et al., 2013; Cotter et al., 2017).

Different metacognitive models and the approaches they derive for the treatment of psychosis have been summarized by Lysaker et al. (2018). These models are distinct and underpinned by different theoretical perspectives. The approach described here is underpinned by the S-REF (self-regulatory executive function) model proposed by Wells and Matthews (1994, 1996) and is not to be confused with metacognition as defined within other models being used within the area of psychosis (e.g., Lysaker et al., 2005). Wells and Matthews propose that it is not the occurrence of mental events (i.e., negative thoughts and emotion) that give rise to prolonged distress, but the resulting perseverative thinking style called the cognitive attentional syndrome (CAS). The CAS is comprised of strategies aimed at managing distressing thoughts and emotions which include worry, rumination, threat monitoring thought control strategies and maladaptive coping behaviors such as avoidance and reassurance seeking (Wells, 2008). The model implicates a central role of the CAS which becomes employed in response to negative thoughts and feelings causing an extension to psychological distress and worsening and extending negative affect. The S-REF model hypothesizes that CAS activity is promoted by underlying metacognitive beliefs both positive and negative in orientation. For example, people hold positive beliefs such as "worrying will help me to be prepared" and on the other hand negative beliefs about the uncontrollability and danger of thoughts and feelings such as "I cannot control my worrying once it begins."

A large body of evidence implicates a central role of metacognition in numerous mental health problems including generalized anxiety disorder, social anxiety, depression, PTSD and psychosis (Wells, 1995; Clark and Wells, 1995; Morrison, 2001; Papageorgiou and Wells, 2001), and metacognitive therapies for such problems are being applied successfully (Wells, 2000, 2008; Normann and Morina, 2018). A metacognitive model of the positive symptoms of psychosis has been developed (Morrison, 2001). This evidence-based model fpsyg-10-02741 January 14, 2020 Time: 17:20 # 3

predicts that metacognitive therapy may help to reduce psychotic like symptoms and target symptoms of co-morbid emotional disorders.

Where metacognitive processes are implicated, it is likely that specific metacognitive therapy for people at high risk of developing psychosis will be effective. Recent evidence has suggested that metacognitive therapy (MCT; Wells and Matthews, 1994, 1996) is a useful alternative to CBT for understanding and treating disorders such as generalized anxiety disorder, post-traumatic stress disorder, obsessive compulsive disorder and depression (Wells and King, 2006; Wells, 2008; Normann and Morina, 2018). MCT is a shorter treatment than traditional CBT, requiring around 6–8 sessions for symptom improvement (Wells, 2008). It has low drop-out rates and appears to be well-tolerated in emotional disorders. MCT could be particularly useful as an alternative treatment for young people at high risk of developing psychosis as it does not directly challenge the patients' belief systems, but rather focuses on the process, of thinking. It is also potentially generalizable to the other axis I disorders which have high co-morbidity within young at-risk individuals (e.g., Leicester et al., 2002).

A single arm feasibility study of 12 sessions of MCT for individuals with psychosis has been conducted (Morrison et al., 2014). This study successfully recruited 10 participants and adherence to MCT was shown to be acceptable; all participants received at least one session and 9/10 received 6 sessions or more (a mean of 10.6). The treatment demonstrated encouraging within-subjects effect sizes on positive symptoms (Cohen's d = 1.27) and delusional beliefs (Cohen's d = 0.71), and on negative symptoms (Cohen's d = 0.62), for which the evidence base in support of CBT is sparse. These positive results for individuals who by definition have a more 'serious' symptom profile suggests that MCT may also be useful for people at high risk of developing psychosis.

This pilot trial provides a preliminary investigation into the acceptability and feasibility of MCT for people meeting ARMS criteria who were experiencing distressing symptoms. It also provides an initial investigation into the efficacy of MCT in producing relief from psychotic like symptoms. In line with standard feasibility aims, the objectives of this study were to assess recruitment rate and to examine the appropriateness, feasibility and acceptability of the intervention and measures. It was hypothesized that MCT would produce symptom relief from unusual or overvalued beliefs (e.g., paranoia) and perceptual experiences (e.g., hallucinations), defined by significantly reduced CAARMS scores at both end of treatment and follow-up.

# MATERIALS AND METHODS

# Design

This study was a single-arm exploratory trial, designed to investigate the feasibility and acceptability of MCT for individuals at ultra-high risk (UHR) of developing psychosis. A National Research Ethics Committee approved the study prior to commencing data collection (13/NW/0238).

# Participants

All participants were being seen by NHS services specifically developed to work with people at high risk of developing psychosis [e.g., an Early Detection and Intervention Team (EDIT) or Early Intervention Service (EIS)] in the community. All participants met criteria for being at UHR of developing psychosis as operationally defined by the Comprehensive Assessment of At-Risk Mental States (CAARMS: Yung et al., 2005). NHS patients are typically referred to such services by primary care clinicians e.g., General Practitioners (GPs) or primary care psychology services, and should be offered assessment, ongoing monitoring of their mental health and Cognitive Behavioral Therapy (CBT).

Our exclusion criteria were: (i) a moderate to severe learning disability; (ii) a neurological impairment of organic origin (e.g., head injury or dementia); (iii) limited command of the English language, sufficient to impede the use of standardized assessments or accessibility of therapy; (iv) currently receiving inpatient care; (v) judged by their case manager to be clinically unstable over the 4 weeks prior to participation; (vi) taking prescribed antipsychotic medication; or (vii) a primary diagnosis of substance dependency.

# Measurements

## Primary Outcomes

The primary outcome measure was to assess feasibility and acceptability outcomes including levels of recruitment into the trial, retention of participants across baseline assessment, intervention and follow-up periods, number of "drop outs," defined as an individual who attended three or less therapy sessions, and adherence to the therapy protocol.

## Secondary Outcomes

A number of secondary outcomes were included. The CAARMS (Yung et al., 2005); a semi-structured interview credited as the gold-standard for assessing "at risk" symptoms. The CAARMS interview comprises six subscales assessing unusual thought content, non-bizarre ideas, perceptual abnormalities, disorganized speech, aggressive behavior, and suicidality over the previous month. Each subscale is rated on two seven-point scales according to the severity and frequency of any endorsed symptom. Scores for each subscale are derived from the product of the severity (0–6) and frequency scores (0–6). A number of studies have shown that the CAARMS has excellent inter-rater reliability, in addition to concurrent, discriminant and predictive validity (Yung et al., 2005).

The Global Assessment of Functioning (GAF; Hall, 1995) was used to measure personal, social and psychological functioning. The GAF is a semi-structured interview measure used in conjunction with the CAARMS and scores range from 0 to 100, with higher scores indicating greater global functioning. The GAF has been widely validated (Jones et al., 1995) and has been used extensively in studies examining UHR samples (Hartmann et al., 2016).

fpsyg-10-02741 January 14, 2020 Time: 17:20 # 4

Anxiety and depression symptoms were measured using the Hospital Anxiety and Depression Scale (HADS; Zigmond and Snaith, 1983), a 14 item self-report questionnaire providing separate total scores for anxiety and depression severity. All items are rated on four-point scales in reference to the previous week and higher scores indicate greater symptom severity. The HADS is considered to possess adequate to good properties of sensitivity, case-finding, concurrent validity and internal consistency (Bjelland et al., 2002). It is a brief and straightforward self-report measure which assesses both depression and anxiety (Hansson et al., 2009) thus reducing the burden of completion for participants. It has been used in similar populations to ours, such as adolescents, young people and adults with psychosis (Bernard et al., 2006; White et al., 2011; Pyle et al., 2019), therefore allowing comparability of our results with similar studies.

Metacognitive beliefs were assessed using the metacognitive questionnaire (MCQ-30; Wells and Cartwright-Hatton, 2004). This is a 30-item measure comprising five subscales: positive beliefs about worry; negative beliefs about uncontrollability and danger of extended processing; cognitive confidence; cognitive self-consciousness; and need to control thoughts. All items are rated on four-point Likert scales ranging from 1 ("do not agree") to 4 ("agree very much"). The MCQ has strong psychometric properties and has been used in studies examining UHR populations (Morrison et al., 2007; Bright et al., 2018).

Activation of the CAS was measured using the CAS scale (CAS-1; Wells, 2008); a 16-item questionnaire assessing worry, threat monitoring, strategies in response to negative thoughts or feelings, and metacognitive beliefs about extended processing and thought control strategies. For this study, only items assessing degrees of worry and threat monitoring were included. Both items are rated on eight-point scales ranging from 0 ("none of the time") to 8 ("all of the time"), in reference to experiences during the previous week. The CAS-1 has strong psychometric properties, including good internal consistency, concurrent and predictive validity (Sellers et al., 2018).

Appraisals of voice hearing were measured using the Interpretations of Voices Inventory (IVI; Morrison et al., 2002), a 26-item self-report questionnaire assessing positive and negative hypothetical interpretations of voices. All items are rated on four-point Likert scales ranging from 1 ("not at all") to 4 ("very much"). The IVI comprises three subscales: metaphysical beliefs, positive beliefs and beliefs about loss of control. The IVI is reliable and valid for use with people defined as high in psychosisproneness (Morrison et al., 2004b).

Metacognitive beliefs about paranoia were assessed using the Beliefs about Paranoia Scale – Short form (BAPS; Gumley et al., 2011), an 18-item self-report questionnaire measuring conviction in positive and negative interpretations. Each item is measured on a four-point Likert scale ranging from 1 ("not at all") to 4 ("very much"). The BAPS can be subdivided into three scales: negative beliefs about paranoia, positive beliefs about paranoia as a survival strategy and normalizing beliefs. The BAPS has strong psychometric properties (Morrison et al., 2011) and has been used with UHR samples (Morrison et al., 2015).

# Procedure

All participants were recruited from EDIT and EIS teams. Individuals were identified and approached by members of their clinical teams to participate in this study. Participants completed a battery of assessments (CAARMS, HADS, MCQ-30, CAS-1, IVI, BAPS) at baseline (pre-therapy), end of therapy (3 months) and 6 months post-therapy. Two clinical measures (HADS, CAS-1) were administered prior to each therapy session. All assessments were conducted by a trainee clinical psychologist or qualified clinical psychologist with extensive training and experience of administering the CAARMS. Throughout the study, all CAARMS scores were reviewed in supervision and ratings were finalized through group discussion. To reduce bias, all follow-up assessments (end of therapy and 6-months post therapy) were conducted by a therapist who was not involved in the delivery of therapy, for each respective participant.

# Intervention

The MCT intervention consisted of 12 sessions over a period of 12 weeks following baseline assessment, and followed the treatment manual developed by Wells (2009). We adapted the metacognitive model of generalized anxiety disorder (Wells, 1995) for use with UHR individuals, in a similar way to the therapy previously described for people with a diagnosis of schizophrenia (Hutton et al., 2014; Morrison et al., 2014). The metacognitive model asserts that psychological distress results from extended processing in response to negative cognitions (comprising thoughts, images, voices etc.). Examples of extended processing include worry, rumination and unhelpful thought control strategies, collectively termed as the CAS (Wells and Matthews, 1996). MCT aims to reduce the CAS by exploring new ways of responding to worrying thoughts and modifying metacognitive beliefs, which contribute to worry, rumination and distress. The MCT intervention consisted of:


6. Developing and reinforcing new plans for processing worry where old plans and new plans are described side by side.

7. Relapse prevention.

fpsyg-10-02741 January 14, 2020 Time: 17:20 # 5

# Therapists

Four therapists delivered the intervention. All therapists received training in the MCT manual and received weekly supervision to ensure adherence to the model. With participant consent, supervisors reviewed audio recordings of therapy to maximize fidelity.

# Statistical Analysis

Data were analyzed using Stata 14.0 (Stata Corporation, 2011). Emphasis was placed on descriptive and summary statistics, and flow across the different stages of the trial. In the absence of normally distributed data, Wilcoxon signed-rank tests assessed differences between assessment scores at baseline and posttherapy, and at baseline and 6-month follow-up. Summary effect sizes were also calculated (Cohen's D) using SD pre as the pre-test value provides an arguably better estimate of the true population value for within-subjects designs. This value is also thought to provide a better comparison to the d statistic in paireddesign experiments thereby making it useful in meta-analysis (Cummings, 2012). Regression, with clustering at the participant level (Rogers, 1993), was used to explore the relationship between meta-cognitive beliefs and worry (CAS-1) across the sessional measures in a 'long-form' of the data.

# RESULTS

The trial ended in December 2015 with a final sample size of 10. Demographic information for the sample is presented in **Table 1**. The consort diagram (**Figure 1**) shows the size of the sample across the different stages of the study. Uptake of the trial was good. Out of 12 referred individuals, 11 met for baseline eligibility assessments. One individual was excluded prior to assessment, and one was found to be ineligible post-baseline assessment due to not meeting CAARMS threshold criteria. Adherence to MCT was adequate with participants completing an average of 8.0 sessions (SD, 4.4; range, 1–12.). Two participants only completed one session due to changes in employment and unstable life circumstances, respectively. Eighty percent of participants completed the posttreatment (12-week) assessment, (one of these participants only


completed the CAARMS assessment and not the questionnaire measures) whereas 60% repeated these assessments at the 6 month follow-up. This was due to clients moving out of area (n = 1), declining assessment (n = 2), and physical health complications (n = 1).

Participants' scores for psychotic like experiences and functioning suggested that they had higher levels of symptoms and were functioning poorer when compared with larger samples of young people meeting UHR criteria (e.g., Morrison et al., 2011). In line with this, only one participant was not in the NEET category whilst 9/10 participants were not in education employment or training. The sample also demonstrated above cut off levels for anxiety (within the severe range) and depression (within the moderate range). Scores on the MCQ were considerably higher than equivalent populations (e.g., Bright et al., 2018), indicative of a sample who had higher levels of psychopathology than seen in previous trials of other interventions e.g., CBT.

Of the eight participants who completed the 12-week assessments, three were still at risk of psychosis, four no longer met ARMS criteria, and one had transitioned to a first psychotic episode, although declined to be referred for further treatment. At the 6 months follow-up (n = 6), two clients were at-risk of developing psychosis and four no longer met ARMS criteria. Summary statistics for primary and secondary outcomes (mean, SD) and effect size analyses (Cohen's d) are presented in **Table 2**. In summary, at 12 weeks participants had significantly lower scores on four out of six CAARMS subscales: non-bizarre ideas (p = 0.018), perceptual abnormalities (p = 0.026), disorganized speech (p = 0.043), suicidal behavior (p = 0.042). HADS scores were significantly lower (anxiety: p = 0.017, depression p = 0.046), as were CAS-1 scores (worry: p = 0.018, threat monitoring: p = 0.028). IVI scores were significantly lower (p = 0.027) and GAF scores significantly higher (p = 0.035).

At 6 months, IVI scores remained significantly lower (p = 0.027) whilst only two subscales described above remained significantly lower: CAARMS non-bizarre ideas (p = 0.027) and HADS anxiety (p = 0.043). There were no significant differences at 12 weeks or 6 months on CAARMS unusual thought content (p = 0.345 and p = 0.068, respectively) or aggressive behavior subscales (p = 0.068 and p = 0.916, respectively). There were also no significant differences in BAPS score at 12 weeks (p = 0.173) or 6 months (p = 0.114).

In line with the mechanism of change, MCQ-30 total scores were significantly reduced at 12 weeks (p = 0.018) and 6 months (p = 0.028), as were three out of five subscales: negative beliefs about uncontrollability and danger (12 weeks: p = 0.018, 6 months: p = 0.026), cognitive confidence (12 weeks: p = 0.018, 6 months: p = 0.026), and negative beliefs about need to control thoughts (12 weeks: p = 0.018, 6 months: p = 0.027). There were no differences at either time-point on positive beliefs about worry (12 weeks: p = 0.172, 6 months: p = 0.068) or cognitive self-consciousness (12 weeks: p = 0.173, 6 months: p = 0.114).

Successful completion of sessional measures was high (98.75%). As can be seen in **Figure 2**, CAS-I worry scores generally declined over the course of therapy, which coincided fpsyg-10-02741 January 14, 2020 Time: 17:20 # 6

with reductions in CAS-I meta-cognitive belief scores. Regression, with clustering at the participant level, suggested that the strength of metacognitive beliefs significantly predicted levels of worry across the sessional measures (β = 0.73, SE: 0.07, p < 0.001, CI: 0.58–0.89).

# DISCUSSION

Our results suggest that MCT is an acceptable treatment for young people with an At Risk Mental State, evidenced by high rates of trial uptake and therapy adherence (80% treatment adherence and an overall average of eight sessions). We observed clinically significant reductions in psychotic like experiences at the post-treatment assessment (CAARMS subscales: Non-Bizarre ideas, Perceptual abnormalities, disorganized speech), retained on one subscale (Non-Bizarre ideas) at the 6-month follow-up point. We also found an important reduction in participants meeting ARMS criteria at the post-treatment assessment (four out of eight) which was retained at the 6-month follow-up. Only one participant made transition to first-episode psychosis across the follow-up period. Trial results also showed statistically significant improvements in anxiety and depression (Hospital Anxiety and Depression Scale) and functioning (Global Assessment of Functioning) at the post-treatment assessment. The significant change in anxiety, but no other secondary outcomes, was retained at the 6 month follow-up assessment. Our effect sizes suggest that the magnitude of the differences found are considered medium to large, although this must be interpreted with caution as given the trial design and small sample the effect sizes are likely to be inflated.

Measures related to beliefs and processes targeted within MCT were assessed over time and during treatment. Results demonstrated clinically significant change on metacognitive beliefs (MCQ-30 total score, beliefs about uncontrollability, need for control and danger and cognitive confidence) at both followup assessment points. This was also the case for people's selfrated beliefs about voices, on a measure (IVI) informed by a metacognitive model. We also observed statistically significant change on processes described within the CAS (worry and threat monitoring) at the post-treatment assessment, although this was not retained at 6 months. As with the previous outcomes, effect sizes on these measures demonstrated medium to large effects across assessments, although the effect sizes need to be interpreted with caution as previously described. Weekly measurement of worry and threat monitoring (recorded via CAS-1 subscales) showed that both reduced, at similar rates, throughout the course of treatment for those who were retained. These results suggest that MCT based upon a specific metacognitive model is capable of changing metacognitive beliefs and processes; which are hypothesized mechanisms within the model and therefore important treatment targets.

These positive results suggest that MCT appears to show potential in reducing psychotic like experiences, anxiety and depression, and increasing functioning for young people at UHR of psychosis. MCT is a relatively brief treatment compared with CBT, and therefore could be a useful treatment within of the contexts of resource restrictions and the importance of timely interventions for young people. It is not possible to conclude TABLE 2 | Summary statistics and outcome data for key variables.

fpsyg-10-02741 January 14, 2020 Time: 17:20 # 7


<sup>∗</sup>One participant failed to complete questionnaire measures. Key: CAARMS, Comprehensive Assessment of At-Risk Mental States; GAF, Global Assessment of Functioning; HADS, Hospital Anxiety and Depression Scale; CAS-1, Cognitive Attentional Syndrome; MCQ, Metacognitive Beliefs Questionnaire; IVI, Interpretation of Voices Inventory. Bold values signify statistically significant values.

from such a small sample about who may benefit the most from CBT versus MCT, or indeed who could be recommended for either treatment. However, if a definitive trial were to replicate our findings this could provide rationale for service users to be offered choice from a range of evidence-based treatments.

We did, however, observe no statistically significant differences in subscales related to psychotic like experiences (CAARMS unusual thought content), positive beliefs about worry, cognitive self-consciousness and beliefs about paranoia. The associated effect sizes for cognitive self-consciousness and beliefs about paranoia were medium to large, small to large on the CAARMS unusual thought content subscale, so it is possible that these findings are related to a lack of statistical power. There appeared to be little effect on positive metacognitive beliefs at both assessment points. This may reflect the relative shortness of attended therapy sessions (average of 8 sessions), and given that modification of positive metacognitive beliefs takes place in the later phase of therapy, it may not have been addressed adequately within the treatment window. It may be that, for this population, modifications should be explored in the length of treatment window offered in order to address this given the documented importance of assertive outreach principles in this area (Morrison, 2017). It is also possible that the relative inexperience of the therapists meant that the efficiency normally derived from MCT was not achieved in this study. As adherence and competency measures were not taken during the trial it's not possible to know if this is the case.

Acceptability of the MCT was high, with eight of ten participants adhering to treatment (operationalized as attendance to least four sessions). The remaining two participants withdrew early (after a single session) due to changes in life circumstances making attendance at therapy sessions difficult (physical health complications and moving out of area). We observed acceptable retention at the post-treatment assessment (80% completion) but higher rates of attrition at the 6-month follow-up (40%). The reasons for two participants not attending the follow-up assessment (described above) were unrelated to the trial procedures. However, two additional participants declined to take part in the 6-month follow-up assessment and it was not possible to explore their reasons for declining. Therefore, we cannot comment on any possible acceptability issues related to trial procedures for either of these participants. It will be important for future trials to explore the acceptability of trial procedures via qualitative interviews or feedback processes.

As would be expected in an exploratory trial of this kind there are a number of important methodological limitations that require consideration. The small sample size both reduces the statistical power and the generalizability of the findings, in part because our sample size did not meet statistical fpsyg-10-02741 January 14, 2020 Time: 17:20 # 8

requirements (e.g., normality) required for hypothesis testing (e.g., Shader, 2015). Small sample sizes also have a higher risk of providing imprecise estimates and therefore we must be highly cautious when interpreting the meaning of these results to the wider population. However, MCT has been found to produce large effects in other groups (e.g., GAD treatment), and therefore is consistent with previous findings. Given the limitations of our study, we are limited in being able to compare our findings to those of larger studies of psychological interventions for ARMS populations, and more data on MCT for ARMS populations is required for verification. However, in line with the aims of pilot studies, we were able to examine the feasibility of processes and procedures (i.e., recruitment, retention, implementation of MCT) in preparation for a larger RCT (Leon et al., 2011).

The trial design did not allow for comparison with a control group, and the assessors were not blind to the presence of the treatment; therefore rater bias and possible non-specific treatment effects have not been guarded against Additionally, we did not complete formal ratings of treatment adherence or therapist competency. This limits any analysis of fidelity to the treatment protocol, although therapists did audio-tape sessions (where consent allowed) and receive supervision by the first author following the treatment manual previously described. A further limitation is that we did not obtain any qualitative feedback from participants on their views of the treatment, or any satisfaction scores. Nonetheless, we found significant effects on a number of outcome measures and potential mechanisms implicated within the MCT model, some of which were derived from self-rated questionnaires which showed statistically significant change with associated medium to large effects.

MCT for individuals meeting ARMS criteria warrants further attention. In the future, it will be important to conduct another pilot trial to further test of the acceptability and feasibility of offering MCT compared with treatment-as-usual under randomized conditions and with a longer follow-up period. The CAARMS assessment seems to be an appropriate outcome measure and would allow for comparison with other trials of ARMS interventions. Further qualitative work is required to explore participants' unique experiences of the trial, and obtain their views on the appropriateness of the CAARMS as a primary outcome measure in a future definitive trial.

# ETHICS STATEMENT

This study was carried out in accordance with the recommendations of A National Research Ethics Committee, North West Greater Manchester West Ethics Committee (Ref. 13/NW/0238) with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the North West Greater Manchester West Ethics Committee.

# AUTHOR CONTRIBUTIONS

SP conceived and designed the study. SP, PM, LM, and JP-C provided the therapy. SP and SB supervised the study. SP, PM, LM, and JP-C performed the data collection. JP-C analyzed the data. SP, JP-C, and LM interpreted the data. SP, LM, JP-C, and SB drafted the manuscript. All authors revised and finalized the manuscript.

# REFERENCES

fpsyg-10-02741 January 14, 2020 Time: 17:20 # 9


# ACKNOWLEDGMENTS

We would like to thank the participants who contributed and the services who were referred to in the study.

risk" clients: investigation of a dilution effect. Schizophr. Res. 170, 130–136. doi: 10.1016/j.schres.2015.11.026


fpsyg-10-02741 January 14, 2020 Time: 17:20 # 10


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Parker, Mulligan, Milner, Bowe and Palmier-Claus. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Metacognitive Therapy for Alcohol Use Disorder: A Systematic Case Series

Gabriele Caselli1,2,3 \*, Francesca Martino1,2,4, Marcantonio M. Spada2,3 and Adrian Wells5,6

<sup>1</sup> Studi Cognitivi, Cognitive Psychotherapy School, Milan, Italy, <sup>2</sup> School of Applied Sciences, London South Bank University, London, United Kingdom, <sup>3</sup> Sigmund Freud University Milano, Milan, Italy, <sup>4</sup> Department of Medical and Surgical Sciences DIMEC, University of Bologna, Bologna, Italy, <sup>5</sup> Division of Clinical and Health Psychology, University of Manchester, Manchester, United Kingdom, <sup>6</sup> Greater Manchester Mental Health NHS Foundation Trust, Manchester, United Kingdom

Alcohol Use Disorder (AUD) is a debilitating condition with serious adverse effects on health and psycho-social functioning. The most effective psychological treatments for AUD show moderate efficacy and return to dysregulated alcohol use after treatment is still common. The aim of the present study was to evaluate Metacognitive Therapy (MCT) as applied to AUD. Five patients were treated using a non-concurrent multiple baseline design with follow-up at 3- and 6-months time points. Each patient received 12 onehour sessions of MCT. Following MCT all patients demonstrated large and clinically meaningful reductions in weekly alcohol use and number of binge drinking episodes that were upheld at follow-up in almost all cases. Metacognitive beliefs, as secondary outcome, also changed substantially. The findings from this study offer preliminary evidence of positive effects associated with MCT in AUD and support the need for a definitive trial of MCT in addictive behaviors.

#### Edited by:

Dana M. Litt, University of North Texas Health Science Center, United States

#### Reviewed by:

Lucia Romo, Université Paris Nanterre, France Paula Odriozola-González, University of Valladolid, Spain

> \*Correspondence: Gabriele Caselli

g.caselli@studicognitivi.net

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 25 July 2018 Accepted: 05 December 2018 Published: 19 December 2018

#### Citation:

Caselli G, Martino F, Spada MM and Wells A (2018) Metacognitive Therapy for Alcohol Use Disorder: A Systematic Case Series. Front. Psychol. 9:2619. doi: 10.3389/fpsyg.2018.02619 Keywords: addiction, alcohol use disorder, metacognition, metacognitive therapy, outcome, treatment

# INTRODUCTION

Alcohol Use Disorder (AUD) involves loss of control over alcohol use, a strong desire or urge to use alcohol, and continued alcohol use in hazardous situations despite awareness about of persistent or recurrent life problems caused by the effects of alcohol (DSM-5, American Psychiatric Association [APA], 2013). The harmful use of alcohol is one of the world's leading health risks and has been implicated in 5.9% of deaths globally (World Health Organization, 2014). Harmful alcohol use has also been associated with a wide range of mental health and social problems such as suicide (Merrill et al., 1992; Demirbas et al., 2003), increased risk of major depression (Boden and Fergusson, 2011), domestic violence (Leonard, 2001), child abuse (Widom and Hiller-Sturmhofel, 2001), and workplace absenteeism (Bacharach et al., 2010). A wide range of approaches have been developed to conceptualize and treat this disorder. Among them cognitive and behavioral models have highlighted the role of learning processes, cognitive biases and dysfunctional beliefs in the etiology and maintenance of AUD. One of the core principles underlying cognitive-behavioral therapy (CBT) for AUD is that alcohol serves as a powerful reinforcer of behavior. Over time, positive (e.g., enhancing social experiences) and negative (e.g., reducing negative affect) reinforcing effects of using alcohol become associated with a variety of internal and external stimuli. The cognitive component of these approaches highlights the role of barriers to change, such as biases, beliefs and expectancies which maintain alcohol use as a coping strategy to deal with negative

affect or to reach desired goals. CBT aims to reduce the strongly reinforcing effects of alcohol by: (1) identifying the problematic situations that lead to alcohol use and teaching patients copingskills to manage them (e.g., assertion drink refusal skills training); (2) increasing engagement in activities that are not related to alcohol use; and (3) removing motivational and cognitive barriers to change (e.g., Kadden, 2001; Donohue et al., 2004; Baillie et al., 2013). On the basis of these principles, a series of CBT protocols for AUD were developed and were extensively evaluated (Marlatt and Gordon, 1985; Monti et al., 1993; Kadden, 1995, 2001) with both abstinence and controlled drinking as treatment goals (e.g., Sanchez-Craig et al., 1984).

The CBT approach has provided valuable insights in the conceptualization and treatment of AUD, however, it is not without limitations. A central limitation of the behavioral component of CBT is that it does not elucidate why only a small proportion of individuals who use alcohol end up losing control over their use. A central limitation of the cognitive component of CBT is the failure to establish if irrational beliefs play a causal role in the etiology and development of AUD rather than being an epiphenomenon of this condition. These structural weaknesses of CBT as applied to AUD may explain its moderate effectiveness when compared to other forms of treatment, including medical management, treatment as usual, or active psychosocial treatments (e.g., Project Match Research Group, 1997; Burtscheidt et al., 2002; Balldin et al., 2003; Litt et al., 2003; Wetzel et al., 2004; Anton et al., 2005; Wolwer et al., 2011; Farren et al., 2014). In addition, treatment effects for CBT appear to diminish over time, especially at 6- to 9-month follow-up (Magill and Ray, 2009).

Drawing on the S-REF model (Wells and Matthews, 1994) it has been argued that a possible reason for CBT's lack of efficacy might be due to residual symptoms and mechanisms that remain present at a metacognitive level (Spada and Wells, 2008; Spada et al., 2009, 2015). Specifically, the modification of the content of biased cognitive beliefs, which is the main focus of CBT, does not directly modify metacognitive beliefs presumed to be driver of maladaptive cognitive processes (e.g., worry, rumination, desire thinking) as implicated in the S-REF.

Over the last twenty-five years the Self-Regulatory Executive Function (S-REF) model has offered novel insights into the role of metacognition in psychopathology (Wells and Matthews, 1994, 1996; Wells, 2000). Central to the S-REF model are the processes which monitor, generate and maintain intrusive and biased cognitive experiences (Wells, 2009). The S-REF model has led to a novel form of psychological therapy, Metacognitive Therapy (MCT; Wells, 2009), which has been applied to the treatment of anxiety and depression with notable results (e.g., Normann et al., 2014). From the metacognitive standpoint, psychological disturbances are maintained by the activation of the Cognitive-Attentional Syndrome (CAS) under conditions of distress. The CAS encompasses repetitive negative thinking styles (rumination and worry), thought suppression, maladaptive threat or self monitoring, and avoidance. The activation of the CAS brings an increase of attentional focus toward distress congruent information and feedback loop which fail to regulate threatening thoughts. The activation, perseveration and escalation of the CAS is linked to the presence of unhelpful metacognitive beliefs. These are beliefs about thinking and ways in which thinking can be controlled. Metacognitive beliefs are either positive (e.g., "Worrying will help me cope") or negative (e.g., "Thoughts are dangerous and should be controlled") and are associated to generic plans for guiding cognition and behavior.

Research undertaken over the last decade has proposed AUD may be conceptualized using this metacognitive perspective (Spada and Wells, 2005, 2006; Spada et al., 2013, 2015; Caselli et al., 2013b). According to this view, it has been argued that metacognitive beliefs lead to the activation of CAS components associated with AUD (such as perseverative thinking about alcohol-related intrusions, the monitoring of internal or external alcohol-related cues, and the reduction of adaptive metacognitive monitoring). Emerging evidence has supported this conceptualization when applied to different forms of perseverative thinking (e.g., desire thinking, rumination and worry) shown to be highly associated with craving and levels of alcohol use in both non-clinical and clinical samples through cross-sectional designs (Caselli et al., 2008, 2012; Goldsmith et al., 2009; Smith and Book, 2010; Caselli and Spada, 2011, 2015; Chakroun-Baggioni et al., 2017), experimental studies (Caselli et al., 2013a,b, 2017; Caselli and Spada, 2011, 2015, 2016) and longitudinal research (Caselli et al., 2010; Martino et al., 2017).

The detrimental interplay between alcohol use and adaptive metacognitive monitoring, another element of the CAS, is widely accepted. In particular impairment of attentional functioning appears to play a fundamental role in determining alcohol effects. For example, alcohol's pharmacological properties can narrow the perception to immediate cues and reduce the capacity for abstract reasoning (Steele and Josephs, 1990). In addition, alcohol reduces self-awareness, conceptualized as the ability to attribute self-relevance in encoding information (Hull, 1981) and neuroscientific evidence suggests that alcohol intoxication impairs neurological systems associated to meta-level processing (Nelson et al., 1998). All these processes are likely to play a relevant role in the effective monitoring of internal states once a drinking episode has started (Spada and Wells, 2006; Spada et al., 2007b). An ineffective monitoring of internal states (termed "metacognitive monitoring"; Spada and Wells, 2006) can lead to higher levels of alcohol use because information on emotional change (e.g., feeling relaxed) and proximity to goals of alcohol use (e.g., achieving a greater level of relaxation) that would serve as a stop signal is not attended to.

The links between metacognitive beliefs and aspects of the CAS in AUD is also now extensive. Research linking metacognitive beliefs on the one hand, and forms of perseverative thinking on the other, is well-established (e.g., Cartwright-Hatton and Wells, 1997; Wells and Papageorgiou, 1998; Papageorgiou and Wells, 2003; Wells and Cartwright-Hatton, 2004). This association has been extensively explored in AUD with similar findings (Spada and Wells, 2005; Spada et al., 2007b; Caselli and Spada, 2010, 2011). For example, cross-sectional research using self-report instruments has demonstrated that metacognitive beliefs are elevated in problem drinking (Spada and Wells, 2009). Furthermore, a longitudinal study showed how beliefs about the need to control thoughts predict levels of alcohol use and relapse

at 3, 6, and 12 months in a sample of problem drinkers (Spada et al., 2009). Finally, in research aimed at uncovering the structure of alcohol-specific metacognitive beliefs in problem drinkers, both positive and negative metacognitive belief systems related to alcohol use were identified (Spada et al., 2007a; Spada and Wells, 2008, 2009).

Taken together, these data support the applicability of the S-REF model to understanding the development and maintenance of AUD and suggest that metacognitive therapy (MCT, Wells, 2009) may be beneficial in treating it. A recent study examined whether a specific MCT technique, detached mindfulness, would be more effective than a control condition in reducing negative meta-appraisal of alcohol-related thoughts, the conviction in maladaptive metacognitive beliefs, and associated distress level and urge to use alcohol in a small sample of patients with AUD (Caselli et al., 2016). Findings suggested that detached mindfulness was associated with a faster change in status. This implies that a targeted focus on modifying the relationship to one's thoughts (rather than simply habituating to them) may be of benefit. The findings also support a broader and more extensive application of a whole MCT package for patients with AUD.

The present study aimed at examining the effects associated with a brief course of MCT in a series of patients with AUD. The main goal of the treatment was controlled or reduced-risk alcohol use. This was suggested as a pragmatic option, with a view to sustain patient engagement, because abstinence as a goal can often represent a barrier (Connor et al., 2016). In addition, from a metacognitive perspective, directly sustaining a controlled drinking goal is more likely to enhance metacognitive control which would otherwise not be achieved through abstinence because negative metacognitive beliefs about the uncontrollability of behavior and thoughts need to be tested through controlled behavior (i.e., continued and controlled alcohol use).

# METHODS

# Design

This case series adopted a non-concurrent multiple baseline (MB, Watson and Workman, 1981) design across individuals with follow-up in order to: (1) test the feasibility and replicability of MCT across different individuals with AUD; and (2) examine if MCT is associated with positive outcomes in these cases. The MB is a well-established design with a wide range of applications and a multitude of publications in the clinical field supporting its use and validity. MB is commonly used in cases where the dependent variable is not expected to return to normal after treatment has been applied (Carr, 2005). This kind of design can offer important advantages. Firstly, repeated measures can help to establish the prediction of a baseline's data path into the subsequent treatment phase and allow for the detection of a difference between the actual data path in treatment and the path predicted from baseline. Secondly, this effect can be replicated across different participants independently of the baselines' length. A detailed data collection with several time points and different baseline length can control for maturation, exposure to the clinical setting, repeated testing, and regression to the mean, increasing the confidence that any observable changes are attributable to the intervention. A predetermined set of baseline lengths was randomly selected and assigned to the five patients of the present study. Baseline length ranges were 3–7 weeks. Treatment was initiated at the predetermined time only if baseline was stable, otherwise extension of the baseline was planned. Stability was defined as an absence of a decreasing trend of at least two consecutive data-points prior to introduction of treatment. Treatment was constituted by 12 one-hour-long sessions as this timeframe had been found to be sufficient to complete the MCT protocol in pilot work. Following the screening assessment, patients received questionnaires on a weekly basis with a view to monitor alcohol use, number of binge drinking episodes and symptoms levels. Following the baseline period, MCT was delivered on a weekly basis. After treatment, patients were followed up at 3 and 6 months, no additional treatment was delivered during the follow-up period. The goal of MCT was to control alcohol use.

# Participants

Patients included in this study were the first five consecutively assessed individuals who met the following criteria: (1) primary diagnosis of AUD as determined by the SCID-5 (First et al., 2015); (2) age 18–65 years; (3) absence of borderline personality disorder; (4) absence of a concurrent psychological treatment; (5) no evidence of physical withdrawal syndrome; (6) no evidence of progressive cerebral traumas or severe cognitive deficits; (7) not actively suicidal; (8) medication free; (9) no concurrent substance use (apart from nicotine) in the previous 6 months; and (10) clear understanding of the Italian spoken language. These criteria were evaluated by the second author and a trained psychologist independently.

### Patient 1

Patient 1 was 25 to 30-years old and reported difficulties in regulating drinking especially during his job. He reported problematic alcohol use since his early teenage years, often associated with cannabis and/or cocaine use. He reported that alcohol use had been the main problematic issue in his life and gave him much trouble (e.g., law problems, brawls, problematic issues in both intimate relationships and workplace). He tried to reduce and/or stop alcohol use many times without long lasting positive outcomes. He reported contacting mental health services, both public and private, often driven by his family. He always rejected pharmacological treatment. He previously undertook 6 months of psychotherapy without any results and he was unable to define the approach employed. The patient also got in contact with Alcoholics Anonymous but abandoned after 2–3 meetings. He met criteria for Major Depressive Disorder of moderate severity.

### Patient 2

Patient 2 was 25 to 30-years old and reported difficulties in controlling alcohol use that began almost 10 years previously as a means of managing anxiety in social situations. He reported that

he began to consider his alcohol use as problematic 2 years earlier, when he had some relational problems with his girlfriend and friends because of his behavior when drunk. He attempted, over the last 2 years, to reduce alcohol use autonomously with some transient positive results but he experienced recurrent relapses. At the beginning of pre-treatment, he thought he had completely lost control over his alcohol use. In addition, he met the criteria for Social Anxiety Disorder. He was medication free and he had never had contact with mental health services.

### Patient 3

Patient 3 was 35 to 40-years old, he was unemployed and reported he was not able to continue his job because of drinking problems. He also reported that difficulties surrounding alcohol use started to become serious 12 years prior, with binge drinking episodes pre-dating this time. During last 10 years he began using alcohol when alone, and on a daily basis, and this habit gradually led to a reduction of social contacts and general withdrawal. He also met criteria for Major Depressive Disorder.

# Patient 4

Patient 4 was 60 to 65-years old and reported that stress related to his job and family difficulties were the main reason for his alcohol misuse. He reported he had never consumed too much alcohol until 10–15 years ago. He was unable to define a specific change in his life circumstance associated with his change in alcohol use, but he reported a general increase in life and work problems occurring at the time. At present, his excessive alcohol use persisted no matter how he tried to reduce it. He got in contact with mental health services during the previous 2 years, but he rejected both the goal of abstinence and pharmacological therapy with Disulfiram. He attended a handful of psychotherapy sessions but did not feel these were effective and he dropped out.

# Patient 5

Patient 5 was 35 to 40-years old and reported that recent problems with alcohol use had lasted 3 years and that drinking too heavily had featured intermittently since teenage years. The patient met criteria for dysthymia but was medication free. The only previous contact with mental health services was 2–3 assessment sessions with a psychotherapist 5 years ago. Patient reported having used other substances (cocaine) but not in the last 12 years.

# Outcome Measures

# Alcohol Use Disorders Identification Test Consumption (AUDIT-C; Bush et al., 1998)

The AUDIT-C includes items 1 to 3 of the 10-item AUDIT which assess alcohol use. Individuals select one of five statements (per question) that most applies to their alcohol use. Responses are scored from 0 to 4 with higher scores representing higher levels of problematic alcohol use. The summary score for the total AUDIT-C ranges from 0, indicating no presence of problematic alcohol use, to 12 indicating severe levels of problematic alcohol use. This self-report measure has been extensively adopted and possesses a well-established validity and reliability (Bush et al., 1998). The Italian version of the measure was used in the current study (Piccinelli et al., 1997).

# Hospital Anxiety and Depression Scale (HADS; Zigmond and Snaith, 1983)

The HADS consists of 14 items on a 4-level Likert scale that refer to how respondents have been feeling over the past week (from "Most of the time" to "Not at all"). The HADS includes two sub-scales (seven item each) assessing anxiety and depression. Higher scores represent higher levels of anxiety and depression. Overall, the scale possesses good validity and reliability and has been widely adopted in a wide range of clinical and non-clinical research settings (Zigmond and Snaith, 1983; Herrmann, 1997; Mykletun et al., 2001; Alati et al., 2004; Wagena et al., 2005). The Italian version of the measure was used (Costantini et al., 1999) which shows a good reliability with alpha equal to 0.89 and 0.88 for anxiety and depression sub-scales, respectively.

## Positive Alcohol Metacognitions Scale (PAMS; Spada and Wells, 2008)

The PAMS consists of 12 items which assess positive beliefs about the need to use alcohol as a cognitive and emotional self-regulation strategy (metacognitive beliefs). Higher scores indicate higher levels of positive metacognitive beliefs. The PAMS possesses a reliable factor structure and good internal consistency and validity in both clinical and non-clinical samples (Spada and Wells, 2008). The Italian version of the measure was used (alpha = 0.88; Di Blasi et al., 2013).

# Negative Alcohol Metacognitions Scale (NAMS; Spada and Wells, 2008)

The NAMS consists of six items which assess negative metacognitive beliefs about uncontrollability and cognitive harm of alcohol use. Higher scores indicate higher levels of negative metacognitive beliefs about alcohol use. The NAMS possesses a reliable factor structure and good internal consistency and validity in both clinical and non-clinical samples (Spada and Wells, 2008). The Italian version of the measure was used (alpha = 0.75; Di Blasi et al., 2013).

# The Penn Alcohol Craving Scale (PACS; Flannery et al., 1999)

The PACS is a 5-item scale that assesses the level of craving for alcohol. Its items refer to duration, frequency, intensity and uncontrollability of craving plus an overall evaluation of the subjective experience of craving during the previous week. Each question is scaled from 0 to 6. This instrument has been shown to possess good psychometric properties (Flannery et al., 1999). The Italian version of the measure was adopted (alpha = 0.80; Caselli and Spada, 2011).

### Quantity Frequency Scale (QFS; Cahalan et al., 1969)

This QFS consists of nine items assessing levels of alcohol use, with three sub-scales assessing the use of beer, spirits and wine. The total scores from the different sub-scales are added to estimate weekly level of alcohol use. The QFS has been extensively used and possesses good reliability and validity (Hester and

Miller, 1995). This instrument was completed on a weekly basis, referring to the previous week's alcohol use.

#### Cognitive Attentional Scale – Alcohol (CAS-A)

A self rating scale was constructed for this study to assess dimensions of the CAS and related metacognitive beliefs that are usually associated with AUD. Items that referred to CAS components included: (1) time spent ruminating on alcoholrelated thoughts; (2) associated distress; and (3) number of binge drinking episodes. Questions on metacognitive beliefs included 10 items referring to both positive metacognitive beliefs (e.g., "I need to drink in order to control my thoughts") and negative metacognitive beliefs (e.g., "I have no control over my drinking"). All dimensions apart from number of binge drinking episodes were rated for the past week on 0–100 scales. The psychometric properties of this instrument have not been evaluated.

# Procedure

We sought and obtained ethics approval for the study from the Ethics Committees of Studi Cognitivi Research Institute Ethics and the School of Applied Sciences at London South Bank University (UREC1503). Participants referred for alcohol-related problems to outpatient clinics in Milan and Modena were invited for an assessment interview in order to determine eligibility for the study. The same invitation was offered to those who had directly contacted the project lead after seeing leaflets and web announcements. All patients were assessed independently by the second author and a psychologist to confirm the diagnosis of AUD and evaluate inclusion and exclusion criteria. After agreement between assessors and informed consent were obtained, an initial and complete assessment was administered. Four participants were excluded from the study because of presence of Borderline Personality Disorder (2 participants) and lack of a primary diagnosis of AUD (2 participants). Weekly ratings were taken for the QSF and CAS-A over the baseline period. The self-report questionnaires were administered to patients on a weekly basis. Once the predetermined baseline length was reached, a fuller assessment was conducted which involved the administration of all self-report measures to be repeated at post-treatment and at 3 and 6-months follow-up. During treatment, QFS and CAS-A were completed at the beginning of each session.

# Treatment

The MCT protocol for AUD consisted of 12 weekly sessions of 45–60 min duration and followed the core MCT steps as developed by the fourth author (Wells, 2009) adapted to the metacognitive formulation of AUD (Spada et al., 2013, see **Table 1**). In the first treatment session an idiosyncratic case formulation based on the metacognitive model of AUD was presented as a basis for a socialization to the model that followed. The latter emphasized how dysregulation of drinking behavior can be caused by alterations in self-monitoring and negative metacognitive beliefs about uncontrollability. Socialization was strengthened by the use of Socratic dialog (e.g., "If you discovered that you had control over your alcohol use how much of a problem would remain?") and the use of metaphors. At the end of TABLE 1 | Summary table of the MCT protocol for AUD.


the first treatment session Adaptive Self-Monitoring (ASM) was introduced as a method to discover the degree of control patients may have over their alcohol use. ASM is an attentional refocusing strategy that involves the orientation of attentional focus toward goal-progress information as it can give appropriate feedback to the cognitive system on when goals are reached, and ongoing drinking behavior can be moderated or stopped. This type of ASM is present in everyday life. For example, the monitoring of an appropriate highway exit to reduce our vehicle speed, change our route, and reach our destination, or the monitoring of cooking time and food appearance to define when to stop cooking. In the case of AUD, it implies focusing on global self and desired goals during alcohol use or simply counting the number of empty glasses on the table. ASM exercises were practiced in session to deliver appropriate information and feedback on selfregulation. Patients were then asked to freely practice ASM as homework.

In the following seven sessions, treatment focused on careful identification of which negative metacognitive beliefs about uncontrollability and/or danger were present and on modifying them. Negative metacognitive beliefs about uncontrollability and danger showed different facets: (1) uncontrollability of alcohol use ("I cannot stop using alcohol when I start"); (2) uncontrollability of thinking about alcohol use ("I cannot stop thinking about using alcohol"); (3) thought-action fusion ("Thoughts about alcohol will make me drink"); (4) abnormal brain beliefs ("I have no control over alcohol use because my brain is abnormal in some way"). Each of these metacognitive beliefs became the target of MCT interventions when present. The application of ASM, controlled drinking experiments, and verbal reattribution were adopted to modify beliefs about uncontrollability of alcohol use. The application of detached mindfulness techniques (Wells, 2009), postponement of perseverative thinking such as rumination, and verbal reattribution were used to modify beliefs about uncontrollability of thinking about alcohol use. Detached mindfulness, metacognitive delivered exposure to thoughts relating to alcohol use with response postponement and verbal reattribution were used to modify beliefs about thought-action fusion. Verbal reattribution, especially the examination of

evidence and counterevidence, and mini-surveys were used to modify abnormal brain beliefs.

In the next two sessions positive metacognitive beliefs about alcohol use became the focus of treatment. To counteract these beliefs an analysis of evidence and counterevidence was undertaken to reinforce knowledge about how the desired outcomes could be better achieved in other ways and behavioral experiments were applied to test this.

In the last two treatment sessions the intervention focused on relapse prevention and the further reappraisal of metacognitive beliefs. This included metacognitive beliefs about the meaning of lapses and relapses. Relapse prevention involved the construction of a replacement plan for situations where using alcohol may take place.

# Training

All patients were treated by the first author who is a Level-2 registered MCT therapist and received training and ongoing supervision in MCT from Professor Adrian Wells.

# Data Analysis and Clinical Significance

The primary goal of this case series was to determine if there is a clear treatment effect following the introduction of MCT. Typically, the visual examination of graphed data provides a reliable test of the treatment effect because only unambiguous effects are likely to be present (Parsonson and Baer, 1992). Weekly scores across baseline, treatment and follow-up on the QFS and metacognitive beliefs are presented in **Figure 1**. In addition, pre-treatment, post-treatment and follow-up scores on standardized measures for each of the five patients are presented in **Table 2**.

To determine whether a change over the course of treatment was clinically significant we adopted a two-fold criterion (Jacobson and Truax, 1991; Bauer et al., 2004). Following this method each patient was allocated to one of four outcomes: reliable deterioration, no change, reliable improvement or recovered. The first three outcomes are derived from the combination of different statistical approaches to reliable change. The Reliable Change Index (RCI, Bauer et al., 2004) approach, which determines whether the change is statistically significant, was applied to AUDIT-C scores. Data to calculate the RCI for the AUDIT-C score was drawn from a large sample of the general population (Aalto et al., 2009), and a minimum change of 3.46 points on AUDIT-C was consequently defined as a reliable change.

To be classified as recovered, patients would have had to demonstrate reliable change on their post-treatment or follow-up scores with these being below a clinical cut-off point for each of the primary outcome measures: (1) AUDIT-C; (2) QFS; and (3) number of DSM-5 criteria for AUD. With reference to the AUDIT-C score, different cut-offs have been established in different countries on the basis of sensitivity and specificity (Anderson et al., 2005): in Italy, total scores equal to or greater than five for men and four for women indicate possible hazardous consumption of alcohol (Struzzo et al., 2006). Data to establish a clinical cut-off for QFS was drawn from the recommendation of the Italian Ministry of Health that defines a safe weekly alcohol use of under 14 weekly units for men and 7 weekly units for women (Società Italiana di Nutrizione Umana [Sinu], 2014). Finally, for DSM-5 criteria for AUD none should have been met for at least 3 months but for less than 12 months (with the exception of craving). This was defined as an established threshold for early remission in line with the specifier for individuals previously diagnosed with AUD (American Psychiatric Association [APA], 2013).

# RESULTS

# Primary Outcomes

Weekly alcohol use appeared relatively stable during the baseline period for all patients (see **Figure 1**). Scores remained constantly above the limit of two standard deviations over the normative mean for the Italian population (Kehoe et al., 2012) as indicated by a normative comparison approach (Jacobson et al., 1986). The weekly mean number of binge drinking episodes at baseline was 3.3 (SD = 1.5) for patient 1, 1.5 (SD = 0.6) for patient 2, 2.4 (SD = 0.5) for patient 3, 3.4 (SD = 0.5) for patient 4, and 4.5 (SD = 0.8) for patient 5. When the treatment was introduced weekly alcohol use and related metacognitive beliefs significantly reduced for all patients. During treatment there were few binge drinking episodes for all patients. In this group of patients, the main effect of treatment appeared in the first half of the treatment, which focused upon using strategies to acquire a greater degree of control over alcohol use. These were maintained in the second half of treatment, which was more focused on the consolidation of new metacognitive knowledge and on relapse prevention strategies. All participants maintained their gains at post-treatment and follow-up, with a level of weekly alcohol use relatively unchanged to that was established during treatment with the exception of Patient 3 who experienced an increase in weekly levels of alcohol use at 6-months follow-up but remined at a lower level compared with the baseline. The levels of weekly alcohol use at posttreatment and follow-up were within one standard deviation of the normative data for the Italian population for all patients. No binge drinking episodes were reported at post-treatment and at 3- and 6- months follow-up. The treatment was welltolerated with no drop-outs and all patients reporting that it was helpful in gaining appropriate control over their alcohol use.

# Secondary Outcomes

The weekly measure of metacognitive beliefs did not change during baseline and showed a substantial reduction during treatment (see **Figure 1**). The decrease in the degree of conviction in metacognitive beliefs was quite rapid for Patients 2 and 5 after the beginning of the treatment. Patients 1 and 4 showed a more gradual decrease within the first half of treatment and remained stable in the second half. Patient 3 showed a constant decrease across treatment. These changes appeared stable at post-treatment and at 3- and 6- months follow-up. Scores for the PAMS and NAMS decreased at posttreatment and follow-up when compared to baseline scores



QFS, quantity frequency scale; AUDIT-C, alcohol use disorders identification test consumption; HADS, hospital anxiety and depression scale; PACS, penn alcohol craving scale; PAMS, positive alcohol metacognitions scale; NAMS, negative alcohol metacognitions scale.

and reached a level within one standard deviation of a nonclinical population as reported by Spada and Wells (2008) (See **Figure 2**). Scores on the PAMS and NAMS decreased, mirroring the weekly metacognitive beliefs measure changes. Similar results were replicated for anxiety, craving, and depression. Scores on HADS and PACS were lower at post-treatment and followup. Again, Patient 3 showed an increase in levels of craving at 6-month follow-up, but this remained lower compared to baseline.

# Clinical Significance

Each patient showed a reliable change for AUDIT-C with a change in score that ranged from 4 to 7 points. This reliable change was confirmed at 3- and 6- months follow-up for all patients with the exception of Patient 3 who reported an increase, but this remained stable when compared to pre-treatment scores. At post-treatment and follow-up all patients scored below the clinical cut-off for AUDIT-C and three patients reported weekly alcohol use below the QFS cut-off. Patient 5 showed a QFS over

the safe cut-off at post-treatment (QFS = 20), 3-months followup (QFS = 5) and 6-months follow-up (QFS = 16). Patient 3 showed QFS below cut-off at post-treatment and at 3-months follow-up but an increase in weekly alcohol use (QFS = 20) at 6-months follow-up. However, levels of weekly alcohol use remained significantly lower for Patients 3 and 5 when compared to pre-treatment QFS scores. None of the DSM-5 criteria for AUD were met at post-treatment and follow-up by Patients 1, 2, 4, and 5, while Patient 3 met one criterion for AUD. Taken together these findings indicate that Patients 1, 2, and 4 were classified as recovered while Patients 3 and 5 were classified as improved.

# DISCUSSION

The aim of this study was to evaluate the preliminary effects associated with MCT as a treatment for AUD. The outcomes of our study provide support the use of MCT as a therapeutic approach for AUD that may be associated with a clinically meaningful improvement in behavioral, cognitive, and affective self-regulation. Substantial reduction in weekly alcohol use and the absence of binge drinking episodes were observed for all patients compared to baseline. This change suggests an early remission from AUD for almost all patients. The reduction in symptoms appeared to remain stable in most cases at 3- and 6- follow-up.

Overall the treatment appears to have been successful and feasible with none of the patients reporting any worsening of psychological symptoms (anxiety and depression) and craving. Furthermore, our findings suggested that MCT might be a viable treatment for AUD as a primary diagnosis, at least with absence of physical withdrawal syndrome, especially when controlled drinking is an accepted or desired treatment goal.

Despite these encouraging results, several significant limitations of the current study need to be noted. Firstly, sample size was small, implying that the generalizability of measured effects should be considered with caution. Secondly, there was

# REFERENCES


no control condition and so it is was not possible to partial out time effects and non-specific factors from the treatment effects. Thirdly, the delivery of treatment by a single individual means we cannot determining the impact of therapist factors on outcomes. Fourthly, the use of self-report measurements may have led to overestimation of treatment effects. Finally, the study lacked any formal assessment of adherence to treatment.

Overall, the outcome in this case series suggests that MCT is a feasible tretament with AUD and appears to be associated with reduced problematic drinking and increased control (at least in the short term). Future studies of MCT for AUD with larger samples and randomized designs are recommended in order to determine whether this approach is efficacious and whether it may provide an alternative to existing treatments.

# ETHICS STATEMENT

This study was carried out in accordance with the recommendations of LSBU code of practice for research with Human Participants, LSBU Research Ethics Committee (UREC 1503); with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the LSBU Research Ethics Committee (UREC 1503).

# AUTHOR CONTRIBUTIONS

GC designed the research and treatment protocol, run therapy sessions, analyzed the data, and wrote the manuscript. FM designed the research protocol, assessed participants, and analyzed data. MS designed the research and treatment protocol, collected and analyzed data, and revised the manuscript. AW created theoretical foundation of the treatment, designed the research and treatment protocol, and revised the manuscript.




adattamento del questionario AUDIT e verifica dell'efficacia d'uso dello short-AUDIT test nel contesto nazionale. Boll. Farmacodipendenze Alcolismo 29, 20–25.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Caselli, Martino, Spada and Wells. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Metacognitive Therapy for Adjustment Disorder in a Patient With Newly Diagnosed Pulmonary Arterial Hypertension: A Case Report

*Lotta Winter1 \*, Franziska Naumann1 , Karen Olsson2 , Jan Fuge2 , Marius M. Hoeper <sup>2</sup> and Kai G. Kahl1*

*1Department of Psychiatry, Social Psychiatry and Psychotherapy, Hannover Medical School, Hannover, Germany, 2Department of Pneumology, Hannover Medical School and German Centre for Lung Research (DZL), Hannover, Germany*

## *Edited by:*

*Hans M. Nordahl, Norwegian University of Science and Technology, Norway*

#### *Reviewed by:*

*Karin Carter, Greater Manchester Mental Health NHS Foundation Trust, United Kingdom Maria C. Quattropani, University of Messina, Italy*

> *\*Correspondence: Lotta Winter winter.lotta@mh-hannover.de*

#### *Specialty section:*

*This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology*

*Received: 24 May 2019 Accepted: 20 January 2020 Published: 12 February 2020*

#### *Citation:*

*Winter L, Naumann F, Olsson K, Fuge J, Hoeper MM and Kahl KG (2020) Metacognitive Therapy for Adjustment Disorder in a Patient With Newly Diagnosed Pulmonary Arterial Hypertension: A Case Report. Front. Psychol. 11:143. doi: 10.3389/fpsyg.2020.00143*

Adjustment disorders (ADs) belong to the worldwide most diagnosed mental disorders and are particularly frequent in patients with an underlying physical illness. Pulmonary arterial hypertension (PAH) is a severe and disabling disease, which significantly impacts on quality of life and has high mortality rates. The authors here present the case of a young female who developed a severe adjustment disorder with both anxious and depressive symptoms after a diagnosis of PAH requiring intensive care treatment due to right heart failure. Psychosocial functioning was severely impaired, and physical health reduced. Following hemodynamic stabilization and the establishment of PAH treatment, the patient was admitted to the Department of Psychiatry, Social Psychiatry and Psychotherapy and received metacognitive therapy (MCT). AD with mixed anxiety and depressed mood was diagnosed according to DSM-V criteria. At the start of treatment, she reported significant mental distress, indicated by a total sum score of the Hospital Anxiety and Depression Scale (HADS) of 20 points. The 6-min walking distance was only 358 m before the patient was exhausted. She then was treated with MCT without further psychopharmacological drugs. After only four MCT sessions, she fully remitted from AD which was accompanied by an 11-point reduction in the HADS (to 9 points). MCT specific scores also improved (MCQ-30 sum score decreased from 77 to 35). Notably, physical capacity improved as well, documented by an improved walking distance (439 m; +22%). This is the first case of a patient with AD in the context of PAH treated with MCT. The case report suggests that MCT is a possible psychotherapeutic treatment option for AD in the context of a potentially life-threatening disease. The study design does not permit an attribution of outcome to MCT but it suggests MCT is a potentially viable and acceptable treatment option.

Keywords: metacognitive therapy, adjustment disorder, pulmonary arterial hypertension, psychotherapy, PAH, MCT

# INTRODUCTION

Adjustment disorder (AD) represents an abnormal stress response that is different from normal adaptive reactions (Casey, 2014). According to DSM-V, AD is characterized by: (A) emotional or behavioral symptoms in response to an identifiable stressor that (B) are of clinical significance and (C) do not meet the criteria for another mental disorder, and (D) do not represent normal bereavement. Typically, AD remits within 6 months if the stressor is terminated; however, a persistent form of AD has been described if the stressor persists (First, 2013; Maercker and Lorenz, 2018). Furthermore, untreated AD poses the risk of persistent AD, and may pave the way for psychiatric disorders other than AD, particularly major depressive disorder and anxiety disorders (O'Donnell et al., 2016).

Epidemiological data are scarce since none of the major epidemiological studies included adjustment disorders among the conditions examined (Myers et al., 1984; Jenkins et al., 1997; Jacobi et al., 2004, 2014; Kessler et al., 2005). However, AD is reported to be common in primary care where rates of the disorder range from 1 to 18% (Casey et al., 1984; Blacker and Clare, 1988), and is also common in elderly persons as shown in a representative community survey (2.3%) (Maercker et al., 2008).

AD has been reported to be almost three times as common as major depression in acutely ill patients (13 versus 5%) (Silverstone, 1996). In potentially life-threatening diseases such as cancer, AD rates as high as 19.4% have been described, and AD has been observed in 15.4% of patients receiving palliative care (Mitchell et al., 2011). In up to one-third of breast cancer patients experiencing recurrence of their cancer, AD has been reported (Okano et al., 2001).

Pulmonary arterial hypertension (PAH) is a rare condition characterized by pulmonary vascular remodeling leading to right heart failure and death. Untreated, the estimated median survival of PAH was 2.8 years (D'Alonzo et al., 1991). Although treatment options have been improved during the last 20 years, PAH treatment is challenging: First, available treatments only reduce the progression of the disease course. Second, patients experience massive physical restrictions, leading to dyspnea, fatigue, exercise-induced syncope, suffocation, and edema, leading to decreased quality of life and decreased social functioning. Third, mortality is still high (3-year survival 70–80%) and for some patients lung transplantation remains the only treatment option. AD may pose further burdens on the patients, reducing their quality of life and psychological well-being (Hoeper et al., 2013; Galie et al., 2016).

Increased levels of anxiety and depression symptoms, and decreased quality of life have been observed in PAH, although AD has not been described so far (Larisch et al., 2014; Somaini et al., 2016). Divergent treatments such as cognitive behavioral therapy and low-intensity psychological interventions (self-help therapy, bibliotherapy, support groups, behavioral activation, mindfulness, meditation, relaxation, and e-mental health interventions) have been proposed for the treatment of AD, and three broader common components of these divergent strategies have been identified: (1) the enabling of individuals to reduce or remove the stressor, (2) interventions to improve coping with the stressor, and (3) stress reduction strategies (O'Donnell et al., 2018). However, to date there is only limited empirical evidence for these treatments in AD and a need for further studies and replication studies evaluating the efficacy of specific interventions in patients with AD have been proposed (O'Donnell et al., 2018). An alternative approach might be to take a more theory-driven perspective and modify the mechanisms that contribute to abnormal stress reactions. Research stimulated by the metacognitive model (Wells and Matthews, 1994) implicates metacognitive beliefs, worry and rumination in the maintenance and exacerbation of stress responses (e.g., Wells and Papageorgiou, 1995), and is supported by evidence that thought control strategies such as worry predict PTSD (e.g., Holeva et al., 2001). This psychological approach (MCT, Wells, 2009) has been found to be effective in both psychological and physical health contexts. The outcome of MCT in cardiac rehabilitation patients (Wells et al., 2018) is currently being evaluated, however, there is evidence of an association between metacognitive beliefs and psychological distress in other health conditions (e.g., Fisher et al., 2017). Overall, the therapy strategies used in MCT possibly prove a good fit to emotional distress in cardiac patients (McPhillips et al., 2018). We therefore examined if using MCT to treat a patient suffering from severe AD in the context of PAH was feasible and associated with symptom reduction.

# CASE PRESENTATION

# Biography

The patient described in this case report is a 34-year-old woman with the diagnosis of hereditary PAH. Several male family members on her father's side had succumbed to the same disease around the age of 35 years. Her mother suffers from depression, and one brother has panic attacks. After finishing high school, the patient became a professional and worked in various physical health fields. She lives together with her boyfriend and in 2017 gave birth to her first child.

# Symptoms

The patient reported dyspnea on exertion after giving birth to a healthy child in 2017. However, despite this fact and the above-mentioned family history, diagnosis of PAH was not made until December 2018 when she was admitted to our hospital as an emergency with right heart failure after pulmonary infection. She recovered with supportive measures and introduction of PAH treatment with macitentan, an endothelin receptor antagonist, and tadalafil, a phosphodiesterase-5 inhibitor. When she returned home, she continued to experience severe limitations in everyday situations. For example, she was not able to carry her child as she felt too weak. In addition, she was afraid of any illness her child could infect her with. She felt incapable of looking after her child on her own and was dependent on other people's support. As soon as she experienced signs of being ill, she went to specialists for check-ups. She was very quickly physically exhausted and had an increased need to sleep. Her situation led to intensive worrying about her self-image, her future, and her health. She cried more than before and experienced panic attacks several times per week. She feared her death and felt guilt toward her family members. Further on she repeatedly kept comparing her current situation to how it was before she was diagnosed, which led to despair and hopelessness. Her everyday life was dominated by anxiety, safety behaviors, and despair. She could hardly be by herself and was dependent on reassurance from others. She was grateful for the internal specialist's referral to the department of psychiatry to seek help.

Consultation by the department of psychiatry resulted in the diagnosis of a severe adjustment disorder and she was registered for treatment.

# Assessment

In February 2019, she had a first appointment at the Department of Psychiatry, Social Psychiatry and Psychotherapy for a diagnostic assessment, and inpatient treatment started eventually in March. Treatment duration was 4 weeks. At both time points before treatment (T0a: diagnostic assessment, T0b: day of hospitalization) as well as at the end of treatment (T1) and 6 weeks after treatment ended (T2) she completed a set of questionnaires including the Hospital Anxiety and Depression Scale (Herrmann-Lingen et al., 2011), and the Metacognition Questionnaire (Wells and Cartwright-Hatton, 2004). At the start of treatment, she reported significant mental distress, indicated by a sum score of the anxiety subscale of the HADS of 13 points, and 7 points on the HADS depression subscale. These scores indicated that anxiety was severe and predominant. The scores of the MCQ-30 show that negative beliefs about uncontrollability and danger of worry were strongest. All scores are presented in **Table 1**. The patient gave written informed consent for the publication of this case report.

# Treatment

For the treatment with Metacognitive Therapy (MCT), the manual (Wells, 2009) was followed. During 4 weeks of inpatient treatment, the patient received weekly MCT sessions lasting 50 min each. In the first session, a personalized case formulation was developed using the generic model (Wells, 2009), which is presented in **Figure 1**. The patient was socialized to the model and the role of the cognitive attentional syndrome (CAS) was illustrated. Further, the patient was asked to rate the intensity of individual positive and negative metacognitive beliefs (**Table 2**). In the second session, Attention Training Technique (ATT; Wells, 1990) was introduced by using the German version of the audio file and the self-attention rating scale. Further, detached mindfulness was introduced by using the phone metaphor. For homework the patient was asked to do the ATT twice a day and practice worry postponement whenever her CAS was activated. In session number three, detached mindfulness was practiced again using the free association task several times. After the second repetition, subjectively difficult words were included in the task. In the beginning of the third session, the patient was also asked to rate the metacognitive beliefs formulated and rated in the first session (**Table 2**). The individual positive and negative metacognitive beliefs had already decreased to almost 0% and so no further challenging was undertaken. In the fourth session with the use of the "old plan – new plan" protocol was used to consolidate the change of strategies and attentional focus and metacognitive beliefs formulated in session 1 were again checked (**Table 2**). The patient was asked to repeat ATT after discharge for another 4 weeks.

Assessment of metacognitive beliefs using the MCQ-30 (**Table 1**) demonstrated a significant reduction in positive and negative metacognitive beliefs, and a significant reduction in maladaptive coping strategies including all elements of the CAS. This improvement was accompanied by a reduction in symptoms of anxiety and depression assessed with the HADS (**Table 1**). Interestingly, we also found an improvement in physical symptoms. As part of the routine assessment

TABLE 1 | Metacognitive beliefs (MCQ-30) and symptoms of anxiety/depression (HADS) according to self-rating scales, and walking distance over the course of MCT treatment and after 6-week follow-up.


*Initial HADS scores pointed to severe mental health problems according to AD. HADS scores for anxiety and depressive symptoms reduced to subthreshold levels after 4 sessions MCT, and even more declined after 6wk follow-up. Metacognitions also improved > 50% and remained stable at 6wk follow-up. Of note, physical capacity markedly improved by 23% measured by an increased walking distance. AD, adjustment disorder; PAH, pulmonary arterial hypertension; HADS, hospital anxiety and depression scale; MCQ-30, metacognitions questionnaire30; POS, positive beliefs about worry; NEG, negative beliefs about uncontrollability and danger of worry; CC, cognitive confidence; NC, need for control; CSC, cognitive self-consciousness.*

TABLE 2 | Rating of the patient's individual metacognitive beliefs in each indicated MCT session.

of the patient.


for patients with PAH, walking distance is regularly measured, and the patient had a 23% increase in walking distance after the end of MCT treatment (**Table 1**). At 6 week follow-up, results concerning metacognitive beliefs and maladaptive coping strategies remained stable, while assessment of anxiety and depression symptoms revealed further improvement (**Table 1**).

It can be reported that the treatment was well tolerated by the patient and no adverse effects could be identified.

# DISCUSSION

Our case report is notable in two ways: first, this is the first description of adjustment disorder as a consequence of a PAH diagnosis. Second, MCT was used for the first time to address AD in a PAH patient, without any further psychopharmacological medication. We chose this approach since according to the metacognitive theory, modifying the mechanisms that contribute to the development and maintenance of mental distress may improve AD.

Psychological distress has been associated with positive and negative metacognitive beliefs in a range of diseases including cancer (Thewes et al., 2013; Quattropani et al., 2017), Parkinson's disease (Brown and Fernie, 2015), epilepsy (Fisher and Noble, 2017), chronic fatigue syndrome (Maher-Edwards et al., 2012), fibromyalgia (Kollmann et al., 2016), multiple sclerosis (Quattropani et al., 2018), and diabetes (Purewal and Fisher, 2018).

If the stress response is abnormal, meaning that it is out of proportion given the intensitiy of the stressor and followed by significant psychosocial impairment, AD can be diagnosed. AD is characterized by cognitive preoccupation with the disease itself, and its imagined consequences for one's life and for significant others, resulting in emotional symptoms such as anxiety and depression, and in avoidance behaviors.

Psychotherapeutic and psychopharmacological treatment options for AD have recently been summarized in three review articles (Domhardt and Baumeister, 2018; O'Donnell et al., 2018; Stein, 2018). They found that the quality of evidence has been ranked low to very low (O'Donnell et al., 2018).

McPhillips et al. (2018) provide qualitative data on why MCT might be more effective than cognitive behavioral therapy (CBT) in the treatment of emotional distress in cardiac patients, although the validity of this hypothesis still needs to be shown. Still, a possible reason may be that content-related strategies like making a distinction between realistic and unrealistic thoughts may leave too much room for prolonged processing and may be ambiguous. A further aspect why the CBT model might be a poor fit is that patients describe diverse realistic negative automatic thoughts encompassing not only physical disease but also other areas of their lives (McPhillips et al., 2018). The perspective of MCT opens the opportunity to address emotional distress without analyzing the content of thoughts, which is often contradictory in AD. Perseverative thinking and underlying metacognitive beliefs can be targeted independently of realistic or unrealistic contents. A further advantage of MCT is its relatively short duration (Normann and Morina, 2018). In general, in CBT, more sessions are needed and the content related strategies used reach their limits in the treatment of AD.

In our patient, the metacognitions "Checking helps to keep me safe" or "Worrying helps me to be prepared" appeared to maintained dysfunctional coping strategies like threat monitoring, worrying, body scanning etc. and therefore preserved experiences of anxiety and insecurity. Over the course of treatment, the conviction in these metacognitions decreased. At the end of treatment, the patient reported new metacognitions like: "You can never be safe, so fighting for safety is useless" and "My body will tell me if it needs attention." The change of metacognitions was accompanied by a decrease in both the anxiety and depression subscale of HADS. According to the metacognitive model, psychological disorders persist because of the effects of a state of thinking, the CAS, on emotional experiences and knowledge (Wells, 2009, p. 721). The CAS is controlled by positive and negative metacognitive beliefs. According to Wells (2009), this presents a range of possibilities for treatment that focus on removing the CAS, modifying metacognitive beliefs, and developing alternative ways of experiencing and relating to inner events (p. 729). In our case, the patient was introduced to the experience of being able to detach from her negative thoughts, reduce her CAS and apply her attention in a more flexible way. These experiences as well as the therapeutic style of addressing her concerns, e.g., with the use of the metacognitive socratic dialogue, lead to a change in metacognitive beliefs as indicated by the scores of the MCQ-30 and the ratings of her individual beliefs. Further modification of metacognitive beliefs was associated with a reduction of clinical symptoms indicated by a decrease of the HADS scores.

An interesting finding belongs to the improvement in physical parameters, i.e., greater walking distance. One can interpret this finding in two ways, either "psychological" as improved confidence of the patient in her physical capacity, or "somatic" as improved physical functioning once the psychological distress was reduced.

Our patient tolerated the intervention and gave positive feedback that she felt well understood and received what she needed. According to McPhillips et al. (2019), the psychological needs of cardiac rehabilitation patients can include the wish not to disclose their concerns. Therefore content-focused therapy like CBT may not be tolerated by these patients. In such cases MCT which is process-focused and allows patients to keep the content of thoughts private may have greater acceptance than other interventions.

Psychopharmacological treatment has also been discussed in AD. Treatment options include the use of benzodiazepines (alprazolam, diazepam, clorazepat, lormetazepam), antidepressants (mianserin, tianeptine, trazodone, viloxazine), plant extracts/ herbals (euphytose, Ginko-Biloba, Kava-Kava), anxiolytics (etifoxine), 5-HT1A agonists (buspiron), and neutraceuticals (s-adenosylmethionine). However, only 11 randomized-controlled trials including 1,195 AD patients have been documented, yielding in part contradictory results (Stein, 2018). Furthermore, psychopharmacological treatment may be limited by drug-induced side effects, including pharmacokinetic and pharmacodynamics alterations such as drug–drug interactions, induction/inhibition of the cytochrome-P-450 system, or indirect drug effects that all may interact with the drugs necessary for treating the underlying disease (Huang et al., 2008; Kahl, 2018; Kahl et al., 2019). This particularly applies to patients with cardiorespiratory impairment (Kahl et al., 2017, 2018). Therefore, nonpharmacological treatments may be favored in patients who develop AD in the context of an underlying physical illness.

# Limitations

As this was an important inpatient treatment, other aspects of being on a psychiatric ward may have influenced the outcome. We rate this factor as low as she was the only patient with AD within the group of patients and was excluded from other forms of psychological interventions. A further limitation can be seen in the lack of a direct evaluation of the therapy process. Further, in the absence of a control condition we cannot rule out a placebo or a therapist effect. Unfortunately, we did not explicitly measure the CAS with the use of a questionnaire. However, according to the metacognitive model, we expect that changing metacognitive beliefs should lead to reduced CAS activity. In accordance with this, decreased worry intensity and less threat monitoring were reported by the patient. Another indicator for reduced CAS activity may be seen in reduced clinical symptoms. In future, a larger sample explored *via* single case or trial methodology is needed to investigate the use and effectiveness of MCT in the treatment of AD in the context of an underlying physical disorder.

# CONCLUSION

We conclude that MCT might be promising for patients with AD with an underlying physical disorder. Future studies examining acute and sustained effects of MCT in patients with AD in the context of different physical disorders are warranted.

# DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

# ETHICS STATEMENT

The studies involving human participants were reviewed and approved by Ethics Committee Hannover Medical School. The patients/participants provided their written informed consent to participate in this study. Written informed consent was

# REFERENCES


obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

# AUTHOR CONTRIBUTIONS

LW was in charge of the therapy plan, was the therapist treating the patient, and wrote the manuscript. FN served as a co-therapist and also carried out the psychological assessments. KO and MH were the attending physicians for any somatic concern. Further on they provided training for the other authors on PAH and wrote parts of the manuscript. JF served as a co-physician and carried out the somatic assessments. KK supervised the treatment and wrote the manuscript.

# ACKNOWLEDGMENTS

We wish to thank the patient for allowing us to report on her treatment.


disorders in old age: findings from a community survey. *Compr. Psychiatry* 49, 113–120. doi: 10.1016/j.comppsych.2007.07.002


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2020 Winter, Naumann, Olsson, Fuge, Hoeper and Kahl. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

fpsyg-10-00162 January 29, 2019 Time: 16:58 # 1

# Brief Metacognitive Therapy for Emotional Distress in Adult Cancer Survivors

Peter L. Fisher1,2,3 \*, Angela Byrne1,2,3, Louise Fairburn1,2,3, Helen Ullmer1,2,3 , Gareth Abbey<sup>1</sup> and Peter Salmon1,2,3

<sup>1</sup> Department of Psychological Sciences, University of Liverpool, Liverpool, United Kingdom, <sup>2</sup> Liverpool Clinical Health, Royal Liverpool and Broadgreen University Hospital NHS Trust, Liverpool, United Kingdom, <sup>3</sup> Psychology Service, Royal Liverpool and Broadgreen University Hospital NHS Trust, Liverpool, United Kingdom

Background: Adult cancer survivors often experience substantial psychological morbidity following the completion of acute cancer treatment. Unfortunately, current psychological interventions are of limited efficacy. This study explored if metacognitive therapy (MCT); a brief transdiagnostic psychological intervention was potentially efficacious and could be delivered effectively to adult cancer survivors with psychological morbidity.

#### Edited by:

Lora Capobianco, Manchester Mental Health and Social Care Trust, United Kingdom

#### Reviewed by:

Maria C. Quattropani, Università degli Studi di Messina, Italy Ana Nikcevic, Kingston University, United Kingdom

> \*Correspondence: Peter L. Fisher peter.fisher@liverpool.ac.uk

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 28 October 2018 Accepted: 17 January 2019 Published: 31 January 2019

#### Citation:

Fisher PL, Byrne A, Fairburn L, Ullmer H, Abbey G and Salmon P (2019) Brief Metacognitive Therapy for Emotional Distress in Adult Cancer Survivors. Front. Psychol. 10:162. doi: 10.3389/fpsyg.2019.00162 Methods: An open trial with 3- and 6-month follow-up evaluated the treatment effects of MCT in 27 consecutively referred individuals to a clinical psychology health service specializing in psycho-oncology. Each participant received a maximum of six 1-hour sessions of MCT. Levels of anxiety, depression, fear of cancer recurrence, post-traumatic stress symptoms, health related quality of life, and metacognitive beliefs and processes were assessed using self-report questionnaires.

Results: MCT was associated with statistically significant reductions across all outcome measures which were maintained through to 6-month follow-up. In the ITT sample on the primary treatment outcome measure, the Hospital Anxiety and Depression Scale-Total, 59% of participants met recovery criteria at post-treatment and 52% at 6-month follow-up, respectively. No participants significantly deteriorated. In the completer sample (N = 20), 80% recovered at post-treatment and 70% at 6-month follow-up. MCT was acceptable to patients with approximately 75% of patients completing all treatment sessions.

Conclusion: MCT, a brief transdiagnostic psychological intervention can be delivered effectively to a heterogenous group of cancer survivors with promising treatment effects. Examining the efficacy of brief MCT against the current gold standard psychological intervention would be a valuable advance toward improving the quality of life of cancer survivors.

Keywords: cancer, survivors, emotional distress, metacognitive therapy, open trial

# INTRODUCTION

fpsyg-10-00162 January 29, 2019 Time: 16:58 # 2

The incidence of cancer in the United Kingdom is projected to increase by 2% over the next 15 years with survival rates also increasing. It is estimated that survival rates have doubled over the past 40 years with a ten-year survival rate of approximately 50% (Cancer Research UK, 2017) in 2016, there were an estimated 15.5 million cancer survivors which is expected to increase to 20.3 million by 2026 (National Cancer Institute, 2018). Psychological morbidity is common in cancer survivors. Approximately 25% of cancer survivors have clinically significant levels of anxiety and depression that could benefit from treatment (Hoffman et al., 2009). Posttraumatic stress disorder symptoms are common in cancer survivors with estimates ranging from 6 to 45% (Swartzman et al., 2017). Fear of cancer recurrence (FCR) is highly prevalent, a systematic review concluded that almost 60% of cancer survivors experience debilitating FCR (Simard and Savard, 2015). Psychological morbidity adversely impacts ongoing cancer care by reducing attendance at follow up screening appointments (DiMatteo et al., 2000; Thewes et al., 2014), health related quality of life (LeMasters et al., 2013) and increases healthcare costs (Carlson and Bultz, 2004; Jansen et al., 2016) and use of healthcare services (Elliott et al., 2011).

The substantial prevalence and associated problems with psychological morbidity in cancer survivors requires effective interventions. Unfortunately, highly efficacious psychological interventions are unavailable (Rehse and Pukrop, 2003; Osborn et al., 2006; Faller et al., 2013). The most widely evaluated and recommended psychological intervention is cognitive behavioral therapy (CBT) but it may be that core components of CBT; labeling cognitive distortions and reality testing negative automatic thoughts (NATs) are clinically limited where NATs will frequently reflect accurate thoughts about cancer recurrence and morbidity (Greer et al., 2010; Cook et al., 2015b). An intervention which does not need to focus on the content of cognition i.e., NATs, but instead focuses on core psychological processes underpinning psychological morbidity may be more efficacious for cancer survivors.

Metacognitive therapy (MCT; Wells, 2009) offers an alternative psychological approach to the treatment of psychological morbidity in cancer survivors. MCT is derived from a trans-diagnostic theory of psychopathology, the Self-Regulatory Executive Function (S-REF) model (Wells and Matthews, 1994, 1996). The model states that psychological morbidity becomes persistent when people use the cognitive-attentional syndrome (CAS) in response to unwanted thoughts. The CAS has three broad main components; (i) perseveration (worry, rumination, over-analyzing, repeatedly questioning one's thoughts); (ii) attentional strategies (a heightened focus on possible signs of threat which can be internal e.g., signs of anxiety or external e.g., reminders of cancer); and (iii) unhelpful coping strategies (e.g., searching the internet for positive outcomes by cancer survivors, avoidance of reminders of cancer).

The S-REF model states that perseveration is guided by positive metacognitive beliefs about the helpfulness of worry and rumination: e.g., "worry will help me be better prepared," "worry will ensure that I complete my daily tasks." Unfortunately, worry and rumination achieve the opposite, because the person experiences more negative thoughts and views more situations as potentially dangerous. The individual repeatedly acts as if unwanted negative thoughts are meaningful which leads to the development of an inflexible way of responding to thoughts. A more flexible response style can help to alleviate perseveration. Similarly, the S-REF model specifies that threat monitoring (e.g., scanning for symptoms or for negative thoughts) is determined by positive metacognitive beliefs. More specifically, a person comes to believe that scanning the environment or one's mind and/or body for symptoms will reduce distress whereas it leads to the persistence of threat and distress. Furthermore, negative metacognitive beliefs about the uncontrollability and danger of worry sustain and increase worry. Modifying negative metacognitive beliefs is fundamentally important in the S-REF model because, if patients believe that worry is uncontrollable, they will not attempt to control it. Therefore, it is possible that through targeting metacognitive beliefs and processes rather than cognitive content, MCT offers a particularly close "fit" with the needs of cancer survivors indicating potential for greater efficacy (McNicol et al., 2013).

The development of MCT for psychological morbidity in cancer is evolving with encouraging evidence for the explanatory and therapeutic utility of MCT. There is increasing evidence for the role of metacognitive beliefs and processes in emotional distress in cancer survivors from cross-sectional and prospective studies (Thewes et al., 2013; Cook et al., 2014, 2015a,b; Butow et al., 2015; Fisher et al., 2018) and in adult cancer patients undergoing chemotherapy (Quattropani et al., 2016, 2017). There have been two tests of the potential efficacy of MCT in cancer survivor. First, an open trial of MCT for emotional distress in adolescent and young adult cancer survivors found clinically significant reductions in anxiety, depression and posttraumatic stress symptoms (Fisher et al., 2015). Second, a multiple baseline study of MCT in four adult cancer survivors (Fisher et al., 2017) reported substantial reduction in anxiety, depression and FCR over six one-hour sessions These studies illustrate that MCT can rapidly alleviate psychological morbidity in cancer patients but before progressing to a randomized controlled trial, further evidence of the potential efficacy and feasibility of delivering MCT is required. The present study therefore examined if MCT delivered over six one-hour individual treatment sessions would result in clinically significant improvements in anxiety, depression, posttraumatic stress symptoms, fear and cancer recurrence and overall quality of life immediately following treatment and over a 6-month follow-up period. The study also examined if MCT would be associated with reductions in the metacognitive beliefs and processes.

# MATERIALS AND METHODS

# Design

An open trial with follow-up at 3 and 6 months evaluated the potential efficacy of brief MCT for adult survivors of cancer experiencing emotional distress. Data was also gathered on recruitment and retention rates. All participants gave written informed consent in accordance with the Declaration of Helsinki. Ethical approval was provided by the National Health Service North West Research Ethics Committee (reference 15/NW/0820).

# Participants and Procedure

fpsyg-10-00162 January 29, 2019 Time: 16:58 # 3

Potentially suitable participants were identified from consecutive referrals to an adult clinical heath psychology service which specializes in psychological interventions for cancer patients. Those patients with elevated scores on the Hospital Anxiety and Depression Scale (HADS; Zigmond and Snaith, 1983) and indicated a willingness to be approached for possible participation in an intervention were provided with an information sheet about the study. Those patients were contacted and invited to attend an assessment appointment to determine their suitability for inclusion. Following the informed consent procedure, clinical and demographic data was obtained by interview and participants completed a range of questionnaires assessing the severity of psychological morbidity (see section on measures). Participants also completed all questionnaires at post-treatment, and again at 3- and 6-month follow-up. All questionnaires were returned to an independent assessor who scored and entered the data.

Twenty seven cancer survivors participated in the study and met the following inclusion criteria: (i) a score of > 15 on the Hospital Anxiety and Depression Scale-Total (HADS-T); (ii) had been diagnosed with cancer ≥ 6 months previously; (iii) were aged 18 years or over; (iv) had completed acute medical treatment for cancer (i.e., chemotherapy, radiotherapy, surgery); (v) were not receiving concurrent psychological treatment; (vi) were not actively suicidal; (vii) reported no current substance use; (vii) were not experiencing a psychotic or organic illness; (viii) were free from psychotropic medication or has been on a stable dose for at least 8 weeks; and (ix) were able to speak and understand English.

# Intervention

Metacognitive therapy was delivered over a maximum of 6 individual face-to face sessions that were 45–60 min in duration. The intervention followed a manualized protocol (Wells, 2009). As the intervention was transdiagnostic, MCT followed the same protocol for each patient in the study regardless of symptom presentation. In session 1, the formulation template used when treating depression served as the basis for the development of an idiosyncratic case formulation for each participant, thus following the approach adopted in previous evaluations of MCT for cancer survivors (McNicol et al., 2013; Fisher et al., 2015, 2017). The next step in treatment is socialization which proceeds by sharing the case formulation and by Socratic Questioning to help the patient understand that each aspect of the CAS and several types of metacognitive beliefs are maintaining emotional distress. MCT then focuses on modifying negative beliefs about uncontrollability of rumination/worry through training in detached mindfulness (DM) and in rumination/worry postponement (Wells, 2009). Patients are helped to understand how naturally occurring thoughts (e.g., "I'm useless," "What if my cancer comes back?," "My family will not be able to cope") do not necessarily lead to perseveration.). Rumination/worry postponement is a behavioral experiment to challenge the negative metacognitive belief that perseveration is an uncontrollable process. Positive metacognitive beliefs about the helpful nature of worry/rumination and the other unhelpful coping responses are also highlighted to the patients and addressed. Final sessions address relapse prevention and involve modifying remaining use of the "cognitive attentional syndrome," reviewing any remaining conviction in positive and negative metacognitive beliefs and consolidating and alternative ways of responding to negative thoughts. Three therapists delivered MCT (PF, AB, and LF). Supervision was provided by PF on a weekly basis.

# Measures

## Hospital Anxiety Depression Scale (HADS; Zigmond and Snaith, 1983)

The HADS is a 14-item self-report questionnaire measuring anxiety and depression (seven items each) over the past week. Each item is rated on a 4-point scale (0–3). Scores for each subscale range from 0 to 21 with higher scores reflecting more sever anxiety or depression. Scores of 11 or more on each of the subscales indicate caseness. Combining the two subscales provides a measure of emotional distress. The HADS-Total is the "gold standard" outcome measure for evaluating the efficacy of interventions on emotional distress in cancer populations, and has excellent psychometric properties (Luckett et al., 2010).

### Impact of Events Scale-Revised (IES-R; Weiss, 2007)

The IES-R is a 22-item self-report questionnaire measuring trauma-related symptoms The total scale score ranges from 0 to −88 with higher scores indicative of more severe trauma symptoms. A total score of ≥ 33 indicates a probable diagnosis of PTSD (Weiss, 2007). The IES-R is validated for use in cancer populations with good psychometric properties (Salsman et al., 2015).

## Fear of Cancer Recurrence Inventory (FCRI; Simard and Savard, 2009)

The FCRI is 42-item self-report questionnaire assessing 7 aspects of FCR. Each item is rated on a 5-point scale (0–4). A total score for the FCRI is obtained by summing scores on the 7 subscales, with higher scores indicating greater severity (range 0–168). The FCRI is the most validated measure of FCR across a wide range of cancer types (Simard and Savard, 2009).

## Functional Assessment of Cancer Therapy-General (FACT-G; Cella et al., 1993)

The FACT-G is a 27 item self-report questionnaire that measures four domains of health-related quality of life (HRQOL). Each item is rated on a 5-point scale from 0 (not at all) to 4 (very much). The FACT-G total score ranges from 0 to 108 with higher scores indicating a better HRQOL. The FACT-G has been used extensively in mixed cancer populations and has excellent psychometric properties (Brucker et al., 2005).

fpsyg-10-00162 January 29, 2019 Time: 16:58 # 4

# Metacognitions Questionnaire-30 (MCQ-30; Wells and Cartwright-Hatton, 2004)

The MCQ-30 measures 5 domains of metacognition by 30 items. Participants rate the extent to which they "generally agree" with statements presented on a 4-point scale from 1 (do not agree) to 4 (agree very much), providing total scores for each subscale ranging from 6 to24. Higher scores indicate greater conviction in metacognitive beliefs. The MCQ-30 assesses: (1) positive beliefs about worry, (2) negative beliefs uncontrollability and danger of worry, (3) cognitive confidence, (4) beliefs about the need to control thoughts, and (5) cognitive self-consciousness. The MCQ-30 has been validated for use in cancer patients (Cook et al., 2014).

## Cognitive Attentional Scale-1 (CAS-1; Wells, 2009)

The CAS-1 is a 10 item self-report questionnaire that assesses metacognitive processes and beliefs. Items 1 to 6 assess the fundamental components of the CAS (perseverative thinking, threat monitoring and unhelpful coping strategies) Each item is rated on a 10-point scale from 0 (none of the time) to 100 (all the time). Items 7 to 10 assess metacognitive beliefs and are not reported in the present study. To provide an overall measure of the CAS, the 6 items were summed and divided by the number of items. The same method has been used previously (Fisher et al., 2016; Heffer-Rahn and Fisher, 2018).

# Statistical Analyses

Intention to treat (ITT) analyses were used to determine the potential efficacy of brief MCT for emotional distress in cancer survivors. Missing data for the non-completers in the study were replaced by using the last observation carried forward (LOCF) method. The LOCF has been considered a conservative approach when evaluating treatment outcomes in open trials. Treatment effects across time (pre-treatment, post-treatment, and 3- and 6-month follow-up) were assessed with repeated-measures analysis of variance (ANOVA); the Greenhouse–Geisser correction was applied when the assumption of sphericity was violated. Main effects were followed by Bonferroni-adjusted pairwise comparisons for each outcome measure. Within group effect sizes were calculated using Cohen's d to assess the magnitude of treatment effects from pretreatment to post-treatment and from pre-treatment to both 3- and 6-month follow-ups. To determine the clinical significance of treatment effects the methodology developed by Jacobson et al. (1984) and Jacobson and Truax (1991) was applied to the HADS-Total. Each patient can be allocated to one of four treatment outcomes: reliable deterioration, no change, reliable improvement, or recovered. The first three outcomes are calculated using from the Reliable Change Index (RCI), which determines whether the magnitude of change is statistically significant. Data to calculate the RCI was drawn from a large non-clinical sample (Crawford et al., 2001). The cut-off score for the HADS-Total was ≤ 13 determined using "criterion a" To be classified as recovered, patients must demonstrate reliable change and their post-treatment or follow-up scores must be below the cut off score. The data were analyzed using SPSS version 24.

#### TABLE 1 | Participant characteristics.


# RESULTS

# Participant Characteristics

Forty-three consecutive referrals were identified as potentially eligible. There were 16 patients who did not fpsyg-10-00162 January 29, 2019 Time: 16:58 # 5



df, degrees of freedom; HADS, Hospital Anxiety and Depression Scale; IES-R, Impact of Event Scale-Revised; FCRI, Fear of Cancer Recurrence Inventory; FACT-G, Functional Assessment of Cancer Therapy-General; MCQ-30, Metacognitions Questionnaire-30; CAS-1, Cognitive Attentional Scale.

TABLE 3 | Within group effect sizes (Cohen's d) for outcome measures at post-treatment and 3- and 6-month follow-up.


HADS, Hospital Anxiety and Depression Scale; IES-R, Impact of Event Scale-Revised; FCRI, Fear of Cancer Recurrence Inventory; FACT-G, Functional Assessment of Cancer Therapy-General; MCQ-30, Metacognitions Questionnaire-30; CAS-1, Cognitive Attentional Scale.

enter the study; 10 did not wish to participate, 3 did not attend the assessment interview 1 patient did not have a have a cancer diagnosis, 1 patient did not meet the threshold for severity of distress with a HADS-T score of less than 16 and 1 patient had a recurrence of cancer.

Twenty-seven patients began the trial of whom 20 completed treatment; a completion rate of 74%. Of the seven patients who did not complete the six sessions of MCT; three patients attended only one session, two patients 2 sessions, one patient 3 sessions and the final patient attended 4 sessions but sporadically and decided that it was not feasible to continue therapy. Reasons for non-completion were; one patient was hospitalized for cancer recurrence, one participant stopped therapy to be able to provide full time care for a relative, 2 participants did not wish to undertake psychological therapy and 3 patients dropped out without providing a reason. The demographic and clinical characteristics of the sample shown in **Table 1**. It is notable that 96% of the sample met caseness for anxiety with 93% also scoring above the clinical cut-off for PTSD. Additionally, 8 of the 27 patients had experienced a cancer recurrence, none of these patients discontinued MCT.

# Treatment Effects

There were significant main effects of time on all outcome measures (**Table 2**). Follow-up Bonferroni pairwise comparisons demonstrated significant differences from pre-treatment to post-treatment, and from pre-treatment to 3-and 6-month follow up on all outcome measures indicating that treatment effects were maintained. Overall, there was significant improvement

#### TABLE 4 | Clinical significance outcomes on HADS-total.


ITT: intention to treat sample; Completers: treatment completers sample.

across all symptom and quality of life measures and significant reductions in metacognitive beliefs (MCQ-30) and processes (CAS-1).

# Effect Size Estimates

fpsyg-10-00162 January 29, 2019 Time: 16:58 # 6

Within group effect sizes for the ITT sample are shown in **Table 3**. There are large pre to post-treatment effect sizes across all outcome measures (0.83–1.66). There are comparable effect sizes across all measures at both follow-up timepoints illustrating that the magnitude of treatment effects is maintained from post-treatment to 6-month follow-up.

# Clinically Significance of Treatment

In the ITT sample, most participants were recovered on the HADS-Total at post-treatment and across the follow-up period. In terms of the proportion of patients that responded to treatment, 81% were improved at post-treatment and 74% at 6-month follow-up. Examination of the recovery rates for those patients that completed treatment shows recovery rates of 80% at post-treatment and 70% at 6-month follow-up. A summary of the clinical significance of treatment outcomes is shown in **Table 4**.

# DISCUSSION

This study provides further support for the potential of brief MCT to alleviate psychological morbidity in cancer survivors. Following six 1-hour sessions of MCT, there were significant reductions in anxiety depression, post-traumatic stress symptoms, FCR and improvements in quality of life. There were also significant reductions in metacognitive beliefs and the CAS as predicted by the metacognitive model (Wells and Matthews, 1994, 1996). Treatment gains were sustained across all measures of psychological morbidity and metacognitive beliefs and processes through to 6-month follow-up. The practical significance as opposed to the statistical significance of the results was assessed using the Jacobson approach to clinical significance. In those patients who completed brief MCT, there were very high recovery rates on the primary outcome variable assessing the severity of general distress; 80% of patients were recovered following six one-hour sessions of individually delivered MCT. The recovery rate of 70% at 6-month follow-up suggests that the effects of the intervention persist beyond treatment completion. Brief MCT appeared acceptable to cancer survivors with approximately 75% of participants starting treatment completed treatment. It is possible that the treatment completion rate can be improved and early drop-outs from treatment prevented by ensuring patients are more effectively socialized to the aims of MCT.

The within group effect sizes on FCR provide the opportunity to benchmark the effects of brief MCT with those reported in recent randomized controlled trial evaluating an integrative approach for FCR. The psychological treatment in the trials conducted by Butow et al. (2017) evaluated an intervention (ConquerFear) based on the treatment components drawn from three theoretical frameworks; common sense model (Leventhal et al., 1992) the self-regulatory model (Wells and Matthews, 1994) and relational frame theory (Hayes et al., 2006). Although the ConquerFear intervention was more efficacious than an attention control condition, the within group effect size for FCR from pre to post-treatment was 0.77. This compares to a within group effect size of 1.66 in the present study. Although, the present study had a much smaller sample size thereby limiting the generalizability of this finding. However, unlike the ConquerFear study, our open trial included participants with depression and severe trauma symptoms indicative of PTSD. Developing specific interventions for each aspect of psychological morbidity for cancer survivors may be unnecessary and integrating treatment components from theoretically inconsistent models could "dilute" treatment efficacy and compromise therapist training (Wells and Fisher, 2015; Byrne et al., 2018).

The present open trial is a valuable step in the translation of MCT from adult mental health populations to cancer survivors and is following the recommended framework for translating psychological interventions to a new population (Craig et al., 2008). The limitations of open trials are well known but should not undermine their place in treatment development research (Craig et al., 2008). No data was collected on either treatment adherence or therapist competency beyond that achievable through weekly supervisory sessions. Subsequent studies should include independent assessment of both treatment adherence and therapist competency to increase confidence in the conclusions drawn and that any treatment effects were attributable to MCT.

A comparatively small sample was used, but the sample appeared representative of cancer survivors referred to the clinical health psychology service. Other limitations include the lack of ethnic diversity and that most of the sample were female, thereby compromising external validity. Treatment outcome was assessed exclusively by self-report questionnaires in the present study. Although exclusive reliance on self-report questionnaires could be considered a methodological weakness, the study was not focused on changes psychiatric diagnosis, rather the study was designed to measure general distress for which the "gold standard" outcome measure for evaluating the efficacy of interventions on emotional distress in cancer was used (Luckett et al., 2010).

Overcoming other limitations of open trials can be achieved through conducting randomized controlled evaluation. It would be valuable to assess the hypothesized mechanisms of change in the context of an RCT against the current recommended treatment approaches, it may be that the treated patients who recover change to most on metacognitive variables regardless of the treatment received. There were statistically significant reductions in all metacognitive beliefs and the CAS over treatment, which were maintained through to the 6-month follow up assessment. This study adds to the extant literature that MCT has the potential to be an efficacious psychological intervention for adult cancer survivors. Given the limited outcomes of currently available interventions, there is an obvious need to conduct a controlled evaluation of the potential of brief MCT to alleviate psychological morbidity in cancer survivors.

# AUTHOR CONTRIBUTIONS

fpsyg-10-00162 January 29, 2019 Time: 16:58 # 7

PF and PS designed the study. AB, LF, and PF were the therapists. PF drafted the manuscript. GA and HU recruited and assessed participants at intake and following treatment. All authors

# REFERENCES


have contributed to drafting and revising the manuscript and approved its submission.

# FUNDING

This study was supported by the United Kingdom Medical Research Council Confidence in Concept Scheme, awarded to the University of Liverpool.


fpsyg-10-00162 January 29, 2019 Time: 16:58 # 8


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Fisher, Byrne, Fairburn, Ullmer, Abbey and Salmon. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Can the Attention Training Technique Reduce Stress in Students? A Controlled Study of Stress Appraisals and Meta-Worry

*Peter Myhr1 , Timo Hursti 2 , Katarina Emanuelsson2 , Elina Löfgren 2 and Odin Hjemdal 3,4 \**

*1MCT-Stockholm, Private Practice, Stockholm, Sweden, 2 Department of Psychology, Uppsala University, Uppsala, Sweden, 3 Department of Psychology, Norwegian University of Science and Technology, Trondheim, Norway, 4 Division of Psychiatry, St. Olavs Hospital, Nidaros DPS, Trondheim, Norway*

#### *Edited by:*

*Lora Capobianco, Manchester Mental Health and Social Care Trust, United Kingdom*

### *Reviewed by:*

*Karin Carter, Greater Manchester Mental Health NHS Foundation Trust, United Kingdom Robin Bailey, Liverpool John Moores University, United Kingdom*

> *\*Correspondence: Odin Hjemdal odin.hjemdal@ntnu.no*

#### *Specialty section:*

*This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology*

> *Received: 09 August 2018 Accepted: 17 June 2019 Published: 10 July 2019*

#### *Citation:*

*Myhr P, Hursti T, Emanuelsson K, Löfgren E and Hjemdal O (2019) Can the Attention Training Technique Reduce Stress in Students? A Controlled Study of Stress Appraisals and Meta-Worry. Front. Psychol. 10:1532. doi: 10.3389/fpsyg.2019.01532*

The present study tested the impact of attention training on cognition; secondary appraisal of perceived stress, and on metacognition; meta-worry in stressed students. Theoretically derived from the Self-Regulatory Executive Function model (S-REF model; Wells and Matthews, 1994a, 1996), the attention training technique (ATT; Wells, 1990) is intended to promote flexible, voluntary external attention and has been shown to reduce symptoms of psychological distress. The present experimental study explored the effects of ATT on cognitive and metacognitive levels of appraisal, namely perceived stress (primary outcome) and meta-worry (secondary outcome). Stressed students were randomized to an experimental ATT group (*n* = 23) or a control group (*n* = 23). The ATT group attended an initial training session followed by 4 weeks of individual (12 min) daily ATT practice. The control group waited for 4 weeks before receiving the intervention. The outcomes were scores on the Perceived Stress Scale 14 (PSS-14) and the Meta-Worry Questionnaire (MWQ) frequency and belief subscales at post study. Both measures decreased significantly following ATT with large pre- to post- effect sizes but there were minimal changes in the control group. The between-group differences were statistically significant. The results add to the literature on the potential effects of ATT by demonstrating effects on the content of cognitive stress appraisals and on meta-worry in an academic setting in a stressed student sample.

Keywords: attention training, stress, meta-worry, S-REF model, experimental study

# INTRODUCTION

In the Self-Regulatory Executive Function model (S-REF; Wells and Matthews, 1994a, 1996), stress reactions are viewed as arising from the activation of a pattern of processing called the cognitive attentional syndrome (CAS) that is marked by self-focused attention and consists of repetitive conceptual processing of negative information (worry, rumination, threat monitoring). In this model more severe stress reactions are linked to a higher level of unhelpful metacognitions, especially those beliefs that thinking cannot be controlled or is harmful (i.e., meta-worry or worry about worrying; Wells, 1994).

One of the first techniques developed to alleviate the CAS was the attention training technique (ATT: Wells, 1990), which involves the practice of specific auditory attention control exercises. The technique shows evidence of effectiveness for a range of anxiety and depression symptoms as reported in systematic reviews (Wells et al., 1997; Papageorgiou and Wells, 1998, 2000; Siegle et al., 2007, 2014; Calkins et al., 2015; Fergus and Bardeen, 2016). Furthermore, there is emerging evidence that the effects are detectable in neurocognitive measures associated with executive control of attention (Knowles et al., 2016).

While the ATT has been applied to clinical states such as anxiety and depression symptoms, the effect of the technique on more general stress-related appraisals has yet to be evaluated. In addition, the impact of ATT on cognition as well as metacognition levels of appraisal in stress remains to be tested. Studies from clinical samples show that the technique can reduce rumination and modify metacognitive beliefs. However, Wells and Matthews (1994a) see the CAS as potentially undermining secondary appraisals of the ability to cope with challenges. Thus, ATT should have an effect on cognitive appraisals of ability to cope with stress as well as impacting on metacognition (i.e., meta-worry). The current study set out to test whether the ATT can work in a stressed sample by impacting on cognitive and metacognitive appraisals.

The transactional theory of stress (Lazarus and Folkman, 1984) views stressful situations as those that are appraised as taxing or exceeding the individual's resources to cope. This process is closely tied to primary and secondary appraisals. Primary appraisals concern the evaluation of events in the terms of initial threat or challenges while secondary appraisal refers to what can be done to cope and the likely success of responses. The S-REF model (Wells and Matthews, 1994a, 1996) on the other hand links stress to the activation of a syndrome of cyclical thinking in the form of worry and rumination coupled with diminished perceptions of metacognitive control over thinking. Wells and Matthews have argued that the deleterious effects of the syndrome are most likely to be observed in situations that are cognitively demanding such as those that are ambiguous or contain uncertainty because they require more cognitive resources that are depleted by the syndrome. Consistent with this idea, Wells and Matthews (1994b) demonstrated a negative association between self-focused attention (a marker for the CAS) and emotion-focused and problem-focused coping only in mixed controllability situations.

In the present study, we set out to test the effects of an ATT intervention on the mechanisms of stress implicated in the transactional theory and the S-REF model. To reach this aim, we therefore assessed stress appraisals consistent with Lazarus and Folkman's model as a primary outcome and metacognitive appraisals in the form of meta-worry consistent with the S-REF as a secondary outcome. Meta-worry was first identified by Wells (1994) as a process of worrying about worry and is a dysfunctional metacognitive appraisal process. It consists of appraising worry as uncontrollable and dangerous and is thought to be closely associated with underlying metacognitive beliefs. Meta-worry is an important process contributing to the CAS in psychopathology including generalized anxiety disorder (Wells, 1995, 2005).

# Aim

In the present study, we tested the impact of the ATT on stress-related appraisals and meta-worry in stressed students. We aimed to address the question: can the ATT reduce stress appraisals and meta-worry?

# MATERIALS AND METHODS

# Design

The study uses a randomized controlled experimental design comparing the ATT with a wait-list control group. There were pre- and post-treatment assessments regarding outcome measures.

# Participants

The majority of the participants were students within the faculty of social sciences. The inclusion criterion was intended to be as broad as possible. However, to be eligible for the study, students had to have a 75% or higher course load. Part-time students with course load lower than 75% were not included. In addition, students had to self-identify as currently "stressed." Participation required attending a 45-min training session and then undertaking the ATT 12 min daily for 4 weeks. The study used a convenience sample. Participation was voluntary, and compensation was not typical apart from three participants who received university credits as part of a psychology class at Uppsala University.

The initial sample consisted of 48 participants of whom 34 (71%) identified themselves as women and 13 (27%) as men, and one (2%) had a different gender identity. The participants ranged from 19 to 43 years of age (*M* = 25.4, SD = 5.2).

### Drop-out

Post-intervention data were collected on 46 of the original 48 participants. One participant withdrew because of limited availability of time and the second for unknown reasons. The dropout rate was 4.17%.

# Material

*Perceived Stress Scale 14 (PSS-14)* was developed by Cohen et al. (1983). It measures the extent to which life situations are evaluated as stressful; thus, it's items tap primary and secondary appraisals according to the transactional stress model. It consists of seven positive and seven negative items all measured on a 5-point Likert scale from 0 (*never*) to 4 (*very often*). An example positive item is: "In the last month, how often have you dealt successfully with irritating life hassles?" An example negative item is: "In the last month, how often have you felt nervous and stressed?" It has shown good psychometric properties (Lee, 2012). It has been tested in Sweden and has good psychometric properties (Cronbach *α* = 0.84, *α* = 0.90) and indications of validity has been shown as it differentiates between a sample with stress disorder and other samples, as well as predicting sensitivity toward change as significant changes were identified from pre to post in a work rehabilitation intervention (Eklund et al., 2014).

*Meta-Worry Questionnaire (MWQ)* measures the level of metaworry (Wells, 2005). The response format measures two aspects of meta-worry: the frequency of meta-worry (frequency scale) and the respective belief in the meta-worry (belief-scale). The frequency of meta-worry contains seven statements, for example: "When I worry, I think: I'm abnormal for worrying." The belief in the same meta-worry thought is measured by responding on a scale from 0 (*I do not believe this thought at all*) to 100 (*I am completely convinced this thought is true*). The psychometric properties are good (for frequency Cronbach *α* = 0.88 and for belief scale *α* = 0.95) and validity has been demonstrated as patients with generalized anxiety disorder score significantly higher than somatic anxiety and no-anxiety groups (Wells, 2005).

# Procedure

Participants were recruited following provision of information about the study that was given in classes at Uppsala University. Information was also given on bulletin boards at Uppsala University, Swedish University of Agricultural Sciences in Uppsala, at the student union offices, students' health services in Uppsala, as well as shared on social media. Those who volunteered to participate were contacted by phone for further information and settling of specific times for ATT training. All communication and questionnaires were in Swedish, with the exception of the original English ATT soundtrack which was used both in the introduction group session and subsequent individual ATT training sessions.

Two students were responsible for conducting the 45-min introductory group sessions with a maximum of five participants at a time. The students were supervised by an MCT-I certified therapist. The session started with a few minutes of psychoeducation to illustrate how thoughts and appraisals (primary and secondary) may influence perceived stress levels in accordance with the transactional theory of stress. This was followed by an introduction to the CAS and the original rational for the ATT, a practice session of the ATT followed.

ATT involves instructing individuals to focus on external sounds and specific spatial locations. The training consists of three parts focusing on (1) selective attention, (2) rapid change of attention, and (3) divided attention (Wells, 2009). We used a 12-min pre-recorded soundtrack in the implementation of practice following the procedure described by Wells (Wells, 1990). After having completed the ATT, participants were encouraged to ask questions to clarify the use of the ATT.

# Conditions

*The ATT group* participants were assessed prior to the training and after 4 weeks of training. After the introductory training session, the participants were instructed to perform ATT daily during the following 4-week period. To encourage adherence participants were asked to register the date and time of their individual training, and they were encouraged to continue training even if they missed a day or two. Frequency of individual training was not collected. Two weeks into the ATT training, the participants received an e-mail reminder to continue ATT training. The participants were encouraged not to engage in other self-help activities targeting stress.

*Wait-list* participants were assessed prior and after the 4 weeks waiting period. They were informed that their training period would be introduced after the post-waiting assessment. During their waiting period they were encouraged to live their lives normally until the training would start. After their 4 week waiting period, their training was identical to the one received by the ATT group.

# Data Collection

The pre- and post-treatment assessments were done *via* a website generated using surveymonkey.com. After the recruitment, participants received an e-mail with a link to the study website where they were provided with further information, after which they provided informed consent before proceeding to the questionnaires. Questionnaires included demographic questions (i.e., gender and age), the PSS-14, and the MWQ. After 4 weeks, participants received a new web link *via* e-mail leading to post-treatment assessment, which included the PSS-14 and MWQ. Filling out the questionnaires took approximately 10 min.

# Randomization

After inclusion, participants were randomized to either the ATT or wait-list. Randomization was undertaken using www. random.org matching on gender identity. Each group consisted of 24 participants. In the ATT group, the gender identity was 17 women and seven men, and in the wait-list condition 17 women, six men, and one prefering not to disclose gender identity.

# Ethics

The procedures for data collection were performed in accordance with the Declaration of Helsinki, as well as the guidelines for professional conduct of clinical psychologist in the Nordic countries. All participants received a written informed consent in addition to the oral information. Participants were informed about the study, that participation was voluntary and confidential, and they had the right to withdraw from the study without giving any reason or it having consequences. All data were stored anonymously. It was specified in the given information that if participants needed further assistance during or after the project, they could contact the two clinical psychologists involved in the project.

The present study was part of the students' thesis, and the guidelines for student projects at Uppsala University were followed. The study was ethically reviewed and accepted by the Department of Psychology, Uppsala University.

# Statistical Analyses

Statistical analyses were undertaken using SPSS version 23. A Shapiro-Wilks normality test indicated that the results were normally distributed. Change scores for the PSS-14 were established by subtracting post-scores from pre-scores. Withingroup effect sizes were estimated using Cohen's *d* where 0.2 indicates small, 0.5 medium, and 0.8 large effect. In the ANOVA, effect sizes were estimated using eta squared (*η*<sup>2</sup> ), where 0.01 indicates small, 0.059 medium, and 0.138 large effect (Clark-Carter, 2010). Changes in PSS-14 and MWQ scores were explored using a mixed two-way ANOVA, followed up with dependent t-tests. The associations between perceived stress and meta-worry and if participants with higher levels of meta-worry experienced ATT as particularly helpful were explored using correlations.

# Results

### Changes in Perceived Stress

A mixed two-way ANOVA was conducted using the PSS-14 as a dependent variable. The independent variables were time and group. Time was pre- and post assessment, and group was either ATT or wait-list. The analyses showed a significant main effect for time, *F*(1, 44) = 23.48, *p* < 0.001, with a large effect size of *η*<sup>2</sup> = 0.53. There was no significant main effect for group *F*(1, 44) = 0.24, *p* = 0.63. There was a significant interaction effect (time × group) *F*(1, 44) = 12.43, *p* = 0.001, with a large effect size of *η*<sup>2</sup> = 0.28. The level of perceived stress among stressed university students was significantly reduced after participating in the ATT, in comparison to the wait-list control group.

Further analyses using follow-up t-tests confirmed a significant reduction in PSS-14 scores in the ATT group *t*(22) = 5.52, *p* < 0.001, with a large effect size but the wait-list group did not show a significant reduction over time *t*(22) = 1.01, *p* = 0.32. Mean values and effect sizes (Cohen's *d*) are reported in **Table 1**.

### Changes in Meta-Worry

To explore changes in frequency of meta-worry, a mixed two-way ANOVA was undertaken. There was a main effect for time, *F*(1, 44) = 8.89, *p* = 0.005, with a large effect size *η*<sup>2</sup> = 0.20, and a significant interaction effect (time × group), *F*(1, 44) = 7.17, *p* = 0.01, also with a large effect size *η*<sup>2</sup> = 0.16. There was no significant main effect for group *F*(1, 44) = 0.27, *p* = 0.60. *T*-tests indicated significant reduction from pre- to post-scores in the ATT group for MWQ-F, *t*(22) = 4.17, *p* < 0.000, the effect size was medium. For the wait-list group, there was no significant reduction for MWQ-F, *t*(22) = 0.21, *p* = 0.84. Mean scores and effect size (Cohen's *d*) are reported in **Table 1**.

To explore changes in beliefs in meta-worry (MWQ-B), a similar mixed two-way ANOVA was conducted. Significant main effects were found for time, *F*(1, 44) = 11.49, *p* = 0.001, with a large effect size of *η*<sup>2</sup> = 0.26. There were no significant interaction



*ATT, attention training; WL, wait-list; PSS-14, perceived stress scale 14; MWQ-F, metaworry questionniare - frequency; MWQ-B, meta-worry questionniare – belief.*

effects (time × group), *F*(1, 44) = 3.68, *p* = 0.06, neither was there a significant main effect for group, *F*(1, 44) = 0.02, *p* = 0.88. Therefore, beliefs decreased overall in both groups but there was no differential effect observed.

# DISCUSSION

We found an effect of the ATT intervention on levels of perceived stress in university students. The intervention led to greater reductions in stress levels compared with a no-treatment waiting period. We also demonstrated an effect in reducing meta-worry frequency scores. However, the effect on metaworry belief was non-significant. The results are consistent with an ameliorative effect of ATT on hypothesized mechanisms of stress in stressed students. The reason for a lack of an effect on meta-worry belief levels is unclear; this may be due to the ATT not being effective on this dimension or lack of power to detect such an effect given the small sample size of the study. It should be noted that the initial scores on this dimension were low with high variability and this may have contributed to the size of the effect observed. The within-group effect sizes show a large effect for the perceived stress and meta-worry outcomes. Individuals were included based on their own evaluation of personal stress levels. The mean score of 31.07 was well above a Swedish non-clinical population (Eklund et al., 2014). For women with stress-related disorders, the mean score of 30.0 has been suggested.

The use of a control group that waited for the length of time over which the ATT group received the intervention controls for the passage of time and spontaneous recovery from stress, but it does not control for the non-specific factors involved in delivering an intervention. We cannot be sure that it is the ATT that caused improvement or other factors such as deviations from normal routines caused by practicing the technique, placebo effects, or expectancies of improvement. We aimed to test a more basic question: does it have an effect? This question is useful to clarify before more rigorous studies are planned.

There are other important limitations of the study that should be considered. We delivered ATT in a dose that is below what is normally recommended for clinical samples, where practicing twice a day rather than once is usually advised. Furthermore, we cannot be sure of the actual level of practice that the students adhered to, which is a major limitation. We also combined the ATT with a rationale that described the role of primary and secondary appraisals in stress and this is not part of the usual rationale for ATT that is grounded in the metacognitive model. We cannot ascertain if this hybrid explanation had a detrimental, positive, or no impact on the effectiveness of the intervention. When the ATT is used as a therapeutic intervention it is typically combined with therapist-led guidance and exploration of subjective experiences to re-shape the clients' maladaptive metacognitions (Wells, 2009).

In conclusion, these findings tentatively add to the research on the effects of ATT by suggesting that it can have an effect on stress responses, which extends the potential utility of ATT to managing stress in non-clinical samples. The results support the further investigation of effects of the technique within this context. Future studies should aim to control non-specific treatment factors and to separate the effects of the attention exercises from the other elements in the package used.

# ETHICS STATEMENT

The procedures for data collection were performed in accordance with the Declaration of Helsinki, as well as the guidelines for professional conduct of clinical psychologist in the Nordic countries. All participants received a written informed consent in addition to the oral information. Participants were informed about the study that participation was voluntary and confidential,

# REFERENCES


and they have the right to withdraw from the study without giving any reason or it having consequences. Data were saved anonymized. It was specified in the given information that if participants needed further assistance during or after the project, they could contact two clinical psychologists involved in the project. The present study was part of a student thesis, and the guidelines for student projects at Uppsala University were followed. The study was ethically reviewed and accepted by the Department of Psychology, Uppsala University.

# AUTHOR CONTRIBUTIONS

PM developed the study and supervised. TH contributed with analyses and supervision. KE and EL ran the experiment. All authors contributed in the writing of the manuscript.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2019 Myhr, Hursti, Emanuelsson, Löfgren and Hjemdal. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# A Preliminary Evaluation of Transdiagnostic Group Metacognitive Therapy in a Mixed Psychological Disorder Sample

*Pia Callesen1,2 , Lora Capobianco1,3 \* , Calvin Heal 4 , Carsten Juul 2 , Sisse Find Nielsen 2 and Adrian Wells1,3*

*1 School of Health Sciences, Division of Psychology and Mental Health, University of Manchester, Manchester, United Kingdom, 2 CEKTOS, Copenhagen, Denmark, 3 Research and Innovation, Greater Manchester Mental Health NHS Foundation Trust, Manchester, United Kingdom, 4 School of Health Sciences, Division of Population Health, Health Services Research and Primary Care, Manchester, United Kingdom*

#### *Edited by:*

*Francisco J. Ruiz, Fundación Universitaria Konrad Lorenz, Colombia*

#### *Reviewed by:*

*Jennifer Jordan, University of Otago, Christchurch, New Zealand Roger Hagen, Norwegian University of Science and Technology, Norway Nexhmedin Morina, University of Münster, Germany*

*\*Correspondence: Lora Capobianco lora.capobianco@manchester.ac.uk*

#### *Specialty section:*

*This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology*

*Received: 17 November 2018 Accepted: 23 May 2019 Published: 20 June 2019*

#### *Citation:*

*Callesen P, Capobianco L, Heal C, Juul C, Find Nielsen S and Wells A (2019) A Preliminary Evaluation of Transdiagnostic Group Metacognitive Therapy in a Mixed Psychological Disorder Sample. Front. Psychol. 10:1341. doi: 10.3389/fpsyg.2019.01341*

Objective: Comorbidity is common among anxiety and depression. Transdiagnostic treatment approaches have been developed to optimize treatment and offer a more unified approach suitable for individuals with comorbidities. Metacognitive therapy (MCT) is a transdiagnostic therapy for psychological disorder and is based on the metacognitive model. The present study is a service evaluation of the outcomes associated with group MCT delivered to unselected patients at a Danish outpatient clinic.

Methods: A total of 131 self-diagnosed patients received 6 sessions of group MCT. Symptoms of anxiety and depression were measured by the Hospital Anxiety and Depression scale (HADS) and metacognition was assessed using the Cognitive Attentional Syndrome-1 (CAS-1). Participants were assessed at pre-treatment, post-treatment, and at 6 months follow-up as per usual clinic protocol. Linear mixed-effects regressions were used to assess the transdiagnostic effects of group MCT. Treatment effect sizes are reported for subgroups based on participant's reason for seeking treatment (anxiety, depression, or comorbid). Effect sizes were not conducted for the depression subgroup given the limited number of participants. Clinically significant change is reported for all subgroups.

Results: Group MCT was associated with large effect sizes for symptoms of anxiety and depression for patients seeking treatment for anxiety (*d* = 1.68), or comorbid (1.82). In addition, 66.7% of patients were classified as recovered at post-treatment, and 12.9% were classified as improved. These results were largely maintained at 6-month follow-up.

Conclusion: These preliminary findings support the continued use of group MCT in the current outpatient clinic and suggest that it may be an efficacious and cost-effective treatment when delivered in "transdiagnostic" groups.

Keywords: metacognitive therapy, transdiagnostic, depression, anxiety, group therapy

# INTRODUCTION

There is increasing evidence demonstrating that anxiety and depression rarely occur alone and instead are highly comorbid (Brown et al., 2001). Brown et al. (2001) evaluated the current and lifetime comorbidity of anxiety and mood disorders and highlighted that, of those with a principal anxiety or mood disorder, the current and lifetime comorbidity with other Axis I disorders was 57 and 81% respectively. Similarly, Lamers et al. (2011) investigated the comorbidity patterns of anxiety and depression in the Netherlands and found that 67% of patients with a depressive disorder had a current comorbid anxiety disorder. Furthermore, among individuals with a current anxiety disorder, 63% had current comorbid depressive disorder. Despite the high rate of comorbidity, psychological paradigms such as Cognitive Behavioral Therapy (CBT) often focus on providing a disorder-specific treatment, whereby separate protocols are used for treating different disorders such as generalized anxiety, OCD, PTSD, and depression. These protocols are typically supported by disorder-specific case formulations and models. However, treatments focusing on disorder-specific models can be problematic as patients often do not present with a single disorder. Therefore, clinicians are required to treat the most pressing disorder even though the patient may be presenting with more than one problem.

The high comorbidity rate among mental disorders supports the need for transdiagnostic models and treatments that focus on the common underlying processes that maintain psychological disorders. CBT is one of the most widely evaluated treatments for psychological disorders. Although CBT is primarily delivered using a disorder-specific protocol, more recent research has aimed to deliver CBT using a transdiagnostic approach. A recent systematic review and meta-analysis of transdiagnostic CBT for anxiety and depression found mixed results on its effectiveness. Of the two studies that compared transdiagnostic CBT to a control condition, only one study (Schmidt et al., 2012) found some evidence for the effectiveness of this approach, while Erickson et al. (2007) did not report significant findings for anxiety.

One transdiagnostic approach to CBT is the unified protocol (UP) for emotional disorders (Barlow et al., 2010). The UP incorporates principles from traditional CBT such as cognitive restructuring and exposure procedures together with advances in emotion regulation research, including an emphasis on increasing patient's awareness of maladaptive cognitions and behaviors (Wilamowska et al., 2010; Craske, 2012; Farchione et al., 2012; Bullis et al., 2015; Laposa et al., 2017). Bullis et al. (2015) evaluated the UP in a group format delivered over 12 sessions. The authors demonstrated medium to large effect sizes on symptom measures of depression and anxiety respectively. However, the study had a small sample size of 11 participants. More recently, Laposa et al. (2017) evaluated the UP in a group format over 14 sessions with 26 participants. There were medium to large effect sizes on measures of anxiety and depression but they noted that participant's scores on the Penn State Worry Questionnaire and Social Interaction Anxiety Scale remained above their clinical cutoffs at post-treatment. Furthermore, the Depression Anxiety Stress Scale-Anxiety sub scale and Quick Inventory of Depressive Symptom scores remained in the moderate range at the end of treatment.

In other areas, third-wave approaches of behavioral and cognitive behavioral therapies such as mindfulness-based stress reduction (MBSR; Kabat-Zinn, 1990) and acceptance and commitment therapy (ACT; Hayes et al., 1999) have also been used to treat transdiagnostic samples. MBSR focuses on cultivating present moment awareness and combines formal and informal mindfulness practices such as mindfulness of the breath, thoughts, bodily sensations, and routine activities. ACT combines psychoeducation with exercises that aim to increase mental flexibility and mindfulness experiences while decreasing avoidance of activities. ACT targets six core processes with the aim of increasing psychological flexibility. The six core processes are: contact with the present moment, values, committed action, self as context, delusion, and acceptance. ACT integrates mindfulness and acceptance processes and commitment and behavior change processes to enhance psychological flexibility (Hayes et al., 2006). In a recent systematic review and metaanalysis, Newby et al. (2015) found a significant difference favoring CBT in comparison to mindfulness/acceptance-based interventions in anxiety symptoms. (CBT, hedge's *g* = 0.88, mindfulness/acceptance, hedge's *g* = 0.61). However, there was no significant difference between treatment type on symptoms of depression (CBT, hedge's *g* = 0.84, mindfulness/acceptance, hedge's *g* = 0.92).

One of the earlier transdiagnostic approaches was presented by Wells and Matthews (1994, 1996) in their Self-Regulatory Executive Function (S-REF) model. They argued for the conceptualization of universal psychological factors across pathologies and asserted that psychological disorder is maintained by a common maladaptive cognitive attentional syndrome (CAS) that should be the target of treatment. The CAS is characterized by increased self-focused attention, repetitive negative thinking involving worry and rumination, and unhelpful coping strategies and behaviors such as attentional threat monitoring, thought suppression, and avoidance. The CAS is a result of an individual's metacognitive beliefs which lead to prolonged negative processing and consequent distress. There are two types of metacognitive beliefs: positive metacognitive beliefs (PMC) and negative metacognitive beliefs (NMC). Negative metacognitive beliefs concern the uncontrollability and danger of worry (i.e., "I cannot control my worry," "my worrying may harm me"). In contrast, positive metacognitive beliefs concern the usefulness of worry (i.e., "worrying helps me cope," "if I worry I'll be prepared). These underlying metacognitive beliefs are considered a major factor driving the CAS. Metacognitive therapy (MCT: Wells, 1995, 2009) was developed based on this model, and aims to remove the CAS and modify positive and negative metacognitive beliefs. MCT has demonstrated significant efficacy across various psychological disorders. Normann et al. (2014) conducted a meta-analysis evaluating the efficacy of MCT for anxiety and depression, where they reported that MCT was a highly effective treatment. When MCT was compared to wait list control on the primary outcome measure, effect sizes favored MCT, *g* = 1.81. In addition, when MCT was compared with CBT, a large effect size was found favoring MCT, *g* = 0.97. Recently, Normann and Morina (2018) conducted an updated systematic review and meta-analysis on MCT for anxiety and depression and found that, when MCT was compared to wait list controls, there was a large pre- to post-treatment effect size, *g* = 2.06. Similarly, when MCT was compared to active control treatments, there was a medium to large effect size in favor of MCT, *g* = 0.68. More specifically, when MCT was compared to cognitive behavior therapy and behavioral activation interventions, a medium to large effect size was found favoring MCT from pre- to post-treatment, *g* = 0.69. Although MCT has been evaluated using an individual treatment format, there is increasing evidence that MCT is effective using a group format. McEvoy et al. (2015) tested group MCT in individuals with GAD who received six sessions of treatment for 2 h plus an additional 1-month follow-up session. The authors found that group MCT was associated with very large effect sizes from pre- to post-treatment on measures of negative metacognitions, worry, and repetitive negative thinking (*d* = 1.75–1.90). In addition, when evaluating reliable and clinically significant change based on Jacobson and Truax (1991) criteria, they found that at post-treatment, 86% of patients had reliably improved and 74% had recovered. Preliminary studies also highlight the efficacy of MCT in transdiagnostic samples (Johnson and Hoffart, 2016; Hagen et al., 2017; Johnson et al., 2017; Capobianco et al., 2018). Johnson et al. (2017) compared transdiagnostic MCT in an individual format with disorder-specific CBT and found that MCT was more effective than CBT (Cohen's *d* = 0.7) in alleviating anxiety symptoms at post-treatment. There was no difference at 12-month follow-up but this may be due to patients accessing other treatments over this period. Capobianco et al. (2018) conducted a pilot feasibility study comparing groupdelivered MCT or MBSR. They noted that while both treatments were acceptable and feasible to deliver in a group format, the preliminary data suggested that MCT might be more effective than MBSR.

In the present study, we aimed to add to the data on the effects associated with transdiagnostic MCT by collating the outcome data of patients who entered into group therapy in a Danish primary care outpatient clinic. The data that were routinely collected allowed us to examine the effects associated with receiving group MCT in a group of individuals selfreporting their reason for seeking treatment in a standard outpatient care setting. Such liberal inclusion criteria and the setting of the treatment are especially informative because they overcome one of the criticisms of tightly controlled trials that use extensive inclusion/exclusion criteria, thus compromising the extent to which participants represent those who are typically seen in outpatient clinics.

# MATERIALS AND METHODS

# Design

The design is essentially a service audit and is therefore an uncontrolled pre-post assessment with 6 months follow-up. Participants attended the Center for Cognitive Therapy and Supervision (CEKTOS), a Danish primary care outpatient clinic. Ethical approval was not sought for the study as data were collected as part of routine clinical practice and evaluated as part of a service audit. However, in accordance with clinical guidelines, patients provided informed and written consent for use of patient data and ethical standards for reporting were adhered to. This is in line with the rules and regulations of the Danish National Ethics Committee. As new patients contacted the clinic, they were offered the choice of group therapy or individual therapy. Recruitment occurred between August 2014 and May 2015; during this time, a total of 145 patients opted to take part in the group therapy being offered, which was 21% uptake rate for group therapy. The Generalized Anxiety Disorder Assessment (GAD-7) and Patient Health Questionnaire (PHQ-9) were administered at pre-treatment as part of general assessment and are reported here to help describe the sample. The Hospital Anxiety and Depression Scale (HADS) and CAS-1 were administered at pre-treatment, mid-treatment, posttreatment, and 6 months follow-up. Participants who opted to take part in group therapy were later approached and asked to provide written and informed consent to allow their anonymized data to be released for use in this evaluation; 14 participants (10%) did not consent for their data to be used for analyses. In addition, seven participants were removed from the analysis as they did not report a reason for seeking treatment, resulting in a total sample of 124 participants. Participants were given the opportunity to withdraw from the treatment at any time during the treatment and follow-up.

# Participants

There were 124 Danish outpatients (87 women, 37 men) treated. The mean age of the total sample was 42.10 (SD = 12.73; age range: 18–68). A total of 51 participants were currently taking medication for anxiety or depression. As there was less than 5% of missing data, means were used for imputing missing values. There was no intake interview or screening of suitability, this was an open treatment in which all consenting patients were deemed suitable and both referred and non-referred clients were eligible. Patients represented a range of different disorders.

# Procedure

Participants completed 6 weeks of group metacognitive therapy. Sessions lasted approximately 2 h. There were 16 groups with an average of eight participants in each. The outcome measures were administered at pre-, mid- and post-treatment, with 3 weeks between questionnaire administrations.

# MEASURES

# Primary Outcome Measure

Hospital Anxiety and Depression Scale The *Hospital Anxiety and Depression Scale* (HADS; Zigmond and Snaith, 1983) is a 14-item scale with two subscales (anxiety and depression). Each item is scored from 0 to 3, with subscale scores greater than 8 being indicative of anxiety and depression. Both subscales demonstrate good internal consistency, good validity and reliability (Zigmond and Snaith, 1983; Herrmann, 1997; Mykletun et al., 2001).

# Secondary Outcome Measure Cognitive Attentional Syndrome

The *Cognitive Attentional Syndrome* (CAS-1; Wells, 2009) assesses the extent to which the cognitive attentional syndrome, a key component of the metacognitive model, is activated in the last week. The CAS-1 is a 16-item measure where the first eight items are rated on a scale from 0 to 8, where 0 indicates none of the time and 8 indicates all of the time. Items 1 and 2 assess the extent to which individuals have been dwelling, worrying, or focusing on possible threat in the past week. Items 3–8 assess various coping behaviors that individuals may be engaging in to deal with negative thoughts (e.g., tried to control emotions, asked for reassurance). The final item assesses the positive and negative metacognitive beliefs that individuals hold (e.g., "*I cannot control my thoughts*," "*analysing my problems will help me find answers*"). The CAS-I demonstrates good internal consistency (Cronbach's alpha = 0.86) (Fergus et al., 2012).

# Pre-treatment Screening Measures Patient Health Questionnaire

This is a 9-item measure that assesses depression in primary health care, where greater scores indicate increasing severity of symptoms (Kroenke et al., 2001). Items are rated on a scale from 0 (not at all) to 3 (nearly every day). The scale has four cutoff points: 5 (mild depression), 10 (moderate depression), 15 (moderately severe depression), and 20 (severe depression). The scale demonstrates good reliability and validity (Cameron et al., 2008).

### Generalized Anxiety Disorder Assessment

This is a brief 7-item measure used to assess symptoms of generalized anxiety disorder in primary health care (Spitzer et al., 2006). Items are rated on a scale from 0 (not at all) to 3 (nearly every day), with greater scores indicating greater severity of anxiety. The scale has three cutoff points: 5 (mild anxiety), 10 (moderate anxiety), and 15 (severe anxiety). The scale demonstrates good internal and test-retest reliability, and good convergent and construct validity (Spitzer et al., 2006).

# Intervention

Group MCT was supported by the Generalized Anxiety Disorder (GAD) protocol as described in the treatment manual by Wells (2009) as this represents the core of the transdiagnostic treatment. The attention training technique (ATT, Wells, 1990) was added in the group treatment sessions, as the ATT helps to address perseverative thinking by promoting attention flexibility and executive control skills and is often used in depression. The ATT also meets the 5-3-20 criterion (Kratochwill et al., 2013) for an evidence-based intervention (Rochat et al., 2018). The 5-3-20 criterion states that an intervention is evidence based if it meets the following criteria: (1) the intervention has a minimum of five single case design studies that either meets standards or meets standards with reservations; (2) The single case design studies are conducted by at least three research teams with no overlapping authorship at three different institutions; and (3) the total number of cases (i.e., participants, classrooms, etc.) across studies totals at least 20. Sessions were delivered by two clinical psychologists trained in MCT and who were supervised by AW. Participants received six weekly sessions of group MCT that lasted approximately 2 h. Sessions focused on a group case formulation, the attention training technique, detached mindfulness, challenging positive and negative metacognitive beliefs, and formulating a personalized plan B. The plan B allowed participants to consolidate what they had learned in therapy and have a summary of how to deal with future negative cognitions.

# Statistical Analysis Plan

Analyses were conducted in STATA (version 15). Multiple imputation was used to impute missing data. Categorical variables were assessed using a Chi-square test. A linear mixed-effects regression incorporating all three time points (pre-treatment, post-treatment, and follow-up) on the total sample (ITT) was applied in order to evaluate the significance of change overall and examine any modifying effects of type of problem on outcome. A sensitivity analysis was also conducted; we used mean imputation to impute missing values at follow-up where missingness was 25.6% for each of the three outcomes. There was less than 1% missingness at pre- and post–treatment, so no imputation was used for these time points. Finally, effect sizes from pre-treatment to post-treatment and pre-treatment to follow-up were based on completers and calculated as Cohen's *d* (Cohen, 1988) using the formula *d* = (*M*1−*M*2)/SDpooled, where *M*1 is the mean at pre-treatment, *M*2 is the mean at posttreatment or follow-up, and SDpooled is the pooled standard deviation. We used the method outlined by Jacobson and Truax (1991) to calculate reliable clinical change based on the HADS total score, with the cut-off score being calculated using criterion "c," which was only conducted on treatment completers. Individuals were classified as recovered if they made a reliable change and were below the cut-off score. Individuals were classified as improved if they made a reliable change but were not below the cut-off score a post-treatment or follow-up. As the sample size for depression subgroup from pre- to posttreatment (*n* = 12) and pre-treatment to follow-up (*n* = 8) was disproportionately smaller than that of the other subgroups, it was not included in the effect size calculation.

# RESULTS

The flowchart (**Figure 1**) shows number of patients contacting the clinic for help, and the number of patients entering the transdiagnostic group intervention. Approximately 9–14 patients call the clinic each day, and of those, approximately 21% chose to participate in the group, while 79% chose to complete individual therapy. Completer analysis was conducted at posttreatment. At follow-up, there was a 70% data return rate.

# Descriptive Statistics

**Table 1** highlights the characteristics of participants based on their self-reported reason for seeking treatment. The total sample included 124 patients, 37 males (29.8%) and 87 females (70.2%). Individuals reported their primary reason for seeking treatment as anxiety, depression, or both (**Table 1**). Participants also reported secondary reasons for seeking treatment which included stress (58 participants), obsessive compulsive disorder (OCD; nine participants), and post-traumatic stress disorder (PTSD; one participant); however, six participants reported secondary reasons for seeking treatment as stress and obsessive compulsive disorder, and an additional six participants reported secondary reasons for seeking treatment as stress and PTSD. All subsequent analyses are based on individual's primary reason for seeking treatment. A Chi-square analysis demonstrated that there was a significant difference in primary reason for seeking treatment (e.g., anxiety, depression, both) by gender, *χ*<sup>2</sup> (2, *N* = 124) = 9.76, *p* = 0.008. There was a greater number of females seeking treatment for anxiety in comparison to males [52 females (41.9%), 11 males (8.9%)], with a similar pattern for those seeking treatment for anxiety and depression [29 females (23.4%), 20 males (16.1%)], while there was an equal gender balance for those seeking treatment for depression (9.7%). Ninety-three participants completed questionnaires at 6-month follow-up [66 women (71.0%), 27 men (29.0%)]. **Table 2** provides an overview of the means and standard deviations for the outcome measures at pre-treatment, post-treatment, and follow-up based on individuals' self-reported diagnosis and for the total sample of treatment completers.

# Outcomes Associated With Group Treatment

#### Hospital Anxiety and Depression Scale

To assess if there were any differences between groups (anxiety, depression, comorbid) over time, a mixed-effect regression was conducted. There was a significant main effect of time, with post-treatment and 6-month follow-up being associated with a respective 10.9 (95% CI 9.7–12.0) and 11.2 (95% CI 9.5–12.9) point reduction in HADS total score compared to baseline (*p* < 0.001), for the entire sample. There were nonsignificant differences between groups on HADS total score at post-treatment, for the depression compared to anxiety group [−1.62 (95% CI −4.68 to 1.44), *p* = 0.300], for the comorbid group compared to anxiety group [−1.65 (95% CI −4.21 to 0.91), *p* = 0.208], and for the comorbid group compared to depression group [−0.03 (95% CI −3.31 to 3.25),

TABLE 1 | Descriptive statistics by reason for seeking treatment.


*Note: M = mean; SD = standard deviation.*

TABLE 2 | Means and standard deviations for outcome measures for all patients (ITT) and treatment completers by reason for seeking treatment.


*Note: HADS = hospital anxiety and depression scale; CAS-1 = cognitive attentional syndrome 1; PMC = positive metacognitions; NMC = negative metacognitions; M = mean; SD = standard deviation.*

*p* = 0.987]. At follow-up however, there was a significant difference between the depression and anxiety groups [−6.48 (95% CI −11.43 to −1.53), *p* = 0.010]. This suggests that the depression subgroup had improved significantly more than the anxiety subgroup by follow-up. There was no such significant difference between the comorbid group and anxiety group [−3.08 (95% CI −6.64 to 0.47), *p* = 0.089] at follow-up, nor between the comorbid group and depression group [3.40 (95% CI −1.81 to 8.60), *p* = 0.201]. These results show that treatment was associated with significant improvements overall (HADS total). These results appear to support the transdiagnostic effect associated with group MCT such that irrespective of reason for seeking treatment, there were significant decreases in levels of distress between pre- and post-treatment and follow-up.

# Positive Metacognitive Beliefs

There was a significant main effect of time, with post-treatment and follow-up being associated with a respective 135.8 (95% CI 121.5–151.1) and 129.0 (95% CI 112.3–145.6) point reduction in positive metacognitive beliefs compared to baseline (*p* < 0.001), for the entire sample. There was a nonsignificant difference between groups on PMC at posttreatment, for depression compared to anxiety groups [−2.41 (95% CI −57.67 to 52.84), *p* = 0.932], for the comorbid compared to anxiety group [−16.04 (95% CI −13.70 to 45.78), *p* = 0.723], and comorbid compared to depression group [18.46 (95% CI −37.79 to 74.70), *p* = 0.520]. Likewise, at follow-up, there was a nonsignificant difference between the depression and anxiety groups [3.96 (95% CI −70.04 to 77.97), *p* = 0.916], the comorbid and anxiety groups [6.26 (95% CI −28.36 to 40.88), p = 0.723], and the comorbid compared to depression group [2.30 (95% CI −73.82 to 78.42), *p* = 0.953]. The results demonstrate that irrespective of reason for seeking treatment, MCT was associated with decreases in positive metacognitive beliefs. The results suggest changes in positive metacognitions were transdiagnostic and occur irrespective of reason for seeking treatment, MCT was associated with significant reductions in positive metacognitive beliefs between pre- and post-treatment and pre-treatment and follow-up.

### Negative Metacognitive Beliefs

There was a significant main effect of time, with post-treatment and follow-up being associated with a 149.9 (95% CI 133.8– 166.0) and 136.2 (95% CI 118.5–153.8) point reduction in NMC compared to baseline (*p* < 0.001) for the entire sample. There was a nonsignificant difference between groups on NMC at post-treatment, for depression compared to the anxiety group [41.70 (95% CI −31.21 to 114.60), *p* = 0.262], for the comorbid compared to anxiety [−0.34 (95% CI −33.05 to 32.36), *p* = 0.984] and compared to depression groups [−42.04 (95% CI 446.79–32.72), *p* = 0.270]. Likewise, at follow-up, there was a nonsignificant difference between the depression and anxiety groups [−7.26 (95% CI −53.50 to 38.97), *p* = 0.758], the comorbid and anxiety [−19.58 (95% CI −56.81 to 17.66), *p* = 0.303] and depression groups [−12.32 (95% CI −57.55 to 32.92), *p* = 0.594]. Overall, the results suggest improvement in negative metacognitions; however, this and the other results should be interpreted with caution due to the difference in number of individuals seeking treatment for anxiety and depression. Those with both anxiety and depression scored higher on NMC, 19 points (−4 to 53) although this was nonsignificant (*p* = 0.07). There were no significant group-by-time interactions in the analyses, suggesting that the nature of presenting problem did not modify outcomes.

# SENSITIVITY ANALYSIS

All findings were robust under the sensitivity analysis where missing values at time 3 were mean imputed.

# TREATMENT EFFECT SIZES

The effect sizes (ES) associated with treatment were calculated based on Cohen's *d* from pre- to post-treatment and pre-treatment to follow-up. Effect sizes were calculated based on subgroup (self-reported reason for seeking treatment; anxiety or comorbid) and for the total sample. Effect sizes were not calculated for the depression subgroup due to the small number of participants within this subgroup from pre-treatment to post-treatment (*n* = 12), and pre-treatment to follow-up (*n* = 8). All effect sizes are displayed in **Table 3**. Overall, the effect sizes are large, highlighting the potential efficacy of group MCT in a "transdiagnostic" sample. Between-subgroup effect sizes were calculated for the anxiety and comorbid subgroups at posttreatment for HADS total, positive metacognitive beliefs, and negative metacognitive beliefs. There was a small between-group effect size on the HADS total and positive metacognitive beliefs, Cohen's *d* = 0.22 and 0.02, respectively, favoring the comorbid anxiety and depression subgroup. There was also a small to medium effect size difference, Cohen's *d* = 0.37, on negative metacognitive beliefs favoring this subgroup. This highlights that there may be a slight advantage for individuals seeking treatment for anxiety and depression on outcomes but this may also be a function of greater initial severity in the comorbid cases.

# Clinically Reliable Change

Reliable change was calculated for the total score of the Hospital Anxiety and Depression Scale for treatment completers. As the


TABLE 4 | Number and percentage of completers that reliably changed.

HADS has varying test-retest reliability scores and few have been calculated for the HADS total score, the average test-retest coefficient for the HADS total was calculated from Michopoulos et al. (2008) who reported a test-retest coefficient of 0.944 and from Spinhoven et al. (1997) who reported a test-retest coefficient of 0.91. Both test-retest coefficients were calculated over a 3-week interval. In order for patients to be classified as having made a reliable change, they had to have made at least a change of 6 points on the HADS total. A cutoff score of 15 was calculated using criterion "c" as outlined by Jacobson and Truax (1991) and used normative data from Crawford et al. (2001). Participants were classified as being improved if they made a reliable change but did not cross the cutoff, were classified as recovered if they made a reliable change and crossed the cutoff, were classified as no change if they did not make a reliable change, and as worsened if they reliably worsened. **Table 4** outlines the number of participants that were classified at post-treatment and at 6-month follow-up. At post–treatment, 20.4% had made no change, 12.9% had improved, 66.7% had recovered, and none had worsened. At 6-month follow-up, 17.2% had made no change, 12.9% had improved, 65.6% had recovered, and 4.3% had worsened from pre-treatment to follow-up.

# DISCUSSION

Until now, most transdiagnostic interventions have not been derived from evidence-based generic models of psychological disorder that articulate common causal factors, but on pragmatic transdiagnostic manuals, which may have contributed to the small to moderate treatment effect sizes observed (Norton, 2008; Norton and Philipp, 2008; Newby et al., 2015). Therefore, we aimed to collate data from a mixed outpatient sample to assess the effects associated with transdiagnostic group MCT, which is based on a highly specified model. The treatment was associated with large effects that were consistent across patient subgroups and across measures. However, effect sizes should be treated with caution, as there was no comparison group. Irrespective of the participants' reason for seeking treatment, the MCT intervention was associated with significant decreases in symptoms of anxiety and depression from pre- to post-treatment and these treatment gains were maintained at a group level over 6-month follow-up. Group MCT was associated with clinically significant changes with 80% of treatment completers having recovered or improved by post-treatment and 79% remaining recovered or improved at 6-month follow-up.


Group MCT has previously been evaluated in Generalized Anxiety disorder and Major Depressive disorder. van der Heiden et al. (2013) evaluated group MCT for individuals with GAD and found large (Cohen's *d* = 2.01) pre-post treatment effect sizes on general anxiety. Similarly, group MCT has demonstrated large treatment effects for depression (Dammen et al., 2015); therefore, results from the current analysis are in line with previous studies evaluating group MCT.

In comparison to other trials of transdiagnostic treatment, the results from the current study offer promising support for the efficacy and potential superiority of group MCT in transdiagnostic groups. Effect sizes (ES) from previous transdiagnostic evaluations such as TD-CBT vary, ranging from small (Cohen's *d* = 0.09, 0.20; Erickson et al., 2007; Norton and Barerra, 2012) to large (Cohen's *d* = 0.93, 1.05, 1.15; Barlow et al., 2017 Laposa et al., 2017; Schmidt et al., 2012). While for mindfulness interventions, ES on symptoms of anxiety (hedge's *g* = 0.08–0.56) and depression (hedge's *g* = 0.22–0.59) (De Vibe et al., 2017) are low to moderate. In comparison, the current study demonstrated larger effect sizes at post-treatment ranging from Cohen's *d* = 1.68 for individuals seeking treatment for anxiety to Cohen's *d* = 1.82 for individuals with Norton both anxiety and depression. The results provide promising support for group MCT especially as the study was conducted within an unselected outpatient clinic. The results are also in line with previous studies of transdiagnostic evaluations of group MCT (Capobianco et al., 2018) that demonstrated large effects sizes at post-treatment (Cohen's *d* = 1.38) and high recovery rates (71% of participants classified as improved at post-treatment).

The strengths of the current study include the use of a heterogenous group of patients and few exclusion criteria, meaning that the results have good generalizability to natural clinical settings. The large overall sample size provides a strong basis for generalizing to other groups of self-selected patients. The study however is not without its limitations. First, we did not use formal diagnoses as participants self-reported their reasons for seeking treatment and therefore we cannot determine whether the self-diagnoses actually represent *bona fide* disorders. However, the range in HADS scores, and scores on the PHQ-9 and GAD-7, show that patients were typically reporting levels of distress within the clinical range. A second limitation is the lack of a comparison or control group which means we cannot be sure that MCT was responsible for the improvement in symptoms and we cannot

# REFERENCES


partial out the effects linked to time such as spontaneous remission. Spontaneous remission rates for anxiety are low (Bruce et al., 2005) while for depression spontaneous remission rates are high. Krøgsboll et al. (2009) found that 35% of improvement in depression could be attributed to spontaneous remission. However, given that the recovery rates for the study are higher than this at both post-treatment (67% recovered) and follow-up (66% recovered), the effects are much greater than would be expected form spontaneous improvements.

The preliminary findings from this study indicate that group transdiagnostic MCT in a sample of help-seeking patients with a mixture of psychological problems was associated with significant clinical gains that were not influenced by the nature of self-reported problems (or comorbidity). These results provide important pilot data for planning a more definitive randomized trial. If it can be substantiated that MCT is responsible for these effects, this treatment would constitute a cost-effective approach for treating mixed groups of patients suffering from a range of disorders.

# ETHICS STATEMENT

Ethical approval was not required for the study as per applicable institutional and national guidelines and regulations. Data were collected as part of routine clinical practice and evaluated as part of a service audit. However, in accordance with clinical guidelines, patients provided informed and written consent for use of patient data and ethical standards for reporting were adhered to. This is in line with the rules and regulations of the Danish National Ethics Committee.

# AUTHOR CONTRIBUTIONS

PC, CJ, and SFN were responsible for data collection and delivery of the therapy. The study is part of the doctoral thesis completed by PC. LC was responsible for data entry and analysis and contributed to the write up of the manuscript. CH was responsible for data analysis and also contributed to the write up of the manuscript. AW was responsible for overall supervision of the study, analysis, and contributed to the write up of the manuscript.

disorders in a large clinical sample. *J. Abnorm. Psychol.* 110, 585–599. doi: 10.1037/0021-843X.110.4.585


depression and anxiety (NESDA). *J. Clin. Psychiatry* 72, 341–348. doi: 10.4088/ JCP.10m06176blu


Wilamowska, Z. A., Thompson-Hollands, J., Fairholme, C. P., Ellard, K. K., Farchione, T. J., and Barlow, D. H. (2010). Conceptual background, development, and preliminary data from the unified protocol for transdiagnostic treatment of emotional disorders. *Depress. Anxiety* 27, 882–890. doi: 10.1002/da.20735

Zigmond, A. S., and Snaith, R. P. (1983). The hospital anxiety and depression scale. *Acta Psychiatr. Scand.* 67, 361–370. doi: 10.1111/j.1600-0447.1983. tb09716.x

**Conflict of Interest Statement:** This evaluation was conducted for partial completion of the first author's PhD, which was supervised by Professor AW.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2019 Callesen, Capobianco, Heal, Juul, Find Nielsen and Wells. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Group Cognitive-Behavior Therapy or Group Metacognitive Therapy for Obsessive-Compulsive Disorder? Benchmarking and Comparative Effectiveness in a Routine Clinical Service

Costas Papageorgiou<sup>1</sup> \*, Karen Carlile<sup>1</sup> , Sue Thorgaard<sup>1</sup> , Howard Waring<sup>1</sup> , Justin Haslam<sup>1</sup> , Louise Horne<sup>2</sup> and Adrian Wells<sup>3</sup>

<sup>1</sup> The Priory Hospital Altrincham, Altrincham, United Kingdom, <sup>2</sup> Mersey Care NHS Foundation Trust, Ashworth Hospital, Liverpool, United Kingdom, <sup>3</sup> Greater Manchester Mental Health NHS Foundation Trust, University of Manchester, Manchester, United Kingdom

#### Edited by:

Roberto Cattivelli, Istituto Auxologico Italiano (IRCCS), Italy

#### Reviewed by:

Michael Simons, RWTH Aachen Universität, Germany Ana Nikcevic, Kingston University, United Kingdom

\*Correspondence: Costas Papageorgiou costas@costaspapageorgiou.com

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 07 July 2018 Accepted: 28 November 2018 Published: 10 December 2018

#### Citation:

Papageorgiou C, Carlile K, Thorgaard S, Waring H, Haslam J, Horne L and Wells A (2018) Group Cognitive-Behavior Therapy or Group Metacognitive Therapy for Obsessive-Compulsive Disorder? Benchmarking and Comparative Effectiveness in a Routine Clinical Service. Front. Psychol. 9:2551. doi: 10.3389/fpsyg.2018.02551 Cognitive-behavior therapy (CBT), delivered in an individual or group format, is the recommended treatment of choice for Obsessive-Compulsive Disorder (OCD), but no studies have benchmarked the outcomes for group CBT in real-world clinical settings. The first aim of this evaluation was to benchmark the outcomes for group CBT in a sample of 125 patients who attended a routine clinical service for OCD. The results showed that the outcomes for the group CBT were comparable to those reported in previous treatment studies. However, consistent with the CBT for OCD literature, 28% of patients receiving CBT reported minimal improvement. The second aim of this evaluation was to carry out a benchmarking analysis for group metacognitive therapy (MCT) to determine if this could provide any advantages in a sample of 95 patients who also attended this clinical service over a subsequent period. The clinically significant results obtained for group MCT improved upon or equaled those obtained for group CBT and those typically found in treatment studies. The group MCT cohort improved significantly more than the group CBT cohort even after controlling for important pre-treatment variables including age, gender, number of diagnoses, symptoms of depression, and psychotropic medication. MCT had significantly higher clinical response rates. Based on international expert consensus criteria, 86.3% of patients in the MCT cohort responded compared with 64% in CBT. The implications of these findings are discussed.

Keywords: cognitive-behavior therapy, metacognitive therapy, group therapy, obsessive-compulsive disorder, benchmarking, effectiveness, routine practice

# INTRODUCTION

Obsessive-Compulsive Disorder (OCD) is a common, debilitating, and chronic mental health problem. Epidemiological studies have estimated the lifetime prevalence of OCD to be approximately 2%, with most individuals with OCD being affected before their mid-twenties (Kessler et al., 2005). OCD has been ranked among the 10 most debilitating disorders in the world

(World Health Organization, 1999). Once developed, OCD tends to have a continuous course in the majority of individuals (84%) and deteriorating (14%) or episodic (2%) courses in others (Rasmussen and Tsuang, 1986). Therefore, in the absence of effective treatment, OCD can persist for many years causing significant functional impairments and reduced quality of life (Koran et al., 1996).

The currently recommended psychological treatment of choice for OCD is cognitive-behavioral therapy (CBT; National Institute for Health Clinical Excellence, 2005), which comprises exposure and response prevention (ERP) with or without OCDfocused cognitive therapy (CT). Meta-analytic studies on the effects of psychological treatments for OCD have concluded that CBT has the highest degree of empirical support (e.g., Rosa-Alcázar et al., 2008; Olatunji et al., 2013; Öst et al., 2015). Öst et al. (2015) conducted the most recent and extensive metaanalysis, which included all randomized controlled trials (RCTs) of CBT for OCD, and the results supported the effectiveness of ERP with or without specific CT strategies with very large effect sizes (ES) for the comparisons of CBT with waiting list (1.31) and placebo conditions (1.33). In addition, other previous metaanalyses focusing only on group CBT for OCD found a mean ES of 1.12 compared with waiting list, which indicates that group CBT is an effective format (Jónsson and Hougaard, 2009). Of particular relevance to the present paper, the results of Öst et al. (2015) also showed that the ES for the comparisons between individual and group interventions (0.17) were small and nonsignificant. Therefore, the empirical evidence shows that CBT is currently the most effective psychological intervention for OCD and the format of CBT (i.e., individual or group) does not affect its outcome.

Whilst the efficacy of CBT for OCD has been established through a number of RCTs, which possess strong internal validity, the generalizability of the findings from these research studies to routine clinical practice is rather limited due to the rigid methodological features of such experimental designs. A central tenet of evidence-based healthcare is a requirement for the objective evaluation of health service interventions for their provision in clinical practice (Sackett et al., 1996). Given the phenomenological characteristics of OCD, such as chronicity and comorbidity (Kessler et al., 2005), it is imperative to determine whether the results from RCTs may be translated into realworld clinical settings. An effective method of achieving this is through benchmarking, which is a type of clinical audit that seeks to examine and improve the quality of treatment by comparing outcomes of a routinely delivered clinical service to those obtained in RCTs. To date, there have been only a few published studies that have benchmarked outcomes for CBT for adults with OCD (Franklin et al., 2000; Rothbaum and Shahar, 2000; Warren and Thomas, 2001; Houghton et al., 2010). Collectively, these studies provide some initial evidence of outcomes comparable to those falling within the benchmarks derived from previous relevant RCTs, but these few studies have a number of key limitations such as small sample sizes of self-selected participants and comparisons with only a limited number of RCTs. Importantly, considering the documented cost and clinical effectiveness of group CBT for OCD, none of the CBT interventions reported in the published benchmarking studies appear to have been delivered in group formats. Therefore, our first aim was to carry out a systematic benchmarking analysis of the treatment outcomes for group CBT for adults who had attended a routine clinical service for OCD in a mental health hospital over a 5-year period. Subsequently, in view of the results of this benchmarking analysis, our second aim was to examine the relative effectiveness associated with introducing an alternative psychological treatment approach: metacognitive therapy (MCT; Wells, 2009), which was also delivered in group formats, and to systematically benchmark this approach. The introduction and evaluation of alternative treatments is clearly supported by the literature that shows that more than a third of patients with OCD have a minimal or no response to CBT or continue to have significant residual symptoms (e.g., Wilhelm, 2000; Fisher and Wells, 2005b).

Metacognitive therapy for OCD developed from a specific metacognitive model of OCD (Wells, 1997, 2009), which was originally grounded on the generic Self-Regulatory Executive Function model (Wells and Matthews, 1994, 1996), where metacognition has prominence in explaining the development and maintenance of emotional disorders. According to the metacognitive model of OCD (Wells, 1997), the experience of intrusive thoughts, which are both universal phenomena but also cardinal clinical features of OCD, is linked with underlying metacognitive beliefs which in turn guide maladaptive thinking referred to as the cognitive attentional syndrome (CAS). The two domains of metacognitive beliefs include (1) beliefs about the significance or dangerousness of intrusive thoughts/feelings and (2) beliefs about the need to perform rituals. The first domain of metacognitive beliefs, also termed fusion beliefs, include: thought-event fusion, the belief that the occurrence of a thought can cause events to happen or that an event has already happened; thought-action fusion, the belief that thoughts alone can make a person carry out unwanted actions or behaviors; and thought-object fusion, the belief that thoughts or feelings can be transferred into objects. Metacognitive beliefs lead to worry and rumination in response to inner cognitive events (e.g., intrusive thoughts), resulting in sustained emotional distress. The second domain, beliefs about rituals, guide responses to these worries and can be expressed in a declarative form (e.g., "I must wash until I stop thinking about germs") or as a plan for monitoring action, which is indicated by a stop criterion or a "stop signal." In Wells' metacognitive model of OCD, the CAS consists of worry, rumination, threat monitoring, and maladaptive behaviors in the form of overt and covert rituals, all of which serve as means of coping with worry linked to obsessions. Whilst this model may be considered as an appraisal theory of OCD, it is distinct in that the nature of the negative appraisal is defined by the CAS and beliefs are solely metacognitive. In contrast, in CBT multiple belief domains are involved including inflated responsibility (Salkovskis, 1985, 1999; Rachman, 1993), intolerance of uncertainty (Carr, 1974), perfectionism (Frost and Steketee, 1997), overestimation of threat and importance of and need to control thoughts (Obsessive Compulsive Cognitions Working Group, 1997). MCT does not prioritize these beliefs but focuses only on metacognitive beliefs about thoughts and

beliefs about rituals. CBT does not formulate beliefs about rituals.

Cross-sectional, prospective, and experimental studies in both clinical and non-clinical populations provide support for the metacognitive model of OCD. Metacognitive beliefs in general, and fusion beliefs in particular, correlate positively with OCD symptoms in non-clinical samples (Cartwright-Hatton and Wells, 1997; Wells and Papageorgiou, 1998; Emmelkamp and Aardema, 1999; Sica et al., 2007). Furthermore, metacognitive beliefs are stronger predictors of OCD symptoms than cognitive beliefs, such as responsibility, intolerance of uncertainty, perfectionism, which explain little or no additional variance (Gwilliam et al., 2004; Myers and Wells, 2005; Myers et al., 2009a). In a prospective study, Myers et al. (2009b) found that, when statistically controlling for worry and overestimation of threat, only fusion beliefs emerged as a significant independent predictor of obsessive-compulsive symptoms, but other beliefs did not. In a routine treatment study, Solem et al. (2009) found that changes in metacognitive beliefs were a better predictor of outcomes than changes in responsibility and perfectionism among patients receiving ERP even after controlling for cognitive factors. Subsequently, Grøtte et al. (2015)replicated and extended this study by using a larger clinical sample and specific measures of metacognition assessing fusion beliefs and beliefs about rituals. Therefore, there is considerable empirical evidence to support the metacognitive model of OCD and the specific and direct role that metacognition plays over and above cognition.

Metacognitive therapy for OCD (Wells, 1997, 2009) directly focuses on modifying metacognitive beliefs and beliefs about rituals. Empirical evidence supporting MCT for OCD has derived from experimental component studies (Fisher and Wells, 2005a). Evidence supporting full MCT for OCD has derived from single case series in children and adolescents (Simons et al., 2006) as well as adults receiving this treatment in both an individual (Fisher and Wells, 2008; Van der Heiden et al., 2016) and group (Rees and van Koesveld, 2008) format. In addition, an RCT comparing individual MCT for adults with OCD with combined MCT and a medication (fluvoxamine) condition has also provided evidence supporting the intervention (Shareh et al., 2010). These studies obtained clinically significant results equal or better to those typically found in RCTs of CBT for OCD (Fisher and Wells, 2005b). In the present evaluation, we aimed to benchmark our usual group CBT for OCD and carry out a further benchmarking analysis for group MCT to determine if this could provide any clinical advantages in a subsequent cohort of adults who attended the same service for OCD in a mental health hospital over a subsequent 5-year period.

# MATERIALS AND METHODS

# Design

This is a benchmarking analysis or clinical/quality audit of a prospectively, routinely delivered clinical service involving treatment as usual (CBT) or MCT for patients with OCD. In view of this, review and approval by a relevant research ethics committee was not required according to institutional or national guidelines.

# Patients

Patients were individuals who were consecutively referred by General Practitioners or Consultant Psychiatrists to a clinical service for OCD in an independent mental health hospital in the North West of England. The suitability to attend this service offering group psychological treatment for OCD was based primarily on patients being 18 years or older and meeting primary DSM-IV (American Psychiatric Association, 1994) criteria for OCD without concurrent diagnoses of organic mental disorders, substance-related disorders, anorexia, mania or psychosis. Unlike research treatment studies, suitability for this routine service was not based on factors such as severity, comorbidity, specific treatment history, motivation, or concomitant pharmacotherapy. During the first 5-year period, a total of 181 patients were referred to the service and 172 of them agreed to attend an initial assessment of suitability. The reasons given for not attending the initial assessment were due to work/university, family, funding or unknown issues. Of the 172 patients who attended for an initial assessment, 166 patients were suitable for the service and agreed to take part in the group treatment. Of these 166 patients, 18 did not attend any of the treatment sessions due to work/university (n = 10), family/health (n = 4), funding (n = 2), or unknown (n = 2) reasons and 23 did not consent for their clinical data to be used for purposes of clinical/quality audit. Note that only the data from patients who had provided written informed consent was used for these purposes. Therefore, the group CBT cohort described here refers to the data from the 125 patients who consented and participated in the service offering group CBT for OCD over this time period. **Table 1** shows the demographic and clinical characteristics of the CBT cohort.

During the subsequent 5 years allotted to MCT, a total of 152 patients were referred to the service and 146 of them agreed to attend the initial assessment of suitability. The reasons given for not attending this initial assessment were due to work/university, illness, funding or unknown issues. Of the 146 patients who attended the initial assessment, 142 patients were suitable and agreed to take part in the group intervention. Of these 142 patients, 14 did not attend any of the sessions due to work/university (n = 9), family/health (n = 1), funding (n = 3), or unknown (n = 1) reasons and 33 patients opted out to their clinical data being used for clinical/quality audit. Note also that only the data from patients who had provided written informed consent was used for this evaluation. Therefore, the group MCT cohort described here represents the data from the 95 patients who consented and participated in the service offering group MCT for OCD over this subsequent time period. **Table 1** summarizes the demographic and clinical characteristics of the MCT cohort.

# Measures

A number of self-report routine outcome measures were administered before and after each intervention. The naturalistic clinical service setting precluded collection of sufficiently appropriate long-term follow-up data, which is very common

TABLE 1 | Demographic and clinical characteristics of group treatment cohorts.


in routine clinical practice. The primary outcome measure was the severity of symptoms of OCD and the secondary outcome measures assessed depression, functional impairment, global improvement, and likelihood to recommend treatment.

# Primary Outcome Measure

The primary outcome measure was the severity of symptoms of OCD as assessed by the self-report version of the Yale-Brown Obsessive Compulsive Scale (Y-BOCS; Baer et al., 1993).

The Y-BOCS is widely considered to be the "gold standard" assessment measure in treatment outcome research in OCD (Frost et al., 1995; Fisher and Wells, 2005b). It is a 10 item measure that assesses the severity of both obsessions and compulsions across five dimensions: frequency, interference, distress, resistance, and control. The Y-BOCS has good test-retest reliability and internal consistency with Cronbach alphas of 0.89 in a non-clinical sample and 0.78 in an OCD sample (Steketee et al., 1996).

# Secondary Outcome Measures

The Beck Depression Inventory (BDI; Beck et al., 1961) was used to assess symptoms of depression. The BDI is a widely used 21 item scale that assesses the presence and severity of depressive symptoms over the previous week using a 4-point severity scale. The reported Cronbach alpha is 0.89 (Beck et al., 1961).

The Work and Social Adjustment Scale (WSAS; Mundt et al., 2002) was used as a measure of functional impairment associated with OCD. The WSAS is a 5-item scale that assesses the degree of impairment in functioning over the previous week using a 9 point rating scale. The reported Cronbach alphas ranged from 0.77 to 0.90 (Pedersen et al., 2017).

In addition to the above measures, patients were asked to complete two further ratings at post-treatment. One of these ratings was a self-report adaptation of Guy (1976) clinicianrated Clinical Global Impression-Improvement (CGI-I) scale, which was developed to be used in pharmacotherapy research trials to provide brief assessments of patient improvements. In the adapted version of this scale, the Self-Ratings of Global Improvement Scale (SRGIS), asked patients to rate on a 7-point scale their response to the following: "Compared to your initial OCD problems just before you started the OCD Treatment Program, please circle a number below to indicate how much you have improved." Patients indicated their response by choosing one of the following: 1 = very much improved, 2 = much improved, 3 = minimally improved, 4 = no change, 5 = minimally worse, 6 = much worse, 7 = very much worse. In the other post-treatment rating, patients were asked to "indicate how likely you are to recommend the OCD Treatment Program you have completed for someone who might be suffering from OCD" by using a rating scale ranging from 0 (I would not recommend it) to 100 (I would definitely recommend it).

# Procedure

Following referral to the clinical service for OCD, all patients were sent a pack containing the following: (1) a letter offering them "an appointment to attend an initial psychological assessment interview with a view to participating in the OCD Treatment Program" and requesting completion of enclosed measures; (2) registration and consent forms; and (3) the battery of pre-treatment measures. The consent form asked patients to decide whether or not to give the hospital permission for their "clinical data to be used anonymously for purposes of clinical/quality audit." All patients attending this interview were assessed by the first author for suitability for the service, which involved diagnostic screening using the Structured Clinical Interview for DSM-IV Axis I Disorders - Patient Edition (SCID-I/P; First et al., 1997). If patients were suitable to attend the service, they were informed at the end of the interview and provided with details of the nature of the respective treatment, including duration, facilitation, and format. This allowed for opportunities to address any specific concerns raised by patients about treatment including apprehension about the group format or expectations about attendance and participation. The patients who agreed to take part in the group treatment were then informed about start dates and encouraged to actively focus on this treatment whilst participating. Patients who had been, or were going to be, prescribed any psychotropic medication were also encouraged to ensure that adequate clinical management of their medication from their General Practitioner and/or Consultant Psychiatrist was regularly in place throughout their group treatment participation. Patients then waited between approximately 1 day and 3 weeks before commencing treatment.

CBT and MCT were delivered in group formats jointly by the first and second authors and each group treatment consisted of 12 2-h weekly sessions over a period of 4 months. The first author is a Clinical Lead and Consultant Clinical Psychologist with extensive training and experience in CBT and MCT for OCD. The second author used to be a serviceuser when she initially attended for individual treatment for OCD. Since achieving full recovery following CBT 14 years ago, she has been co-facilitating each group treatment session over the entire period of the clinical service for OCD and gaining considerable experience under supervision. CBT for OCD followed the treatment approach advocated by Salkovskis and Kirk (1989, 1997) but also that of Wilhelm and Steketee (2006) in order to comprehensively extend the focus beyond inflated responsibility and to other cognitive domains implicated in OCD such as overestimation of threat, intolerance of uncertainty, and perfectionism. MCT for OCD followed the treatment approach of Wells (1997, 2009) and the published treatment protocol (Wells, 2009). The delivery rather than the content of each treatment modality was adapted for use in the group format. Common to both interventions was the content of sessions 1, 8, and 12 where the primary focus was on psychoeducation about OCD and its treatment and motivational enhancement (session 1), how significant others (a family member, friend, or colleague of each patient attended this session) could support the patient in maximizing therapeutic gains (session 8), and therapy blueprint and relapse prevention (session 12). There were other common general features of the two treatments including conceptualization, socialization, exposure to feared stimuli, and verbal and behavioral reattribution strategies were used to change beliefs and behaviors, but for each of these features the content and focus was different. Specifically, during CBT the focus was on extensively challenging relevant cognitive belief domains and implementing self-directed ERP whilst the focus during MCT was to challenge metacognitive beliefs in OCD (i.e., metacognitive beliefs about intrusions and beliefs about rituals and stop signals). In addition, during MCT patients were introduced to detached mindfulness as an alternative means of responding to their intrusions and instructed to postpone worry and rumination. MCT implemented metacognitively focused exposure aimed at testing fusion beliefs. At session 12, all patients were re-administered the Y-BOCS, BDI, and WSAS and they were also asked to provide ratings of global improvement using the SRGIS and ratings of likelihood to recommend treatment.

# Overview of Analyses

fpsyg-09-02551 December 6, 2018 Time: 15:8 # 6

We examined the outcomes of the group CBT and the group MCT against other previous research treatment studies of CBT to gauge the relative effects of these interventions when delivered in routine clinical practice. Statistical analyses were conducted using within-subjects t-tests to examine changes in outcome variables within each group treatment. Mixed model ANCOVAs were computed to examine differences in improvement in Y-BOCS between the CBT and MCT interventions. These were followed by between-group ANCOVAs on post-treatment variables. In non-randomized evaluations like this, it is important to control for potential threats to internal validity that are not minimized by a randomization method. Therefore, we controlled for the following pre-treatment factors: age, gender, number of diagnoses, symptoms of depression, and medication status in all of the mixed model analyses with additional controls of the pre-treatment Y-BOCS in the post-treatment betweengroups ANCOVAs. We did not control for WSAS when assessing Y-BOCS outcomes because of the measurement overlap as both scales assess interference or disability associated with OCD. Of most relevance to service provision, the clinical significance of the effects of each treatment was examined and compared using international expert consensus criteria for OCD.

# RESULTS

# Benchmarking of Treatment Outcomes for the Group CBT Cohort

The demographic and clinical characteristics of the CBT cohort are shown in **Table 1**. In comparison to those reported in previous RCTs and other research treatment studies of CBT for OCD (for reviews, see Jónsson and Hougaard, 2009; Öst et al., 2015), the group CBT cohort had a more balanced gender distribution, considerably higher number of referrals from secondary care (i.e., Consultant Psychiatrists), more comorbidity, and greater number of patients who were prescribed psychotropic medication. The remaining demographic and clinical characteristics of the CBT cohort were consistent with previously published data. On the whole, our CBT cohort seemed to be a group of patients with more complex OCD presentations than those previously reported in treatment studies.

We next examined the attrition rates for the entire course of CBT. Attrition was defined as a patient who takes part in at least the first group treatment session, but then withdraws before completion of the intervention (Öst et al., 2015). During the 5 year course of the group CBT, 12 (9.6%) patients dropped out of this intervention. This compares relatively well to the previously reported drop out rates, which have ranged from 11.4% for CT to 32% for the combined ERP, CT and antidepressant medication (Öst et al., 2015). The analyses presented here are based on intention to treat. Therefore, the attrition rate for the group CBT suggests that patients found this intervention acceptable. In addition, the mean number of group CBT sessions attended was 11.42 (SD = 0.86, range: 8–12) in mean group sizes of 7.7 (SD = 1.71, range: 6–11), and both of these sets of data are consistent with those reported in previous group CBT for OCD studies (Jónsson and Hougaard, 2009).

The descriptive and summary statistics for the primary and secondary outcome measures before and after each group CBT intervention are presented in **Table 2**. At pre-treatment, the CBT cohort displayed mean Y-BOCS scores indicating severe obsessive-compulsive symptoms and the mean BDI and WSAS scores were suggestive of moderate levels of depression and functional impairments, respectively. As shown in **Table 2**, all of these scores decreased from pre-treatment to post-treatment. The repeated measures t-tests indicated that these within-group changes were all significant in terms of Y-BOCS [t(124) = 24.52, p < 0.0005], BDI [t(124) = 13.35, p < 0.0005], and WSAS [t(124) = 12.35, p < 0.0005].

It is well-known that antidepressant medication is effective in the treatment of OCD (e.g., Soomro et al., 2008). Therefore, because a large proportion of the patients in the CBT cohort were taking medication, we examined within-group pre-post effect sizes (Hedges' g) based on the Y-BOCS scores for those with and without medication to estimate any effects associated with the combined treatment. In the medicated sub-group (n = 98), the effect size was 2.39 compared to the non-medicated subgroup (n = 27), which was 2.59. It is important to note that the medicated sub-group displayed more severe pre-treatment symptoms of OCD that the non-medicated sub-group.

For the entire CBT cohort, the resulting pre-post Y-BOCS change score was 13.28. This compares favorably to those reported in previous CBT for OCD studies, which have ranged from 5 to 12.6 (Jónsson and Hougaard, 2009). In addition, the within-group pre-post effect size (Hedges' g) based on the Y-BOCS scores was very large (ES = 2.38), and this was also comparable to those previously reported, which have ranged from 1.47 for medication, 2.06 for ERP, 2.21 for CT, and 2.95 for combined ERP, CT and medication (Öst et al., 2015). We computed the self-ratings of global improvement data and the results indicated that 25 (20%) of patients rated their improvement following group CBT as "very much improved," 65 (52%) as "much improved," and 35 (28%) as "minimally improved."

Finally, we examined the extent (0–100%) to which patients were likely to recommend the group CBT for someone who might be suffering from OCD. At post-treatment, the mean score for this scale was 92.96 (SD = 11.71). This indicates a significant degree of satisfaction with the treatment experienced, as patients were highly likely to recommend it.

# Benchmarking of Treatment Outcomes for the Group MCT Cohort

**Table 1** displays the demographic and clinical characteristics of the group MCT cohort. In comparison to those reported in previous studies of CBT for OCD (Jónsson and Hougaard, 2009; Öst et al., 2015), the MCT cohort was slightly younger, had a more


TABLE 2 | Means, SD (in parentheses), and summary statistics for the primary and secondary outcome measures before and after each group treatment.

Y-BOCS, self-report version of the Yale-Brown Obsessive Compulsive Scale; BDI, Beck Depression Inventory; WSAS, Work and Social Adjustment Scale.

balanced gender distribution, substantially more referrals from secondary care, higher comorbidity, and more patients taking psychotropic medication. The duration of OCD was consistent with previously reported data. **Table 1** shows that compared to the CBT cohort, the MCT cohort was significantly younger. Therefore, the group MCT cohort also seemed to be a group of patients with more complex OCD presentations than those previously reported in treatment studies.

During the 5-years running of group MCT, 7 patients (7.4%) dropped out of this therapy. This compares relatively well to the previously reported drop out rates in CT and for the combined ERP, CT, and medication treatment, but also to the drop out rates found for our CBT cohort although there was no significant difference in drop out rates between the treatment cohorts [χ 2 (1) = 0.58, p = 0.447]. The analyses shown here are based on intention to treat. Therefore, this attrition rate indicates that patients found this intervention acceptable. In addition, the mean number of group MCT sessions attended was 11.33 (SD = 0.95, range: 8–12) with mean group sizes of 7.75 (SD = 1.85, range: 5–10), and both of these sets of data were comparable with those published in previous group CBT for OCD studies (Jónsson and Hougaard, 2009).

**Table 2** shows that at pre-treatment the MCT cohort displayed mean Y-BOCS scores indicating severe obsessive-compulsive symptoms and the mean BDI and WSAS scores were suggestive of moderate levels of depression and functional impairments, respectively. As shown in **Table 2**, all of these scores decreased from pre-treatment to post-treatment. The repeated measures t-tests indicated that these within-group changes were all significant in terms of Y-BOCS [t(94) = 24.06, p < 0.0005], BDI [t(94) = 11.52, p < 0.0005], and WSAS [t(94) = 13.66, p < 0.0005].

Similar to the CBT cohort, a large proportion of the patients in the MCT cohort were taking medication. Therefore, we also examined within-group pre-post effect sizes (Hedges' g) based on the Y-BOCS scores for those with and without medication to estimate any effects associated with the combined treatment. In the medicated sub-group (n = 75), the effect size was 2.81 compared to the non-medicated sub-group (n = 20), which was 3.57. It is also noteworthy that the medicated sub-group displayed more severe pre-treatment symptoms of OCD that the non-medicated sub-group.

For the entire MCT cohort, the pre-post Y-BOCS change score was 15.23, which is highly comparable to those reported in previous CBT for OCD studies (Jónsson and Hougaard, 2009). In addition, the within-group MCT pre-post effect size (Hedges' g) based on the Y-BOCS scores was very large (ES = 2.89), which was comparable to those previously reported, almost equating to the ES of 2.95 for the combined ERP, CT and medication (Öst et al., 2015). We then computed the SRGIS data and the analyses indicated that 24 (25.3%) of patients rated their improvement after the group MCT as "very much improved," 62 (65.3%) as "much improved," and 9 (9.4%) as "minimally improved."

Finally, we examined the extent to which patients were likely to recommend the group MCT for someone who might be suffering from OCD. At post-treatment, the mean score for this scale was 95.26 (SD = 8.1). This indicates a significant degree of satisfaction with the treatment experienced, as patients were highly likely to recommend it.

# Comparisons Between Group CBT and Group MCT on Primary and Secondary Variables

There were no significant differences between the CBT and MCT cohorts in terms of mean number of sessions attended [t(218) = 0.79, p = 0.214] or the mean group sizes [t(218) = −0.21, p = 0.416]. When comparing patients' ratings of likelihood to recommend treatment, the analyses indicated that there was no significant difference between the group CBT and group MCT in terms of these ratings [t(218) = −1.64, p = 0.062)]. This would imply that any actual outcome differences between the two group treatments are less likely to be due to non-specific factors such as satisfaction, acceptability or credibility. However, as **Tables 1**, **2** show there were significant pre-treatment differences in terms of age, Y-BOCS, BDI, and WSAS.

A mixed model ANCOVA with cohort (CBT vs. MCT) as the between-groups factor and time (pre-treatment and posttreatment) as the repeated-measures factor was computed on the primary outcome variable (i.e., Y-BOCS). The covariates were the following pre-treatment variables: age, gender, number of diagnoses, BDI, and medication status. There was a significant interaction involving group and time [F(1, 213) = 4.03, p = 0.046], which showed that the MCT cohort improved significantly more than the CBT cohort over the 12-week course of treatment. Follow-up between-group ANCOVA controlling for pre-treatment Y-BOCS and all other covariates (i.e., age,

gender, number of diagnoses, BDI, and medication) showed no significant post-treatment group differences on Y-BOCS score [F(1, 212) = 1.09, p = 0.296]. Therefore, at post-treatment the Y-BOCS scores were similar but the MCT cohort showed a greater level of improvement than the CBT cohort over time.

When statistically controlling for pre-treatment age, gender, number of diagnoses, BDI, and medication status, mixed model ANCOVAs demonstrated no significant group by time effects on the following post-treatment outcomes: BDI [F(1, 214) = 0.79, p = 0.374] and WSAS [F(1, 213) = 3.69, p = 0.056]. Moreover, when controlling for WSAS at pre-treatment and all other pre-treatment variables, follow-up ANCOVAs on posttreatment WSAS score showed that the group effect was not significant [F(1, 212) = 0.47, p = 0.494]. Similarly, for the BDI at post-treatment when controlling for pre-treatment BDI, Y-BOCS and the other covariates, the group effect was not significant [F(1, 212) = 0.04, p = 0.852]. However, when statistically controlling for pre-treatment Y-BOCS, age, gender, number of diagnoses, BDI, and medication, ANCOVA on the patients' ratings of global improvement indicated greater improvement following MCT than CBT [F(1, 212) = 8.37, p = 0.004].

In order to examine the relative clinical significance of the group interventions, we applied the international expert consensus criteria for defining treatment response, remission, and recovery in OCD (Mataix-Cols et al., 2016). The consensus definitions involve a twofold criterion and can be operationalized as follows: response is defined as a ≥35% reduction in Y-BOCS scores plus CGI-I rating of 1 ("very much improved") or 2 ("much improved") lasting for at least 1 week; partial response is defined as a ≥25% but <35% reduction in Y-BOCS scores plus CGI-I rating of at least 3 ("minimally improved") lasting for at least 1 week; and it is assumed that no response is defined as <25% reduction in Y-BOCS scores. In the absence of CGI-I ratings, we relied on the patients' ratings of global improvement using the SRGIS, which maintains the same criteria. However, we were unable to apply the criteria to estimate rates of remission, recovery, or relapse as the criteria required the Clinical Global Impression-Severity (CGI-S) ratings, 1- and 12-month followup data, which we did not have available. **Table 3** displays the proportion of patients achieving criteria for response on Y-BOCS at post-treatment for the group treatment cohorts. The MCT cohort displayed an overall higher clinical response rate than the CBT cohort and this difference was significant [χ 2 (2) = 12.97, p = 0.0015].



Based on international expert consensus criteria for OCD (Mataix-Cols et al., 2016).

# Treatment Resource Requirements

In routine clinical services, especially those with scarce resources, the amount of treatment required to achieve a clinical response or significant clinical improvement is an important economic factor. An advantage of group treatment delivery is that a higher volume of patients can be treated over a specified period of time. Therefore, using previous formulae (i.e., number of treatment sessions x number of hours per treatment session x number of therapists divided by number of patients per group) for calculating basic cost-savings (Jónsson and Hougaard, 2009), we estimated the mean number of therapist hours required to treat each patient in each group treatment cohort. For the CBT cohort, with two therapists treating groups with a mean size of 7.7, 2 h per week over 12 weeks, equates to a total 6.23 h per patient to achieve a 64% clinical responder rate. For the MCT cohort, with two therapists treating groups with a mean size of 7.75, 2 h per week over 12 weeks, equates to 6.19 h per patient to achieve an 86.3% clinical responder rate. Clearly, if only one therapist facilitates each group session over the course of treatment, then the mean number of hours needed to treat each patient becomes 3.12 and 3.10 for the group CBT and group MCT, respectively. Both group treatments could potentially create considerably greater costeffectiveness when compared with individual therapy although it must be noted that the longer-term effects have yet to be established.

# DISCUSSION

Given that CBT is the recommended treatment of choice for OCD, but few systematic studies have documented whether the results based on this recommendation can be translated into real-world settings, our first aim was to benchmark outcomes for group CBT in a routine clinical service. In a large group of patients with relatively more complex OCD presentations than previously reported, the results demonstrated that a 12-week course of group CBT led to significant improvements in OCD, depression, and functional impairments. At post-treatment, the scores from primary and secondary outcome measures fell within normal/mild ranges. The results of benchmarking indicated that the outcomes of group CBT were equal to those found in research treatment studies (Jónsson and Hougaard, 2009; Öst et al., 2015). Of particular relevance to the results obtained is the low attrition rate found given that 78.4% of patients in the CBT cohort were prescribed medication. Studies have reported that treatments with medication alone or in combination with ERP or CT tend to produce the highest attrition rates (Öst et al., 2015). The results of this benchmarking evaluation contribute to the generalizability of the findings from research treatment studies but extend it to group CBT and more complex OCD presentations.

The results of our initial benchmarking analysis based on patients' ratings of global improvement revealed that 28% of the patients who had received group CBT reported only minimal improvement. This finding is not surprising, and consistent with literature showing that a significant proportion of patients have a minimal or no response to CBT for OCD (e.g., Wilhelm, 2000; Fisher and Wells, 2005b). It supported

our second aim to address limitations by introducing MCT for OCD and examining its comparative effectiveness. The results demonstrated that a 12-week course of group MCT led to significant improvements in OCD, depression, and functional impairments. At post-treatment, the scores from primary and secondary outcome measures fell within normal/mild ranges. The clinically significant results obtained for the group MCT appeared to be better or equal to the group CBT cohort and those typically found in RCTs of CBT for OCD, especially given the low attrition rate in a group for whom 76.8% were prescribed medication. The results of the group MCT also contribute to a growing body of empirical evidence attesting to the effectiveness of this intervention in group formats for generalized anxiety disorder in children (Esbjørn et al., 2018) and adults (Van der Heiden et al., 2013; McEvoy et al., 2015), depression (Dammen et al., 2015), antidepressant and CBT-resistant depression (Papageorgiou and Wells, 2015), and transdiagnostic patient samples (Capobianco et al., 2018).

During the course of each intervention, patients found both group treatments equally and highly acceptable and satisfactory as evidenced by equivalent low attrition rates and lack of significant differences in the number of sessions attended and the patients' treatment recommendations. However, the effect size for the MCT cohort was higher than that obtained for the CBT cohort and the patients' ratings of global improvement coupled with treatment response rates suggests that patients receiving MCT benefitted more from this intervention. The results show that the MCT cohort improved significantly more over the 12-week course than the CBT cohort after controlling for important pre-treatment variables including age, gender, number of diagnoses, symptoms of depression, and medication. Therefore, even though the Y-BOCS scores of the treatment cohorts were similar at post-treatment, the MCT cohort displayed a greater level of improvement than the CBT cohort over time. This is likely to be due to the MCT cohort having higher scores at pre-treatment as control of pretreatment Y-BOCS in the post-treatment analysis showed no differences between the conditions in final level of Y-BOCS score.

The clinical significance of the comparative findings is the most informative given the motivation to reduce the number of patients showing minimal or no response to treatment. Using the twofold international expert consensus criteria (Mataix-Cols et al., 2016) applied to the Y-BOCS and SRGIS, there was a reduction following MCT of 10.5% in non-responders when

# REFERENCES


compared with CBT, a reduction of partial responders by 11.8% but an increase in clinical responders by 22.3%. The difference in response rates was statistically significant.

Our analyses represent a naturalistic evaluation to benchmark treatment outcomes but the obvious limitations of the present evaluation are associated with the strengths of RCTs and other research treatment studies. That is, apart from the SCID-I/P, there was a lack of clinician-administered tools, untreated control conditions, treatment fidelity and adherence checks, and independent raters or assessors. Importantly, we were unable to control for type of pharmacotherapy, which would have enabled us to determine the impact of different drugs on outcome. However, examination of within-group treatment effect sizes for patients without medication suggests a greater change in MCT compared to CBT, but these sub-group analyses are based on a small number of patients. Finally, we were not able to collect any meaningful follow-up data due to the routine clinical nature of the service within an independent mental health hospital. Nevertheless, the data are likely to represent the types of outcomes that can be achieved in clinical settings.

In conclusion, both CBT and MCT were effective interventions when delivered as group treatments in a naturalistic clinical setting. We found that MCT appeared to show some advantage over CBT. Most notably, when compared to CBT, MCT appeared to reduce the rate of non-responders and partial responders whilst significantly increasing the rates of clinical response.

# ETHICS STATEMENT

This study is a benchmarking analysis or clinical audit of a prospectively, routinely delivered clinical service rather than a research treatment study. All patients in this paper provided consent for their clinical data to be used anonymously for purposes of clinical/quality audit.

# AUTHOR CONTRIBUTIONS

CP, KC, ST, HW, and JH contributed to the development of the service. CP, KC, ST, HW, JH, and LH contributed to the data collection, scoring, and recording. AW, CP, and LH contributed to the data analysis and interpretation. All authors contributed to writing the manuscript.



undergoing treatment with exposure and response prevention. Behav. Res. Ther. 47, 301–307. doi: 10.1016/j.brat.2009.01.003


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Papageorgiou, Carlile, Thorgaard, Waring, Haslam, Horne and Wells. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Group Metacognitive Therapy for Generalized Anxiety Disorder: A Pilot Feasibility Trial

Svein Haseth<sup>1</sup> \*, Stian Solem2,3, Grethe Baardsen Sørø<sup>1</sup> , Eirin Bjørnstad<sup>2</sup> , Torun Grøtte1,2 and Peter Fisher1,4

<sup>1</sup> Nidaros DPS, St. Olavs Hospital, Trondheim, Norway, <sup>2</sup> Department of Psychology, Norwegian University of Science and Technology, Trondheim, Norway, <sup>3</sup> Department of Research and Development, St. Olavs Hospital, Trondheim, Norway, 4 Institute of Psychology, Health and Society, University of Liverpool, Liverpool, United Kingdom

#### Edited by:

Lora Capobianco, Manchester Mental Health and Social Care Trust, United Kingdom

#### Reviewed by:

Bruce Fernie, King's College London, United Kingdom Costas Papageorgiou, Priory Hospital Altrincham, United Kingdom

\*Correspondence: Svein Haseth Svein.Haseth@stolav.no

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 09 August 2018 Accepted: 29 January 2019 Published: 14 February 2019

#### Citation:

Haseth S, Solem S, Sørø GB, Bjørnstad E, Grøtte T and Fisher P (2019) Group Metacognitive Therapy for Generalized Anxiety Disorder: A Pilot Feasibility Trial. Front. Psychol. 10:290. doi: 10.3389/fpsyg.2019.00290 Background: Individual metacognitive therapy (MCT) for generalized anxiety disorder (GAD) is well established, but only one study has investigated the effectiveness of Group MCT (g-MCT) for GAD. The aim of the current study was therefore to evaluate the feasibility and effectiveness of g-MCT for GAD within a community mental health setting whilst addressing limitations evident in the previous study.

Methods: The study used an open trial design, and 23 consecutively referred adults with GAD completed 10 sessions (90 min) of g-MCT, delivered by two therapists trained in MCT. Diagnoses were assessed by trained raters using the Anxiety Disorder Interview Schedule-IV. All patients but one had previous psychosocial treatment, and 17 (73.9%) had at least one comorbid axis-I disorder. Self-reported symptoms were assessed using the Penn State Worry Questionnaire, the Generalized Anxiety Disorder-7, and the Patient Health Questionnaire-9 at pre- and post-treatment as well as 3-month followup. Feasibility was assessed using rates of patients who declined group treatment in favor of individual treatment, patients not able to attend due to pre-scheduled dates for sessions, and drop-out rate.

Results: Of 32 eligible participants, six patients (19%) declined g-MCT in favor of individual MCT, and three (9%) were unable to attend due to scheduling conflicts. No patients dropped out during treatment, but two patients did not complete the selfreport questionnaires at 3-month follow-up. g-MCT was associated with significant reductions in worry, anxiety, depression, metacognitive beliefs, and maladaptive coping. According to the standardized Jacobson criteria for recovery, 65.3% were recovered at post-treatment, whereas 30.4% were improved and 4.3% showed no change. At 3-month follow-up, the recovery rate increased to 78.3%. Moreover, recovery rates were comparable for patients with- and without comorbidity. Number of therapist hours per patient was 6.5 and the treatment has now been implemented as a standard treatment option at the clinic.

Conclusion: g-MCT for GAD is an acceptable treatment which may offer a costeffective alternative approach to individual MCT. Recovery rates and effect sizes suggested that g-MCT could be just as efficient as individual MCT and cognitive behavioral therapy.

Keywords: metacognitive therapy, generalized anxiety disorder, GAD, outcome, metacognition, group therapy

# INTRODUCTION

fpsyg-10-00290 February 12, 2019 Time: 17:50 # 2

Generalized anxiety disorder (GAD) is a common disorder associated with a chronic course and significantly reduced quality of life (Spitzer et al., 2006; American Psychiatric Association, 2013). It is characterized by excessive and uncontrollable worry related to multiple events or activities, with a duration of six months or more (American Psychiatric Association, 2013). Associated symptoms include restlessness, fatigue, difficulties concentrating, irritability, muscle tension, and sleep difficulties (American Psychiatric Association, 2013).

Cognitive behavioral therapy (CBT) is currently an evidencebased treatment for GAD (Hoyer et al., 2011). Meta-analyses show that CBT leads to a reduction in anxiety symptoms more so than treatment as usual or a waiting list (Mitte, 2005; Hunot et al., 2007; Covin et al., 2008). However, based on the criteria for clinically significant change (Jacobson and Truax, 1991), only 50–60% of patients with GAD recover at 6-month follow-up after CBT (Fisher and Durham, 1999). Thus, since a considerable proportion of GAD patients do not recover following CBT, more effective interventions are required.

Metacognitive Therapy (MCT) for GAD is an alternative treatment to CBT. MCT focuses on changing thought processes rather than thought content (e.g., Wells, 1995). MCT is derived from the self-regulatory executive function (S-REF) model (Wells and Matthews, 1994, 1996). Maintenance of psychological problems is linked to the activation of the cognitive-attentional syndrome (CAS) consisting of repetitive thinking (worry and rumination), threat monitoring, and maladaptive coping behaviors. The CAS is a product of an individual's metacognitive beliefs and knowledge. Central to the metacognitive model of GAD (Wells, 1995, 1997, 2009) is that individuals' thoughts and beliefs about worry (i.e., metacognitive beliefs) contribute to the development and maintenance of the disorder. Worry is often triggered by negative intrusive thoughts in the form of "what if " questions, e.g., "What if I'm involved in an accident?". Thereafter, the use of worry is related to the activation of positive metacognitive beliefs about the advantages or benefits of worrying (Wells, 2009). Examples of such positive beliefs are "Worrying makes me prepared, and focusing on threat keep me safe."

Symptoms of GAD escalate when negative metacognitive beliefs about worry are activated. Two types of negative beliefs are important: negative beliefs about the uncontrollability of worry (e.g., "I have lost control over my thoughts") and negative beliefs about the possible dangers of worry ("If I do not stop worrying, I will lose my mind"). The activation of negative metacognitive beliefs leads to worry about worry (also called "meta-worry" or "Type 2-worry"), which intensifies worry, anxiety, and other maladaptive coping strategies. The model proposes that individuals with GAD tend to use worry as a coping strategy to safeguard against perceived threats and dangers. Examples of other frequently used coping responses among GAD patients are thought suppression, threat monitoring, distraction, avoidance, and reassurance seeking. These coping strategies backfire and consolidate the belief that worry is uncontrollable.

The metacognitive model of GAD (Wells, 1995, 1997, 2009) proposes that both positive and negative metacognitive beliefs need to be modified to enable people to disengage from worrying in response to trigger thoughts. Furthermore, the model specifies that counterproductive coping strategies need to be modified if people are to successfully reduce worry.

So far, four studies have evaluated MCT for GAD delivered individually for outpatients. Wells and King (2006) conducted an open trial (N = 10), where a range of 3–12 weekly MCT sessions were delivered. There were significant improvements in symptoms of worry, anxiety, and depression at post-treatment [within-group d's between 1.12 (health worry) and 2.78 (traitanxiety)] and follow-up (within-group d's between 1.10 and 2.58), and 87.5% of the patients met criteria for recovery on traitanxiety (STAI-T) at post-treatment, and 75% were recovered at 6- and 12-month follow-up.

The second study was conducted by Wells et al. (2010) and was a randomized controlled trial (N = 20, 10 in each condition) where MCT was compared with applied relaxation (AR) in patients with GAD. Treatment sessions lasted 45–60 min and were held once per week for 8–12 weeks. MCT was significantly more effective in reducing GAD symptoms than AR. Following criteria (Fisher and Durham, 1999) for clinically significant change (PSWQ; cut-off ≤47, reliable change index: 7), the recovery rate was 80% in the MCT group at post-treatment, compared with 10% in the AR group. At 6-month follow-up, the recovery rate was 70% in the MCT group and 10% in the AR group, while the figure was 80 and 10%, respectively, at 12-month follow-up. High recovery rates combined with a large withingroup effect size (d = 3.41) indicated that MCT was an effective treatment for GAD.

van der Heiden et al. (2012) investigated the effectiveness of MCT and intolerance of uncertainty therapy (IUT). Each treatment consisted of a maximum of 14 weekly sessions of 45 min. Both MCT and IUT were associated with significant reductions in symptoms of GAD at post-treatment and 6-month follow-up, but MCT was found to be significantly superior to IUT. The within-group effect sizes for worry (PSWQ) in the MCT group were high at both post-treatment (d = 1.67) and followup (d = 1.66), and the between-group effect sizes were 0.96 at post-treatment and 0.78 at follow-up. In the MCT intention-totreat group, 60% met criteria for recovery on PSWQ (cut-off ≤53,

reliable change index: 7) at end of treatment and 62% at followup. The corresponding recovery rates for the IUT group were 37% and 47%, respectively.

Nordahl et al. (2018) compared the efficacy of MCT and CBT for GAD. Both CBT and MCT produced significant reductions in worry (PSWQ) in comparison to the wait list group. However, MCT was found to be more effective than CBT. In the MCT condition 65% were classified as recovered post-treatment in comparison to 38% in the CBT condition, and the difference was maintained at 2-year follow-up.

In summary, previous research indicates that individual outpatient MCT for GAD is well established. According to, the (National Institute for Health and Clinical Excellence [NICE], 2011) guidelines, MCT is a recommended treatment for GAD. However, group MCT (g-MCT) for GAD has only been examined in one open trial (van der Heiden et al., 2013). This study used large groups (10–14 patients) which may limit participation of some group members and not allow therapy to be implemented with sufficient specificity to address individual needs. In addition, two out of the four therapists had not received training in MCT thereby potentially limiting treatment adherence and competency. The sample consisted of 33 outpatients, treatment sessions lasted 90 min and were held weekly for 12–14 weeks. There were significant reductions in worry, anxiety, and negative metacognitive beliefs. In the intention-to-treat sample, the between group effect sizes at post-treatment and 6-month followup were 1.24 and 1.29, respectively. In terms of recovery, 55% of participants met criteria for clinically significant criteria at posttreatment recovery rate at post-treatment (cut-off: ≤53, reliable change index: 7).

Treatment in a group can be an attractive alternative to individual treatment for several reasons. A similar effect as individual treatment will result in group treatment being more cost-effective by cutting down on long waiting lists leading to more effective use of the therapists' time. One assumption is that MCT will be well-suited to a group format because it is based on a transdiagnostic model. A recent study supported the use of g-MCT for a transdiagnostic sample (Capobianco et al., 2018). The study found that g-MCT was more effective than Mindfulness Based Stress Reduction in treating symptoms of anxiety and depression. Furthermore, patients with GAD may worry about different events, activities, life events and will frequently have different comorbid disorders, but MCT focuses on changing the attitudes and beliefs one has around thought processes (i.e., worrying and rumination) and is less concerned with the actual idiosyncratic thought content of each patient. Patients can help each other identify shared maladaptive metacognitive beliefs and coping strategies whilst their worry content differ.

Despite the appealing aspect of group treatment, a comparison of effect sizes, recovery-, and attrition rates with previous studies of individual MCT indicates that g-MCT may be less effective. Furthermore, the dropout rate was higher in g-MCT (27%) than in individual treatment studies (van der Heiden et al., 2012: 18%; Wells and King, 2006 and Wells et al., 2010: 0%). In addition to the limitations of the van der Heiden et al. (2013) study, the authors also suggested several possible reasons for the differences from individual MCT. First, the large group size (10–14 patients per group) may have reduced the acceptability of the treatment modality and contributed to the high drop-out rate. Second, there may have been less time to identify and challenge each patient's idiosyncratic metacognitive beliefs, given the group size. Third, therapist factors may have comprised the effectiveness of the intervention as only two out of four therapists were trained in MCT, and there was no supervision in delivering g-MCT.

In summary, even though van der Heiden et al.'s (2013) results indicated that g-MCT was effective in reducing GAD symptoms, many questions remain regarding the feasibility of g-MCT, such as recruitment, group size, and retention. Consequently, the primary aim of the current study was to benchmark and evaluate the feasibility of g-MCT for adult patients with GAD. Moreover, to explore whether smaller groups would be more feasible and effective, as only 4–6 patients were included in each group. The study was conducted at a Norwegian psychiatric outpatient clinic without a control group. The secondary aim of the study was to evaluate the effectiveness of g-MCT, with the hypothesis being that g-MCT will be associated with significant reductions in symptoms of GAD and depression, as well as reductions in positive- and negative metacognitions, maladaptive coping strategies, and avoidance.

# MATERIALS AND METHODS

# Participants

The sample consisted of 23 participants, of which 22 were women (95.7%). The average age was 29.70 years (SD = 9.21). Further demographic characteristics are shown in **Table 1**. The four patients using antidepressants reported to use either Zoloft or Cipralex. Three of these four had been on a stable dose for years, while the fourth started medication 4 months before treatment. No changes were made to medication during treatment. In addition, two patients used medicine for sleep related problems.

Diagnosis was established using the Anxiety Disorder Interview Schedule (ADIS-IV, Brown et al., 1994). To be included in the present study, GAD had to be the primary diagnosis. None of the participants had known serious somatic illnesses, psychosis, post-traumatic stress disorder, known cluster Aor B personality disorders, were suicidal, or suffered from drug addiction. Seventeen (73.9%) participants had comorbid disorders. Fourteen had one comorbid disorder (OCD = 4, depression = 2, panic disorder = 3, social anxiety disorder = 1, specific phobia = 1, health anxiety = 1, ADHD = 2). Three patients had two comorbid diagnoses (one with panic disorder and depression, one with OCD and depression, and one with OCD and social phobia).

# Procedure

The clinic has a population catchment of approximately 130,000 people. Patients were referred to the clinical service from their GP, student health services, and mental health clinics. The first group started in September 2016 and the last group started in October 2017. Patients included in the study were consecutive referrals.



Patients diagnosed with ADHD were already diagnosed with ADHD as described in their referral.

Pre-treatment assessment consisted of the ADIS-IV (Brown et al., 1994) and completion of self-report questionnaires. The ADIS-IV was conducted by independent investigators (clinical psychologists not involved with the treatment) trained in diagnostic interviewing. Patients received no treatment whilst waiting for treatment to start. The wait time period was 3–4 months.

Five groups were held, each with 4–6 patients. The groups were held at Nidaros DPS, St. Olavs Hospital. Patients were offered 10 weekly group sessions, each with a duration of 90 min. All self-report questionnaires were completed at pretreatment, post-treatment, and at 3-month follow-up. The first groups completed questionnaires on pen and paper at the clinic, while the more recent groups completed questionnaires online. In addition, the Generalized Anxiety Disorder Scale-Revised (GADS-R; Wells, 2009) was distributed before the beginning of each treatment session. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The study was approved by the Regional Committees for Medical and Health Research Ethics in Norway (REK; 2013/2155, Helse Midt, https://helseforskning.etikkom.no/) and conducted without external funding.

# Therapists

All groups were led by two therapists; a psychiatric nurse and a clinical psychologist. Both had completed training in MCT and were registered level 1 and level 2 therapists respectfully. Video supervision was conducted with a master clinician in MCT. Furthermore, several groups had been conducted for training purposes before the open trial was initiated.

# Treatment

The g-MCT had a specific structure and followed the treatment manual for GAD (Wells, 2009). Sessions one and two focused on creating a group case formulation. Participants were helped to create their own personal case formulation. Participants were socialized to the metacognitive model and introduced to the concept of detached mindfulness (detached mindfulness; Wells, 2009). Sessions three and four focused on challenging metacognitive beliefs regarding uncontrollability of worry and the belief that they would lose control if they worried too much. In order to clarify conflicting and dysfunctional metacognitions, the group was divided into two smaller groups and they constructed arguments for worry being controllable or not, and if they could lose control or not. The participants then discussed and challenged each other's beliefs, with help from the therapists.

In sessions five and six the primary aim of MCT was to reduce negative beliefs about the dangers of worry. Both verbal and behavioral strategies were used to challenge metacognitions. Examples of verbal strategies were questioning the evidence of metacognitive beliefs and searching for counterclaims (as with beliefs about uncontrollability in earlier sessions). Thereafter, in session 7 and 8, positive beliefs about worry were challenged and modified.

The last phase of therapy (session 9 and 10) focused on relapse prevention. The group members made a summary of their case formulation (therapy blueprint) and a summary ("old and new plan") of how they used to respond to negative thoughts in the past and contrasted this with their new adaptive responses to worrying thoughts.

# Measures

The Penn State Worry Questionnaire (PSWQ; Meyer et al., 1990) is a 16-item self-report questionnaire measuring the severity of worry, both in terms of frequency, intensity and uncontrollability. Each item is rated from 1 ("not at all typical of me") to 5 ("very typical of me"). The total score ranges from 16 to 80, where a higher score indicates higher levels of pathological worry. It has excellent internal consistency (Cronbach α = 0.93) and good psychometric properties (Meyer et al., 1990). Cronbach's alpha in the current study was 0.97.

Generalized Anxiety Disorder-7 (GAD-7; Spitzer et al., 2006) is a self-report questionnaire with seven items assessing symptoms of GAD. Patients answer how much during the last two weeks they have been bothered by each symptom. The answer options range from 0 ("not at all") to 3 ("almost every day"), resulting in a total score between 0 and 21. A clinical cut-off point of 10 has been suggested. GAD-7 has been shown to have excellent internal consistency (Cronbach α = 0.92) and good test-retest reliability (r = 0.83). It has also demonstrated good criterion, construct, factorial, and procedural validity (Spitzer et al., 2006). Cronbach's alpha in the current study was 0.89.

The Patient Health Questionnaire-9 (PHQ-9; Kroenke et al., 2001) is a self-report questionnaire designed to measure symptoms of depression using nine items corresponding to the nine criteria for depression. The patient answers how troublesome each problem has been during the past two weeks, where each question is scored on a scale of 0 ("not at all") to 3 ("almost every day"). The total score range from 0 to 27, of where a cut point of 10 identifies major depression with good sensitivity and specificity (Kroenke et al., 2001). The

PHQ-9 has demonstrated excellent internal reliability (Cronbach α = 0.86) and test-retest reliability, as well as good construct and convergent validity (Kroenke et al., 2001). Cronbach's alpha in the current study was 0.90.

Generalized Anxiety Disorder Scale-Revised (Wells, 2009) is a self-report inventory based on the metacognitive model of GAD. The first items cover GAD symptoms, time spent worrying, as well as how often a range of coping and avoidance behavior have been done the last week. These items are scored on a scale from 0 to 8. In addition, the GADS-R assesses negative and positive metacognitive beliefs related to worry (Wells, 2009), each measured on a scale from 0 ("I do not believe this at all") to 100 ("I'm completely convinced this is true"). Cronbach's alpha for the coping items was 0.94, 0.79 for avoidance items, and 0.94 for the metacognitive belief items (0.94 for negative beliefs and 0.93 for positive).

# Data Analysis

The feasibility of g-MCT was operationalized and visualized through the participant flow chart (**Figure 1**), of where recruitment and retention rates are important feasibility outcomes. The results are contrasted with the g-MCT study of van der Heiden et al. (2013).

A repeated measures ANOVA was used to investigate changes in worry and symptoms of anxiety and depression. The same test was used to measure changes in metacognitions, coping strategies, and avoidance. There was no significant skewness or kurtosis on pre-treatment measures. Mauchly's test of sphericity was not significant for all analyses using repeated measures ANOVA, except for PHQ-9, negative beliefs, and positive beliefs.

Effect sizes (Cohen, 1992) were calculated with Morris and Deshon's equation no. 8, which controls the correlation between pre- and post-treatment values of the dependent variable. Following Jacobson and Truax (1991) and Fisher (2006), recovery (clinically significant change on the PSWQ) was calculated with the following criteria: cut-off = 47, reliable change index = 7. The study uses a cut-off point and a reliable change index that has been applied to a large group of GAD patients and use the standardized criteria as described in Fisher (2006). These criteria have been used in all other MCT studies for GAD except for the van der Heiden et al. (2013) study. Using the standardized criteria allows benchmarking of the results and allows a reasonable comparison between individual and group MCT. Along with effect sizes, recovery rates were used to compare the treatment effectiveness of the current study with previous studies of both individual and group based MCT for GAD.

Two patients did not complete questionnaires at followup. These values were replaced using last observation carried forward (one classified as improved and one as a treatment nonresponder). There were no other missing values at pre-treatment, post-treatment, or follow-up. Missing values for session-tosession data were not replaced.

Lastly, the potential influence of comorbid disorders on treatment outcome was investigated using independent t-tests. The PSWQ, GAD-7, and PHQ-9 scores of patients with and without comorbid disorders were compared at pre-treatment, post-treatment, and 3-month follow-up.

# RESULTS

# Feasibility

As shown in the participant flow chart (**Figure 1**), 45 patients were referred to and assessed for inclusion in the current study. Twenty-three patients were entered into the study and 22 patients were excluded. The most common reason for exclusion was that GAD was not the primary diagnosis (n = 9). Furthermore, two patients were excluded due to serious somatic disorder, and another two patients were given inpatient treatment instead of outpatient treatment because of their symptom severity and low level of functioning. Six patients preferred individual treatment instead of group treatment, and three patients could not participate in g-MCT due to practical difficulties. Therefore approximately 75% of suitable patients were included in the study. More specifically, 28.1% i.e., 9 of the 32 offered g-MCT declined.

Patients attended a mean of 8.9 (SD = 1.3) sessions. More specifically: one patient attended five sessions (due to scheduling conflicts), two received seven sessions, four received eight sessions, seven received nine sessions, and nine patients attended all ten sessions. Number of sessions were not significantly correlated with symptoms at post-treatment (r = 0.32 and p = 0.13) or follow-up (r = 0.35 and p = 0.10). Patients were asked to give their feedback on treatment acceptability in the tenth and final treatment session. For each group, all patients reported that they would have preferred group treatment rather than individual treatment because they were able to meet other patients which enabled them to learn from each other, and that the group setting reduced stigma related problems.

After completion of the open trial, the two therapists reported that delivering treatment in a group format was clinically appropriate and that the small group format need not prevent any patients from fully participating in the therapy. Furthermore, the clinicians plan to continue to use g-MCT in their routine clinical practice as it is cost-effective and reduces the length of time patients have to wait for treatment.

No patients dropped out during treatment, but two patients did not complete the self-report questionnaires at 3-month follow-up.

# Treatment Effect

**Table 2** shows the mean and standard deviations for preand post-treatment scores and 3-month follow-up. A repeated measures ANOVA was conducted to investigate changes. Mauchley's test was not significant on any of the analyses (except for PHQ-9, and negative- and positive metacognitions), and Wilks' lambda was therefore used. The results show significant improvements and large effect sizes for all measures. Linear mixed model analysis was also attempted with these data. However, all slopes went in the same direction as the results were unambiguous. Furthermore, there were no significant fixed effects only a clear effect of time. Model fit did not significantly improve when including attendance rate and age into the model compared to a simple model.

TABLE 2 | Repeated measures ANOVA testing change in symptoms and metacognitions.


Greenhouse–Geisser correction used for PHQ-9, and negative- and positive beliefs. Effect sizes (Cohen's, 1992) were calculated using Morris and Deshon's equation no. 8 controlling for correlation between pre- and post-treatment value for the variable in question. PSWQ, Penn State Worry Questionnaire; GAD-7, Generalized Anxiety Disorder-7; PHQ-9, Patient Health Questionnaire-9; GADS-R, Generalized Anxiety Disorder Scale-Revised.

Changes in symptoms were significant from pre-treatment to post-treatment, and there were non-significant changes from post-treatment to follow-up for all three measures. In addition to tests of statistical significance, clinically significant change was investigated. Only one patient did not respond to treatment. A summary of recovery rates are displayed in **Table 3**.

Patients with comorbid disorders did not have significantly more symptoms than patients with no comorbidity at any of the three times of assessment. For PSWQ there was no significant difference at pre-treatment, t(21) = 0.96, p = 0.35, at posttreatment, t(21) = 1.82, p = 0.08, or follow-up, t(21) = 1.27, p = 0.22. Five of the six (83.3%) patients without comorbid disorders were recovered at follow-up compared to 76.5% for patients with comorbid disorders. For GAD-7 there was also no difference at pre-treatment, t(21) = 0.36, p = 0.73, at posttreatment, t(21) = 0.55, p = 0.73, or follow-up, t(21) = 0.71, p = 0.49. Same observation was made for PHQ-9 at pretreatment, t(21) = 0.61, p = 0.55, at post-treatment, t(21) = 1.34, p = 0.19, and at follow-up, t(21) = 0.32, p = 0.76.

# Metacognitive Changes From Session to Session

Generalized Anxiety Disorder Scale-Revised was completed by patients before every session to measure changes in symptoms, worry, metacognitions, coping strategies, and avoidance.

TABLE 3 | Recovery rates (percentages) at post-treatment and follow-up.


PSWQ, Penn State Worry Questionnaire; GAD-7, Generalized Anxiety Disorder-7; PHQ-9, Patient Health Questionnaire-9. Cut-off values for GAD-7 and PHQ-9 was set at >10. Improved = at least 7-points improvement on PSWQ or below cut-off. Recovered = criterion for improved and scoring 47 or less on PSWQ. 91.3% of participants scored above cut-off on GAD-7 at pre-treatment, and 73.9% scored above cut-off on PHQ-9. The two patients that scored below cut-off on GAD-7 at pre-treatment were not classified as recovered (probably due to low pre-treatment values).

**Table 4** shows a general decrease in all MCT related factors from session 1 to session 10. In general, the graph shows that treatment was associated with reductions in symptoms, worry, negative- and positive metacognitions, maladaptive coping strategies, and avoidance.

# Comparison With Other GAD Trials

For benchmarking purposes, uncontrolled effect sizes (all outcome measures using the PSWQ) were compared to the previously mentioned studies of MCT for GAD (Wells et al., 2010; van der Heiden et al., 2012, 2013; Nordahl et al., 2018). **Figure 2** shows effect sizes (using pooled standard deviations) from pre-treatment to post-treatment and from pretreatment to follow-up for the various studies. The results suggested that patients in the current study had obtained large reductions in symptoms of worry that were comparable even with individual MCT for GAD. Patients in the current study had quite high scores on PSWQ at pre-treatment, whereas posttreatment and follow-up scores were comparable with results from individual MCT. T-tests comparing the results of the current study with that of Wells et al. (2010) showed that the current study had a significantly higher PSWQ pre-treatment score, t(31) = 2.86, p = 0.007, while there was no significant difference at post-treatment, t(31) = 0.14, p = 0.889 and followup, t(31) = 0.55, p = 0.587.

The average number of therapist hours per patient in this study was 6.5 h (10 session × 1.5 h × 2 therapists<sup>∗</sup> 5 groups/23 patients = 6.5), which accounts for fewer hours per patient compared to Wells et al. (2010) and van der Heiden et al. (2012) which had 10–12 sessions (45–60 min each) per patient.

# DISCUSSION

The aims of the current study were to evaluate the feasibility and effectiveness of g-MCT for patients with GAD within the context of an ordinary psychiatric clinic. As only a small proportion of patients declined g-MCT in favor of individual MCT and no patients dropped out during treatment, g-MCT appeared to be an acceptable treatment modality. Furthermore, g-MCT was associated with significant reductions in worry and symptoms of anxiety and depression. There were also significant reductions in all MCT related factors such as positive metacognitive beliefs, negative metacognitive beliefs, and maladaptive coping strategies (including avoidance behavior). Session to session ratings indicated that the reduction in symptoms, metacognition, and coping behavior coincided with each other. However, due to the design of the study, the results provide no clarity with respect to causal relationships. In sum, large effect sizes and high recovery rates indicate that g-MCT is an effective treatment for GAD.

With respect to treatment feasibility, 23 patients received treatment, while 22 patients were excluded. GAD not being the primary diagnosis (n = 9) was the most common reason for exclusion. Six patients (19 %) declined g-MCT in favor of individual MCT, and three patients (9%) were unable to attend due to scheduling conflicts. Thus, 28% of participants who were offered treatment chose not to participate. This rate is slightly higher compared to a previous RCT study [19.8% (20 of 101 eligible patients)] offering individual treatment (Nordahl et al., 2018). Group treatment could also be less flexible than individual treatment which could exclude patients with set or busy schedules. On the other hand, a positive aspect is that none of the included patients dropped out during treatment, suggesting that g-MCT was accepted by the participants. Furthermore, the average number of therapist hours per patient in this study was 6.5 h, which accounts for fewer hours per patient compared to studies using individual therapy (typically 10–12 sessions). Thus, g-MCT appear to be a cost-effective treatment method.

According to benchmarking analyses, patients in the current study had quite high scores on PSWQ at pre-treatment, while post-treatment and follow-up scores were comparable to previous investigations of individual MCT for GAD (Wells et al., 2010; van der Heiden et al., 2012; Nordahl et al., 2018). The recovery rate (PSWQ) at post-treatment in this study was 65.3%, which is somewhat lower than Wells et al. (2010). This might be explained by the high pre-treatment scores in the current study. However, the recovery rate increased to 78.3% at 3-month follow-up, which is in line with results from individual MCT. The group study of van der Heiden et al. (2013) showed somewhat lower recovery rates than the current


TABLE 4 | Changes on GADS-R from session to session.

fpsyg-10-00290 February 12, 2019 Time: 17:50 # 8

Changes from session to session (pre-treatment to 3-month follow-up) in GAD symptoms, worry, negative− and positive metacognitions, maladaptive coping strategies, and avoidance. All scores are transformed to a 0–8 scale.

study. It could be speculated that this is related to differences in group size (4–6 patients vs. 10–14 patients per group), but it could also be related to therapist factors, as two of their four therapists had not received MCT training. When comparing uncontrolled within effect sizes for studies on MCT for GAD, the current study showed promising results. However, the effect size estimation could be inflated and influenced by the relatively small sample size. The results are also encouraging when compared to recovery rates in CBT. As previously mentioned 50–60% are recovered following CBT for GAD (Fisher and Durham, 1999), and only 38% were recovered in a recent study (Nordahl et al., 2018).

Group-MCT was associated with significant reductions in positive and negative metacognitions. The reduction was greater for the negative metacognitive beliefs than for positive beliefs. A possible explanation could be that patients reported fewer positive than negative metacognitive beliefs at the start of treatment.

Treatment was also associated with reduction in symptoms of depression and comorbidity did not affect treatment outcome. This is an appealing aspect of treatment given the high rate of comorbidity (and overlap in symptoms) between GAD and depression. This finding is also consistent with studies showing that MCT has an effect on comorbid disorders (e.g., Johnson et al., 2017; Capobianco et al., 2018; Papageorgiou et al., 2018). The fact that treatment reduced comorbid symptoms of depression is also consistent with a metacognitive understanding of common underlying psychological processes

in emotional disorders, and therefore supports a transdiagnostic utility of MCT.

The study is not without limitations. The most obvious is the open trial design lacking a control group. Therefore, the study is unable to control for random fluctuations, spontaneous recovery, or effect of external variables. Evaluation of treatment effectiveness was also based on self-reported symptoms, which poses certain limitations such as social desirability. However, this effect could also be present for interview based ratings. Diagnostic re-assessment at long term follow-up is ongoing. Another issue is that it was a predominantly a female sample, as well as a probable overrepresentation of patients with comorbid OCD. A strength of the study is however that treatment outcomes were comparable for patients with and without comorbid disorders. Furthermore, there was no official measure of adherence. However, video supervision was conducted with an international expert in MCT and several groups had been conducted for training purposes before the open trial was initiated. Another issue is that diagnostic interviews were not videotaped and there is no measure of interrater agreement. Sample size is also an issue for the comorbidity analyses and comparing results across treatment studies is not always straightforward as samples and conditions may vary.

# CONCLUSION

In conclusion, the results of this study show that g-MCT was a suitable and effective treatment for patients with GAD.

# REFERENCES


Treatment was associated with significant reductions in worry, anxiety, dysfunctional metacognitions, and coping strategies. It was also associated with significant improvement in symptoms of depression, which supports the transdiagnostic effects of MCT. Effect sizes were high and recovery rates were comparable to previous studies. The study supports further evaluation of group-MCT for patients with GAD using larger sample sizes and controlled designs.

# AUTHOR CONTRIBUTIONS

SH, SS, and PF were responsible for designing the study. SH and GS conducted the therapy. PF supervised the therapists. EB and SS wrote the first draft of the manuscript and conducted statistical analyses. EB and TG were responsible for diagnostic interviews. SS acted as principle investigator and was responsible for getting ethical approval. All authors have contributed in revising the manuscript and approved its submission.

# ACKNOWLEDGMENTS

We would like to thank all patients participating in the study. We would also like to thank Tonje Grønning Andersen for help with the LMM analysis.



Wells, A., Welford, M., King, P., Papageorgiou, C., Wisely, J., and Mendel, E. (2010). A pilot randomized trial of metacognitive therapy vs applied relaxation in the treatment of adults with generalized anxiety disorder. Behav. Res. Ther. 48, 429–434. doi: 10.1016/j.brat.2009.11.013

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer BF declared a past co-authorship with one of the authors PF to the handling Editor.

Copyright © 2019 Haseth, Solem, Sørø, Bjørnstad, Grøtte and Fisher. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A Randomized Controlled Trial of Metacognitive Therapy for Depression: Analysis of 1-Year Follow-Up

Odin Hjemdal<sup>1</sup> \*, Stian Solem<sup>1</sup> , Roger Hagen<sup>1</sup> , Leif Edward Ottesen Kennair<sup>1</sup> , Hans M. Nordahl2,3 and Adrian Wells4,5

<sup>1</sup> Department of Psychology, Norwegian University of Science and Technology, Trondheim, Norway, <sup>2</sup> Department of Mental Health, Norwegian University of Science and Technology, Trondheim, Norway, <sup>3</sup> Nidaros DPS, Division of Psychiatry, St. Olavs Hospital, Trondheim, Norway, <sup>4</sup> Division of Clinical and Health Psychology, The University of Manchester, Manchester, United Kingdom, <sup>5</sup> Greater Manchester Mental Health NHS Foundation Trust, Prestwich, United Kingdom

#### Edited by:

Roberto Cattivelli, Italian Auxological Institute (IRCCS), Italy

#### Reviewed by:

Giancarlo Dimaggio, Centro di Terapia Metacognitiva Interpersonale (CTMI), Italy Sverre Urnes Johnson, Modum Bad Psychiatric Center, Norway

> \*Correspondence: Odin Hjemdal odin.hjemdal@ntnu.no

#### Specialty section:

This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology

> Received: 16 April 2019 Accepted: 25 July 2019 Published: 08 August 2019

#### Citation:

Hjemdal O, Solem S, Hagen R, Kennair LEO, Nordahl HM and Wells A (2019) A Randomized Controlled Trial of Metacognitive Therapy for Depression: Analysis of 1-Year Follow-Up. Front. Psychol. 10:1842. doi: 10.3389/fpsyg.2019.01842 This paper reports the 1-year follow-up results from a randomized controlled trial (RCT), which examined the efficacy of metacognitive therapy (MCT) for unipolar depression compared to a waiting condition. Thirty-nine patients with major depression were offered MCT and were divided into two conditions; immediate MCT with 10 weekly sessions or a waiting period that had a 10-week delayed MCT start. Two participants dropped out during the waiting condition. Thirty-four patients participated in the follow-up assessment. Based on the intent-to-treat sample and all patients, 67% were classified as recovered, 13% improved, and 20% were unchanged at 1-year follow-up. For the completers sample 73% recovered, 12% improved, and 15% were unchanged. Five of the 31 patients (13%) that were in remission at post-treatment experienced relapse at 1-year follow-up. Within-group effect sizes were large for reductions in symptoms of depression (d = 2.09) and anxiety (d = 1.16) at 1-year. Treatment response was associated with reductions in rumination, worry, and metacognitive beliefs as predicted by the metacognitive model, but reductions in metacognitions independently predicted reductions in depression scores from pre-treatment to 1-year follow-up. The results suggest that treatment gains are stable at 1-year follow-up. The study sets the stage for future research, which should evaluate MCT over a longer term and compare it with active treatments using suitably powered RCTs.

#### Keywords: depression, metacognitive therapy, 1-year follow-up, rumination, worry

# INTRODUCTION

Depression is one of the most common psychiatric disorders, with a high degree of comorbidity (Kessler et al., 2003), and is the leading cause of disease burden worldwide (World Health Organization [WHO], 2018). The consequences are significant in terms of lost work productivity, mortality, and lower quality of life (Simon, 2003). The risks associated with depression are profound with the majority of suicides committed by depressed individuals (Hawton et al., 2013).

Perhaps the most challenging aspect of depression with respect to treatment is its recurrent nature. As many as 85% of those that recover from major depressive disorder will have a second episode within 15 years of naturalistic follow-up, and additional episodes will increase the relapse

probability by 18% (Mueller et al., 1999). Despite being recognized as a commonly occurring disorder, many patients do not receive the best recommended treatments (Kessler et al., 2008). Furthermore, for those that receive an active treatment, a major problem for depressed patients is the high relapse rate at follow-up (Steinert et al., 2014).

Cognitive-behavioral therapy (CBT) is a recommended treatment for adult unipolar depression (Butler et al., 2006). However, findings suggest that relapse rates are from 29 to 39% within 1 year, and between 40 and 60% within a period of 2 years (Hollon et al., 2006; Vittengl et al., 2007; Dobson et al., 2009). For behavioral activation the 1-year relapse rates were reported as 50%, with continued medication being 53% and medication withdrawal 59% (Dobson et al., 2009). Antidepressant medication has a similar efficacy to CBT in treating depression but relapse rates are between 29 and 60% within one to 2 years (Parker et al., 2008). There is a clear need to develop more effective treatments for depression and to reduce relapse rates after treatment.

Metacognitive therapy (MCT; Wells, 2009) is a treatment that may offer an advance, because it targets specific processes thought to increase risk of depression. It is based on the Self-Regulatory Executive Function model (S-REF; Wells and Matthews, 1994, 1996; Wells, 2000), which proposes that low mood and depression is prolonged by perseverative thinking styles, such as depressive rumination, worry, and other unhelpful self-regulation strategies. This thinking style, called the cognitive attentional syndrome (CAS; Wells and Matthews, 1994) is influenced by positive and negative metacognitive beliefs about uncontrollability and danger of rumination and worry as well as maladaptive executive control of attentional processes.

Empirical studies, such as those of Papageorgiou and Wells (2003) and Solem et al. (2016) have confirmed theoretically consistent relationships between positive metacognitive beliefs, rumination, negative metacognitive beliefs and depression consistent with the model. The model predicts that recovery from depression requires reductions in rumination, worry, and dysfunctional metacognitions, as well as changes in metacognitive beliefs (Wells, 2009). Clarifying the mechanisms of change in MCT may help expand and elaborate understanding of depression and refine the delivery of treatment (Hoffart et al., 2018). Studies should therefore explore if the effects of MCT for depression are related to changes in the hypothesized causal variables.

A meta-analysis of the effects of MCT for anxiety and depression showed that the treatment is effective (Hedges' g = 2.06 compared to wait-list) and potentially more effective than recommended treatments such as CBT at post-treatment (Hedges' g = 0.69 − 0.37) (Normann and Morina, 2018). Across studies of depression, most of which have been small-scale to date, recovery rates for MCT typically range from 66–79% at post-treatment (Wells et al., 2012; Papageorgiou and Wells, 2015; Hjemdal et al., 2017).

A platform trial of treatment-resistant depression with 12 patients found that 66.6% of patients treated with MCT were recovered at post-treatment and 58.3% at follow-up (Wells et al., 2012) using the stringent criteria of Frank et al. (1991). Similarly, a case series by Callesen et al. (2014) reported that three out of four depressed patients were recovered, and a group MCT study by Dammen et al. (2015) found that 91% of 11 patients recovered at 6-months follow-up. At 1-year follow-up, 70% remained recovered, and 80% at 2-year follow-up (Dammen et al., 2016). In 2015, Papageorgiou and Wells also published a trial for group MCT for antidepressant and CBT resistant depression using a baseline-controlled design. The study included 10 patients and showed that 70% were recovered at post-treatment and 6-months follow-up (Papageorgiou and Wells, 2015). An opentrial with 10 comorbid depressed patients also reported 70% recovery rates at 6-month follow-up (Hjemdal et al., 2017). Whilst promising, these trials are small scale and the data must be considered preliminary.

The first RCT included 39 patients with major depression and compared individual MCT with waitlist (Hagen et al., 2017). Results indicated that 79.5% were recovered at post-treatment and 69.2% at 6-months follow-up. Whilst the recovery rates of MCT are very promising, longer follow-up data from randomized trials is required to assess the effects of MCT for depression.

In the present study, we conducted a follow-up analysis of the Hagen et al. (2017) patients 1-year after finishing treatment. Further, we examined the levels of anxiety, rumination, worry, and dysfunctional metacognitions at 1-year follow-up. The study tested whether gains made in these constructs were different at 1-year follow-up in recovered and non-recovered patients.

# MATERIALS AND METHODS

# Participants

The total sample consisted of 39 participants of which 59% were women (n = 23). The mean age was 33.7 years (SD = 10.42) ranging from 18 to 54. Three participants were of Asian ethnicity while the remaining were ethnic Norwegian. A total of 41% were single, 38% were married/cohabitants, 13% had partners, and 8% were divorced/separated. With respect to employment, 31% worked full time, 21% had part-time jobs, 21% were students, while 33% received social/welfare benefits. The group had on average 1.08 (SD = 1.28) children. Patients who were treated with SSRI were included if they were on a stable dosage and agreed to maintain this dosage throughout the study. However, only three used SSRIs. The majority of participants had been treated previously for their depression (76.9%). With respect to their highest obtained education, 5% had completed elementary school, 44% had completed high school, 13% finished college, and 38% had a master's degree.

Mean age of onset for the first depressive episode was 26.2 years (SD = 11.7) and patients had suffered from depression on average for 7.6 years (SD = 7.1). In the study 84.6% (33 patients) were diagnosed with recurrent depression (one mild, 21 moderate, 11 severe), and 15.4% (six patients) with single depressive episode (three moderate, three severe). Comorbidity was common as only 33% had depression as their single diagnosis. Different additional axis-I disorders were present in 41% of the sample (10 with generalized anxiety disorder, two with panic disorder, and single incidents of social phobia,

hypochondriasis, trichotillomania, and eating disorder not otherwise specified). With respect to axis-II disorders, 33% were diagnosed with such (three with avoidant personality disorder and 10 with obsessive compulsive personality disorder). Only three reported having received psychological treatment between post-treatment and 1-year follow-up, 28 reported no additional treatment and eight had missing data on this issue.

# Procedure

The RCT was registered at ClinicalTrials.gov (NCT01608399). The Regional Medical Ethics Committee in Norway (REK-Midt ref. no. 2011/1138) provided their ethical approval. The main inclusion criterion was a DSM-IV diagnosis of primary unipolar depression (including mild, moderate, and major). Participants with a single episode of depression or recurrent depression were included. Further inclusion criteria incorporated signing the written informed consent form and being 18 years or older, accepting random allocation, and not receive multiple therapies at the same time. Patients were excluded if they suffered from a known somatic disease, were psychotic, suicidal, had PTSD, cluster A or cluster B personality disorder, substance dependence, and they had to accept random allocation, and not receive multiple therapies at the same time. In all, 105 participants attended a diagnostic interview of which 63% (n = 66) of those were excluded. Reasons for exclusions among were: other primary diagnosis (30%), GAD as the prominent diagnosis (27%), cluster A or B personality disorder (15%), no psychiatric diagnosis (12%), subclinical depression (8%), somatic disease (3%), PTSD (2%), substance dependence (2%), and psychosis (2%).

Participants were recruited between 2013 and 2015. They were treatment-seeking individuals referred by their GP or self-referral. Adverts describing the study were placed in newspapers, in letters to GPs, and on social media. Referred patients were given a telephone screening to ensure that they had symptoms resembling depression. Those that did were offered an appointment to meet with a trained assessor for a diagnostic interview. Further information about the study was given and they were given the informed consent to sign. The assessment covered inclusion and exclusion criteria. The diagnostic interviews used the Structured Clinical Interview for the DSM-IV axis-I (SCID-I; First et al., 1995), as well as the Structured Clinical Interview for the DSM-IV axis-II (SCID-II; Gibbon et al., 1997). The assessment team interviewed patients at pre- and post-treatment. Those accepted into the trial were randomly assigned to begin 10 sessions of MCT treatment either immediately or after a 10-week wait period. Follow-up assessment (1-year) was accomplished by mailing paper versions of the questionnaires to participants, who filled them out in their own homes. Results from the two treatment conditions (immediate and delayed) were included in the analyses of the follow-up data. **Figure 1** displays a flow chart of the study.

# Instruments

The Beck Depression Inventory (BDI; Beck et al., 1961) assesses severity of depressive symptoms. The BDI has 21-items that are rated on a 0–3 scale. The reported Cronbach's alpha of BDI is between 0.86 and 0.89 (Beck et al., 1961, 1988). The BDI is a reliable and valid measure of depressive symptoms (Beck et al., 1988). BDI total scores can be classified accordingly: 0–9 minimal depressive symptoms, 10–18 mild depressive symptoms, 19–29 moderate depressive symptoms, and 30–63 severe depressive symptoms.

The Ruminative Response Scale (RRS; Nolen-Hoeksema and Morrow, 1991) assesses rumination in response to depressed mood (e.g., think "Why do I have problems other people don't have?" or "think about how sad you feel"). The RRS has 22 items that are rated on a 1 to 4 scale, and scores range from 22 to 88. Higher scores indicate higher levels of rumination. Psychometric properties with Cronbach's alphas have been reported between 0.88 and 0.92 (Luminet, 2004).

The Positive Beliefs about Rumination Scale (PBRS; Papageorgiou and Wells, 2001b) assesses beliefs about the benefits of rumination (e.g., "Ruminating about my feelings helps me to recognize the triggers for my depression" and "I need to ruminate about the bad things that have happened in the past to make sense of them"). The PBRS has nine items using a 1–4 scale, and scores range from 9 to 36. Good psychometric properties have been documented with Cronbach's alpha of 0.89 (Luminet, 2004).

The Negative Beliefs about Rumination Scale (NBRS; Papageorgiou and Wells, 2001a) assesses beliefs about uncontrollability and harm as well as interpersonal consequences (e.g., "rumination can make me physically ill," "I can't stop myself from ruminating," "only weak people ruminate"). The NBRS has 13 items using a 1 to 4 response scale, and scores range from 12 to 52. Good psychometric properties have been documented with Cronbach's alphas between 0.80 and 0.83 (Luminet, 2004).

Metacognitions Questionnaire-30 (MCQ-30; Wells and Cartwright-Hatton, 2004) assesses levels of metacognitive beliefs. The MCQ-30 has 30 items which are rated 1–4, with higher scores indicating higher levels of maladaptive metacognitions. Scores range from 30 to 120. The psychometric properties are good with Cronbach's alpha for the total score of 0.88 (Spada et al., 2008).

Penn State Worry Questionnaire (PSWQ; Meyer et al., 1990) assesses levels of worry. The PSWQ has 16 items which are rated on a 1–5 scale, with higher scores indicating higher levels of worry. Scores range from 16 to 80. The psychometric properties are good with Cronbach's alpha of 0.93 (Brown et al., 1992).

The Beck Anxiety Inventory (BDI; Beck and Steer, 1990) assesses severity of anxiety symptoms. The BAI has 21 items which are rated on a 0 to 3 scale. BAI total scores can be classified accordingly: 0–7 minimal anxiety, 8–15 mild anxiety, 16–25 moderate anxiety, and 26–63 severe anxiety. The BAI has good psychometric properties with Cronbach's alpha of 0.92 (Steer et al., 1993).

# Therapists

Four therapists all of whom were clinical psychologists and were trained in MCT delivered therapy. Treatment was supervised by the last author (AW) and the supervision was based on videotaped recordings of the sessions. In addition, the therapists met every second week for peer supervision.

# Treatment

fpsyg-10-01842 August 7, 2019 Time: 18:7 # 5

Treatment consisted of 10 sessions and followed the manual of MCT for depression (Wells, 2009). The main components of the treatment involve in the following sequence; (1) case conceptualization and (2) socialization to the MCT model for depression, (3) learning triggers for rumination, (4) attention training, (5) challenging beliefs about uncontrollability of rumination, (6) challenging other negative metacognitive beliefs, (7) challenging positive metacognitive beliefs, (8) eliminating coping strategies, and (9) relapse prevention.

# Statistics

Shapiro-Wilk test of normality was not significant for the study outcome variable. Effect sizes were calculated with Cohen's d. To evaluate clinically significant outcomes, the corrected Jacobson criterion (Jacobson et al., 1999) reported in Christensen and Mendoza (1986) was used. This meant a cut-off of 14 points and below on the BDI and based on the current sample an estimated reliable change index of 9.49, which was rounded down to 9.

Bivariate Pearson's correlations were run in order to examine the association between change score from pre-treatment to 1 year follow-up 1BDI score, and pre-treatment as well as change scores from pre-treatment to 1-year follow-up on: 1BAI, 1RRS, 1NBRS, 1PBRS, 1PSWQ, and 1MCQ-30.

A multiple hierarchical regression analysis explored if changes from pre-treatment to 1-year follow-up in rumination, worry or metacognition predicted change in depressive symptoms from pre-treatment to 1-year follow-up, thus the outcome variable was change from pre-treatment to 1-year follow-up 1BDI score. In the first step gender and age was entered, in the second step using the forward selection method change scores from pre-treatment to 1-year follow-up were entered for 1RRS, 1PSWQ and 1MCQ-30. Note that pre-treatment scores for the waiting list patients were assessed post-waiting list before starting treatment.

# Missing Values and Imputation of Data

Missing data in the intention to treat (ITT) analyses were replaced using last observation carried forward. Two participants allocated to waiting list (delayed treatment) dropped out during the waiting period (one moved and one started treatment at a private practice psychologist) and did not provide data after pre-treatment. A further two from the waitlist condition did not complete all 10 treatment sessions. In the MCT immediate treatment group all participants completed treatment. All except one of the remaining participants completed self-report questionnaires at 6 month and 1-year follow-up. There was very little missing data on individual BDI items (0.4%) and BAI items (0.8%).

# RESULTS

The results displayed in **Table 1** show a significant change in BDI, BAI, MCQ-30, NBRS, PBRS, RRS, PSWQ, and MCQ-30 from pre- to post-treatment, 6-month and 1-year follow-up. The uncontrolled effect sizes varied between 2.53 and 1.16. The highest effect sizes were for levels of rumination and depressive symptoms. **Table 1** displays mean and standard deviations for all of the outcomes. At 1-year follow-up there was a small but statistically significant increase in BDI symptoms, but the mean score remained low at 8.85 and the effect size was 2.09.

# Clinically Significant Change Analyses

On the BDI, based on Jacobson criteria (Jacobson et al., 1999) at 1-year follow-up the response rates are presented in **Table 2**. Statistics for ITT and completer samples are presented for the entire combined samples (immediate MCT plus delayed MCT) and for the immediate MCT subgroup seperately. The proportion of recovered patients is higher in the completers dataset compared to the ITT data-set as might be expected. We will concentrate on the ITT data here as it is more conservative since we would expect depressed patients to recover over time and therefore using LOCF is likely to reduce the time effect. Seventy per-cent (70%) of the immediate MCT patients were recovered

TABLE 1 | Means, standard deviations at pre-treatment, post-treatment, and 6-month and 1-year follow-up with mixed modeling for the BDI, BAI, MCQ-30, NBRS, PBRS, RRS, PSWQ (N = 39).


BDI, Beck Depression Inventory; BAI, Beck Anxiety Inventory; MCQ-30, Metacognitions Questionnaire-30; PSWQ, Penn State Worry Questionnaire; NBRS, Negative Beliefs about Rumination Scale; PBRS, Positive Beliefs about Rumination Scale; RRS, Ruminative Response Scale. The reported data are based on intention-to-treat. Missing data is replaced using last observation carried forward ∗∗∗ p < 0.001.

at 1 year follow-up (n = 20), whilst this figure was 66.7% in the combined sample (n = 39). A further 15 and 12% of patients were reliably improved and some patients were classified as no change (15 and 20.5%) respectively in the two sub-groupings.

To check if the recovery rates for the entire sample depended on severity of depression, clinically significant change was also calculated separately for the subgroups moderate and severe depression. For the moderate depression subgroup (n = 24), 60% were classified as recovered at 1-year follow-up (72% improved). For the severe depression subgroup (n = 14) 79% were recovered at 1-year follow-up (93% improved).

**Table 3** presents the clinically significant change score from post-treatment to 1-year follow-up. None of the patients that were classified as unchanged at post-treatment had changed their status at 1-year follow-up. Of the improved category one out of five patients changed their classification to recovered. Among the recovered group at post-treatment, 25 patients remained in the recovered category, while one changed to the improved group. Five patients that were recovered at post-treatment changed to unchanged at 1-year follow-up (indicating a relapse rate of 12.8%). The overall picture is that the large majority of recovered patients remained the same both at post-treatment and at 1 year follow-up. The fluctuation is relatively limited, and the data are based on intention-to-treat which is conservative in

TABLE 2 | Clinically significant change in depressive symptoms for the MCT immediate treatment group (n = 20) and the total combined sample (N = 39).


MCT ITT, MCT immediate treatment group with intention to treat; completers, analysis of those completing all data; All, the total sample (immediate MCT plus delayed MCT); ITT, Intention-to-treat; BDI, Beck Depression Inventory; No patients deteriorated; BDI criteria for improvement, patients that had a 9-point reduction on the BDI or BDI of 14 or less; BDI criteria for recovery, patients that had a 9-point reduction on the BDI and a score of 14 points or lower.

TABLE 3 | Change in clinical improvement rates from post-treatment to 1-year follow-up (N = 39).


BDI criteria for improvement, patients that had a 9-point reduction on the BDI. BDI criteria for recovery, patients that had a 9-point reduction on the BDI and a score of 14 points or lower.

this regard. The results suggest a small proportion of patients relapsing following MCT at 1-year follow-up.

To explore any effect of pre-treatment symptom severity on longer term outcomes pre-treatment scores on the BDI, BAI, RRS, PSWQ, NBRS, PBRS, MCQ-30 and BAI were correlated with changes from pre-treatment to 1 year follow up on BDI. None of the pre-treatment scores correlated with the 1BDI.

Next, we explored the possible association between change in these predictors over the longer-term (pre-treatment to 1-year follow-up) and longer-term change in depression (pre-treatment to 1-year follow-up). The results of these analyses are displayed in **Table 4**. It is evident that changes in all variables were positively associated with change in depression. The highest correlation was BAI which probably reflects the overlap of symptoms of anxiety and depression. Of the theoretical predictors (purported causal factors) metacognitive belief change (MCQ30) showed the strongest positive association.

Finally, we ran a hierarchical multiple linear regression to explore the best independent predictor amongst change in the predictive mechanisms (MCQ30, RRS, PSWQ). In the first step we controlled for age and gender and neither were significant, in the second step the forward method resulted in 1MCQ-30 as a significant predictor. Neither 1RRS nor 1PSWQ were significant predictors. The summary statistics are presented in **Table 5**.

# DISCUSSION

The follow-up results from this RCT showed that the effects gained with MCT at post-treatment were largely maintained at 1-year follow-up for depressive and also for anxiety symptoms. The clinically significance analyses showed that 70.0% of the MCT immediately treated intent-to-treat sample, and 73.5% in the completers sample achieved recovery at 1-year follow up for individual MCT. For the total ITT sample this figure was 66.7%. These results appear to be consistent with previous studies with 1-year follow-up of group MCT depression in

TABLE 4 | Bivariate Pearson's correlations between the BDI 1-year follow-up score and the change scores from pre-treatment to 1-year follow-up of BDI, BAI, RRS, PSWQ, NBRS, PBRS, and MCQ.


1, changes from pre-treatment to 1-year follow-up scores; BDI, Beck Depression Inventory; BAI, Beck Anxiety Inventory; MCQ-30, Metacognitions Questionnaire-30; PSWQ, Penn State Worry Questionnaire; NBRS, Negative Beliefs about Rumination Scale; PBRS, Positive Beliefs about Rumination Scale; RRS, Ruminative Response Scale. <sup>∗</sup> p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001.



Predictors are age, gender, change scores from pre-treatment to 1-year follow-up 1RRS, 1PSWQ and 1MCQ-30 using forward variable selection. 1, changes from pre-treatment to 1-year follow-up scores; BDI, Beck Depression Inventory; MCQ-30, Metacognitions Questionnaire-30; PSWQ, Penn State Worry Questionnaire; RRS, Ruminative Response Scale.

an open trial (Dammen et al., 2016) and for individual MCT reported in a previous platform trial for treatment resistant depression (Wells et al., 2012). The current study extends previous findings by including a larger sample size and a randomized controlled design. Results suggest that the majority of patients benefitted from MCT.

One of the major challenges with depression treatment has been the recurring nature of the disorder. Relapse rates for CBT have been reported as 29–39% at one year and up to 60% for antidepressant treatments with a range between 40 to 60% within 2 years (Gloaguen et al., 1998; Hollon et al., 2006; Vittengl et al., 2007; Dobson et al., 2009). In the current study five of the patients who were recovered at post-treatment relapsed which is a rate of 12.8%, and in addition one was classified as improved instead of recovered at follow-up. Of the five patients who were improved at post-treatment, four remained improved at follow-up, while one had improved further and classified as recovered. The beneficial effects of MCT seem to cut across the severity of symptoms with 60% recovered in the moderate depression subgroup and 79% recovered in the severe depression subgroup. Also, there were no pretreatment (t = 0.54, p = n.s.) nor post-treatment (t = 1.09, p = n.s.) differences in level of depressive symptoms in the current sample between patients with and without personality disorder. This suggests that the treatment effects may not be dependent on the presence or absence of at least some co-morbid personality issues.

Some exploratory results from the present paper showed that the pre-treatment values were not associated with the 1-year follow-up values of depression while the change scores from pre-treatment to 1-year follow-up were. This suggests that pre-treatment severity was not associated with depression improvement levels over the 1-year period. The results of bivariate correlations showed that change in patients' metacognitive beliefs, rumination and worry over the 1 year rather than pre-scores on depression symptom severity were associated with changes from pre-treatment to 1-year follow-up depression scores. Consistent with the Metacognitive model (Wells, 2009), reduction in rumination, worry and maladaptive metacognitions appeared to be associated with improvements in depression over the longer term. Among these processes, the regression showed that changes in maladaptive metacognitions was an independent predictor of changes in depression scores from pre-treatment to 1-year follow-up. Similar findings have been found in studies of predictors of outcome in OCD treatment (Solem et al., 2009). Future studies should explore metacognitions measured session by session, which will highlight the possibility to disaggregate both the within and between effects.

Metacognitive therapy could be more effective than other treatments (e.g., Nordahl et al., 2018; Normann and Morina, 2018) and could have good long terms outcomes. There are different explanations as to why MCT might offer an efficient and long-lasting treatment for depression. MCT for depression aims to: (1) increase awareness of metacognitive processes and reduce of rumination, worry and threat monitoring; (2) facilitate control of these processes and greater attentional flexibility, and (3) modify negative and positive metacognitive beliefs (Wells, 2009). The S-REF model (Wells and Matthews, 1994), which is the founding of MCT, hypothesizes that the cognitive attentional syndrome maintains disorder and MCT is designed to directly target this mechanism. In the present study those who recovered had a considerably larger reduction in rumination, negative and positive beliefs about rumination, negative metacognitions, and worry, than patients who did not recover. Furthermore, this mechanism is thought to underlie most forms of psychopathology and so MCT may be particularly effective at dealing with multiple morbidities, thereby reducing parallel problems that may confer risk of relapse (e.g., Nordahl, 2009; Hjemdal et al., 2017). The present findings are consistent with the metacognitive theory, and are in line with other studies showing that metacognitions and rumination are important factors for the level of symptoms of depression (Papageorgiou and Wells, 2003; Wells, 2009; Solem et al., 2016). The research is also in line with research showing that change in metacognition is associated with change in symptoms (Solem et al., 2009).

# Limitations

One of the limitations of the study is the small sample size. In addition, five patients (12.8%) did not attend the follow-up assessment, and two of these dropped out early when they were randomized to the waitlist condition. We used last observation carried forward to deal with these missing data. This method has been criticized, but depression is known to recover over time, and we retained the last scores of patients that dropped out which reduces this effect of time. Another limitation is that follow-up assessment was based on self-reported data. Future studies should include additional diagnostic evaluations and compare MCT to other active treatment.

The limitations reported in Hagen et al. (2017) are also valid for the current study. There was only informal assessment of treatment adherence and therapist competence. Adherence was, monitored through supervision but there was no formal assessment of adherence to the treatment manual. As previously reported, there were no differences between therapists in terms of patient outcomes. This suggests that therapist differences did not

affect the results. Future studies should include active treatment as a comparison condition. However, the course of untreated depression may serve as a benchmark for assessing the true benefits of an active treatment. Posternak and Miller (2001) reported that the decrease in depressive symptomatology can be between 10 and 15% on average without treatment.

The sample included cluster C personality disorders which applied for 33% of the sample, but other personality disorders were not included. The results are therefore limited to cluster C personality disorder and predominantly OCPD and avoidant personality disorders.

# CONCLUSION

Large improvements in depression and anxiety symptoms were observed. Improvement was associated with reductions in rumination, worry and metacognitions. The treatment gains were sustained at 1-year follow-up. Improvement in metacognitive beliefs (a hypothesized mechanism) showed a unique positive association with improvement in depression symptoms over 1 year. The current low relapse rates (12.8%) indicate that MCT is a potentially effective treatment for depression, but further studies comparing MCT for depression with other treatments are needed.

# REFERENCES


# DATA AVAILABILITY

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

# ETHICS STATEMENT

All subjects gave a written informed consent in accordance with the Declaration of Helsinki. The trial was registered at ClinicalTrials.gov and approved by the Regional Medical Ethics Committee in Norway (ref. no. 2011/1138).

# AUTHOR CONTRIBUTIONS

RH, OH, SS, and LK conducted the therapy in the trial. AW supervised the therapists. All authors have contributed in the writing of the manuscript.

# ACKNOWLEDGMENTS

The authors wish to thank the participants in this study and the student assistants Marte Rauø Strand and Charlotte Bjørkli.



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer SJ declared a past collaboration with the authors HN and AW to the handling Editor.

Copyright © 2019 Hjemdal, Solem, Hagen, Kennair, Nordahl and Wells. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A Comparison of Metacognitive Therapy in Current Versus Persistent Depressive Disorder – A Pilot Outpatient Study

*Lotta Winter1 \*, Julia Gottschalk1 , Janina Nielsen1 , Adrian Wells2,3 , Ulrich Schweiger <sup>4</sup> and Kai G. Kahl1*

*1 Department of Psychiatry, Social Psychiatry and Psychotherapy, Hannover Medical School, Hannover, Germany, 2 Division of Psychology and Mental Health, University of Manchester, Manchester, United Kingdom, 3 Greater Manchester Mental Health NHS Trust, Manchester, United Kingdom, 4 Department of Psychiatry and Psychotherapy, University of Lübeck, Lübeck, Germany*

#### *Edited by:*

*Ana Fonseca, University of Coimbra, Portugal*

#### *Reviewed by:*

*Sverre Urnes Johnson, Modum Bad Psychiatric Center, Norway Roger Hagen, Norwegian University of Science and Technology, Norway*

> *\*Correspondence: Lotta Winter winter.lotta@mh-hannover.de*

#### *Specialty section:*

*This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology*

> *Received: 28 March 2019 Accepted: 09 July 2019 Published: 06 August 2019*

#### *Citation:*

*Winter L, Gottschalk J, Nielsen J, Wells A, Schweiger U and Kahl KG (2019) A Comparison of Metacognitive Therapy in Current Versus Persistent Depressive Disorder – A Pilot Outpatient Study. Front. Psychol. 10:1714. doi: 10.3389/fpsyg.2019.01714*

Background: Metacognitive therapy (MCT) is a modern approach with demonstrated efficacy in current major depressive disorder (MDD). The treatment aims to modify thinking styles of rumination and worry and their underlying metacognitions, which have been shown to be involved in the initiation and perpetuation of MDD. We hypothesized that metacognitive therapy may also be effective in treating persistent depressive disorder (PDD).

Methods: Thirty depressed patients (15 with MDD; 15 with PDD) were included. Patients in both groups were comparable on depression severity and sociodemographic characteristics, but PDD was associated with more former treatments. Metacognitive therapy was applied by trained psychotherapists for a mean of 16 weeks.

Results: We observed a significant improvement of depressive symptoms in both groups, and comparable remission rates at the end of treatment and after 6 months follow-up. Furthermore, we observed significant and similar levels of improvement in rumination, dysfunctional metacognitions, and anxiety symptoms in both groups.

Limitations: The study is limited by the small sample size and a missing independent control group. The effect of the therapeutic alliance was not controlled. The quality of depression rating could have been higher.

Conclusions: We demonstrated that metacognitive therapy can successfully be applied to patients with PDD. The observed results were comparable to those obtained for patients with current major depressive disorder. Further studies with larger groups and a randomized design are needed to confirm these promising initial findings.

Keywords: metacognitive therapy, MCT, major depressive disorder, persistent depressive disorder, thinking style, metacognition, psychotherapy

# INTRODUCTION

Major depressive disorder (MDD) is a debilitating and often treatment-refractory mental health problem and a significant cause of lost life years (Whiteford et al., 2013). This has led to a call for a continuous process of innovation in the field of depression treatment (Hollon et al., 2014; Riihimaki et al., 2014). Comorbidity and chronicity are common in MDD and frequently complicate the course of disease (Penninx et al., 2011).

To date, little is known about the underlying reasons for chronic disease courses, and how to manage them. Former studies pointed to the role of previous episodes and subclinical symptoms as course modifiers in MDD, leading to the development of chronicity (Hardeveld et al., 2010, 2013; Seemuller et al., 2014). Furthermore, maladaptive cognitive processes such as rumination and worrying have been shown to negatively influence the course of depression (Nolen-Hoeksema et al., 2008; Lyubomirsky et al., 2015). Rumination is defined as a negative pattern of responding to distress by repetitively focusing on the meanings, causes, and consequences of one's depressive symptoms and has been linked with symptom severity and a chronic disease course in MDD (Kuehner and Weber, 1999; Wiersma et al., 2011; Gan et al., 2015). Klein (2010) summarized earlier onset, higher comorbidity rate, more extreme personality traits, higher levels of at least some cognitive biases, and greater suicidality as differences between chronic and non-chronic depression.

For the treatment of chronic depression several factors to consider when choosing treatment have been suggested (Kriston et al., 2014). A conclusion to date is that chronic depressions appear to require somewhat different approaches to treatment than non-chronic depressions (Klein, 2010).

In the following study, we aimed to examine the effects of a modern and effective form of psychological treatment, metacognitive therapy (MCT; Wells, 2000), in patients with current MDD and persistent depressive disorder (PDD). MCT differs significantly from other forms of psychotherapy in its focus on metacognitive processes and metacognitive beliefs as well as on regulating thinking styles, in contrast to traditional cognitive therapy where cognitive content is the target of psychotherapeutic intervention. MCT is based on the Self-Regulatory Executive Function model (Wells and Matthews, 1994) in which psychological disorder is caused and maintained by a transdiagnostic process of extended negative thinking and coping behaviors that lead to failures of effective selfregulation. The treatment focuses on changing cognitive processes and facilitating metacognitive modes of processing which can overcome inflexibility of attentional control that contributes to sustained repetitive negative thinking of rumination, worrying, and threat-monitoring. It is assumed that these thinking styles are also maintained by metacognitive beliefs, the latter are considered as a common factor in psychopathology leading to exacerbation of negative affect. In the metacognitive model of depression, positive metacognitive beliefs about the value of rumination in solving problems and overcoming low mood are thought to commonly occur. However, negative metacognitive beliefs concerning the uncontrollability of rumination and depressive thinking are considered central. The typical patterns of repetitive negative thinking in depression consist of rumination, threat-monitoring, and dysfunctional coping strategies (Wells, 2009). Consistent with this model, repetitive negative thinking is a transdiagnostic factor (Drost et al., 2014), rumination is an independent contributor to the maintenance of depressive symptomatology (Lyubomirsky et al., 2015; Yilmaz et al., 2015), and metacognitive beliefs contribute to depressive symptoms and rumination (Papageorgiou and Wells, 2003).

Some preliminary studies have examined the effects associated with MCT in current, recurrent, and postpartum MDD (Bevan et al., 2013; Callesen et al., 2014; Farahmand et al., 2014; Jordan et al., 2014; Normann et al., 2014; Dammen et al., 2015; Hagen et al., 2017). In each study, MCT was associated with large effect sizes and high levels of remission (Normann and Morina, 2018). So far, two studies have examined purely treatment-refractory cases (Wells et al., 2012; Papageorgiou and Wells, 2015), but no study to date has directly compared outcomes in current MDD versus PDD. As the effect sizes found in these studies were similar to non-refractory depression, the primary hypothesis of our pilot study was that MCT should have similar effectiveness in treating current MDD and PDD. We also predicted that the underlying maladaptive thinking style will change with treatment response in both subgroups.

# MATERIALS AND METHODS

The study was approved by the local ethics committee (No. 1343-2012). Participants were consecutively recruited from a waiting list of the psychotherapy outpatient clinic of the Department of Psychiatry, Social Psychiatry and Psychotherapy of the Hannover Medical School. They were either referred by local psychiatrists, local general practitioners, or from other departments of the Hannover Medical School. All patients gave written informed consent to participate in the study.

# Participants

Inclusion criteria were a diagnosis of current MDD or PDD according to DSM-IV and an age between 18 and 70 years. Exclusion criteria were: cognitive impairment, current substance use disorder, psychotic disorder, bipolar disorder, acute medical conditions such as cancer or heart failure and suicidality requiring inpatient treatment.

Fifty patients were contacted for a first screening. Of these, 20 were excluded as they did not meet study criteria or were not willing to participate. Thirty patients gave informed consent. Of these, according to DSM-IV, 15 patients were diagnosed as having current MDD and 15 patients were diagnosed with PDD. Depressive symptoms had lasted for at least 2 years in patients with PDD. Furthermore, all patients with PDD had former treatment with antidepressant medication, and 11 PDD patients had at least one trial of cognitive TABLE 1 | Baseline demographics and sample characteristics of the patients (intention to treat, *n* = 30).


*MDD, Current major depressive disorder; PDD, Persistent depressive disorder; GAD, Generalized anxiety disorder; OCD, Obsessive-compulsive disorder; PTBS, Post-traumatic stress disorder; PD, Personality disorder; ADHD, Attention-deficit hyperactivity disorder.*

behavioral therapy (CBT) without response (**Table 1**). In summary, PDD patients suffered from persistent and treatment-resistant depressive symptoms.

For all patients who were on antidepressant medication when entering the study, two criteria were mandatory: they had to be on the current dose for at least 3 months before starting with MCT and they had to agree not to change the medication or dose until the end of therapy. A total of 10/15 patients with MDD and 12/15 patients with PDD reported former depressive episodes.

Of all, 27 patients reached the post-treatment evaluation, and 20 patients reached the 6 months follow-up. Baseline pre-treatment data are given in **Table 1**.

# Assessment and Design

A comprehensive pre-treatment assessment included a mental status examination, a semi-structured interview to document sociodemographic information and to screen the available psychiatric and medical information for the presence or absence of exclusion and inclusion criteria. Comorbidity status was assessed using a standardized diagnostic interview (SCID-1/ SCID-2) (Wittchen et al., 1997). Depression severity was assessed using the German version of the clinician-rated 21-item Hamilton Depression Scale (Ham-D) (CIPS – Collegium Internationale Psychiatriae Scalarum, 2015). The severity of anxiety symptoms was assessed with the German version of the Beck Anxiety Inventory (BAI) (Kabacoff et al., 1997). Problematic metacognitive processes were evaluated using German versions of the Positive Beliefs about Rumination Scale (PBRS) (Papageorgiou and Wells, 2001b), the Negative Beliefs about Rumination Scale (NBRS) (Papageorgiou and Wells, 2001a), the short form of the metacognitions questionnaire MCQ-30 (Wells and Cartwright-Hatton, 2004), and the ruminative response scale (RRS) of the response styles questionnaire (Nolen-Hoeksema and Morrow, 1991).

Before therapy started, a preparatory session was used to give feedback on the diagnoses assessed and to set the patient's personal therapy goals. The treatment for depression followed the treatment manual by Wells (2009) which is available in German (Wells, 2011). Therapy was terminated by agreement between therapist and patient when subjective therapy goals were met (T1). Mean treatment duration was 16 weeks (±8) with a frequency of one session per week in both groups. Six months after the end of therapy, patients were contacted and assessed for follow-up (T2). Twenty patients reached the 6-month follow-up examination (T2).

The main outcome criterion was improvement of depressive symptoms evaluated with the HamD. Secondary outcome parameters were changes in BAI and the evaluation of the PBRS, NBRS, MCQ-30, and the RRS.

A complete set of questionnaires and interviewer-rated measures was administered at pre-treatment (T0), post-treatment (T1), and at the 6 months follow-up (T2).

Response, remission, and recovery rates based on depression symptoms were evaluated. Response was considered as a 50% symptom reduction. Remission was defined when a score ≤ 7 was reached on the HamD. Recovery was defined when a remission at T1 was stable for at least 6 months (T2) (DGPPN et al., 2009, adapted: June 2015).

# Therapist Competence

All therapists were graduates of the MCT Institute1 diploma program which is a 128-h training curriculum in metacognitive therapy that includes supervision by accredited MCT supervisors. Therapists for both groups were the same. Adherence and competence were checked by monthly expert supervision.

# Statistics

Statistical analysis was performed using IBM SPSS (version 24). Categorical variables were compared using Chi-square tests. Group comparisons at the beginning of the study were analyzed using *t* tests. To determine depressive symptoms over the course of the study, mixed model ANOVA with HamD sum score as the dependent variable, group as between-subjects variable, and time at pre-treatment (T0), post-treatment (T1), and follow-up (T2) as the repeated measures factor was performed. If the assumption of sphericity was violated, the Greenhouse Geisser correction was applied. Analyses were performed using intention-to-treat principles and using last observation carried forward (LOCF) for missing data. Effect sizes were analyzed using Cohen's *d* effect sizes.

# RESULTS

Patients with MDD and PDD were similar concerning gender distribution (66.6 versus 53.3%), age (42.9 ± 13.2 versus 39.3 ± 8.4 years), partner status (53.3% partnered in both groups), and employment status (93.3 versus 80% employed) (for *p*, see **Table 1**). Patients with PDD had significantly more trials of former psychotherapy during the present episode (80 versus 0% in the MDD group), and all patients in the PDD group had at least one former trial with antidepressant medication,

1 www.mct-institute.com compared to none in the MDD group. Considering the mean HamD score at T0 (**Table 2**), severity of depression was similar in both groups before therapy started. In both groups, severity was moderate to severe.

Comorbidity status was similar in both groups, with slightly more anxiety disorders in the MDD group, and slightly more personality disorders in the PDD group.

Depressive symptoms significantly improved in both groups as measured at T1 with HamD declining from 20.7 ± 5.7 to 5.1 ± 4.1 (*d* = −3.1) in PDD and from 21.2 ± 4.6 to 7.3 ± 5.7 (*d* = −2.7) in MDD. The mixed model ANOVA showed no statistically significant interaction effect between time and group (*F*(1.1,32) = 0.3, *p* = 0.62). Regardless of group, a significant main effect for time (*F*(1.1,32) = 127.6, *p* < 0.01) could be found. There was no statistically significant main effect for group, meaning that the groups did not differ significantly (*F*(1,28) = 1.2, *p* = 0.28).

With age as a covariate in a mixed model ANCOVA, there remained no statistically significant difference between the two groups (*F*(3,25) = 2.49, *p* = 0.083). A main effect for age could be found though (*F*(1,27) = 8, *p* < 0.05 at T1 and *F*(1,27) = 5.9, *p* < 0.05 at T2). After stratification by age, no interaction effect between time and age could be found (*F*(1.2, 32.2) = 1.5, *p* = 0.23). Comparing HamD scores and effect sizes in the stratified sample, it could be seen that the effect on symptom reduction was even stronger in the group of younger patients (HamD at T0: 20.7 ± 5.6, T1: 4.4 ± 4.2, T2: 3.3 ± 3.8, *d*(T0,T1) = −3.3, *d*(T0, T2) = −3.6) than the effect found in the group of older patients (HamD at T0: 21.2 ± 4.6, T1: 8.3 ± 5.3, T2: 7.1 ± 5.7, *d*(T0, T1) = −2.6, *d*(T0,T2) = −2.7).

In summary, all depressive symptoms improved significantly in all patients independent of type of depression and age.

Results of the secondary outcomes are presented in **Table 2**, demonstrating that both groups were similar in reduction of BAI sum scores, positive and negative beliefs about rumination scale score, and ruminative response scale scores.

TABLE 2 | Means, standard deviations, effect sizes, and summary statistics for the primary and secondary outcome measures comparing cases of persistent depression vs. cases of current major depression [intention to treat analysis (*n* = 30)].


*MDD, Current major depressive disorder; PDD, Persistent depressive disorder; HamD, Hamilton depression scale; BAI, Beck anxiety inventory; PBRS, Positive beliefs about rumination scale; NBRS, Negative beliefs about rumination scale; MCQ-30, Metacognition questionnaire; RRS, Ruminative response scale; d = 0.2–0.4: small effect; d = 0.5–0.7: medium effect; d ≥ 0.8: large effect.*

The ANOVA of the intention to treat analysis yielded significant improvements in the pre-treatment, post-treatment, follow-up comparison on all variables (main effect time: BAI: *F*(1.4, 38.9) = 14.9, *p* < 0.01; PBRS: *F*(2, 56) = 18.4, *p* < 0.01; NBRS *F*(1.6, 44.8) = 32.2, *p* < 0.01; MCQ-30: *F*(1.5, 42.3) = 27.6, *p* < 0.01; RRS: *F*(1.4, 37.9) = 33.2, *p* < 0.01). As **Table 2** shows, no interaction effects between the two groups could be found. None of the main effects for group were statistically significant. In brief, both groups improved to a similar degree on secondary outcome parameters over time.

In both groups, large effect sizes (Cohen's *d*) were found evaluating the pre-treatment post-treatment (T0-T1)- and pre-treatment-follow-up (T0-T2)- comparison of the HamD scores (T0-T2: PDD: *d* = −3.4, MDD: *d* = −2.9). Looking at the secondary outcome parameters, large effect sizes can be reported analyzing the scores of the PBRS (T0-T2: PDD: *d* = −1.6, MDD: *d* = −0.8), NBRS (T0-T2: PDD: *d* = −1.7, MDD: *d* = −1.1), MCQ-30 (PDD: *d* = −1.4, MDD: *d* = −1.2) and RRS (T0-T2: PDD: *d* = −1.9, MDD: *d* = −1.4). On the BAI, a large effect size was observed in PDD (T0-T2: *d* = −1.3) compared to a medium effect size in the MDD group (T0-T2: *d* = −0.7).

Response, remission, and recovery rates are presented in **Table 3**. Improvement of depressive symptoms was slightly better in PDD (response: 80%, remission: 80%, recovery: 80%) than MDD (response: 73.3%, remission: 66.7%, recovery: 66.7%) assessed by clinician-rated HamD. In no patient symptoms deteriorated. Treatment gains were ongoing in both groups after 6-month follow-up (HamD at T2: PDD: 4.2 ± 3.9, *n* = 12; MDD: 5.9 ± 6, *n* = 8), although there was more missing data in the MDD group at T2.

The number of MCT treatment sessions was only slightly different in each group (MDD: 15.5 ± 8.7 sessions versus 17.8 ± 7.5 sessions in the PDD group) (data not shown). Using Spearman's rank-order correlation, we did not find a correlation between clinical improvement and number of MCT sessions (data not shown).

All patients improved concerning HamD sum scores. Considering single items of the HamD scale, both groups improved most (≥2 points difference between T0 and T1) on item 1 ("depressed mood"). PDD patients also improved markedly (>2 points) on item 7 ("work and interests") and item 10 ("anxiety – psychic"). Concerning remaining items in remitted patients of both groups, no item could be identified as outstanding. All items in both groups of remitted patients had a mean score below 1 at the end of treatment (T1).



*HamD, Hamilton depression scale; MDD, Major depressive disorder; PDD, Persistent depressive disorder.*

# DISCUSSION

The main results of our study are that metacognitive therapy in patients with MDD and PDD was associated with significant improvement in terms of HamD sum score, response and remission rates at post-treatment and at 6 months after the end of treatment. Of particular interest, effect sizes and clinical outcomes were broadly comparable across patients with PDD and MDD. A slight difference can be seen in the level of recovery where we observed an almost 15% difference favoring those with PDD. While caution should be exercised in interpreting these data, they do suggest that MCT is associated with large improvements in both groups of patients. Furthermore, our results are encouraging in that patients with high levels of non-response to previous treatment may profit from MCT. Our results are not biased by different dosages of MCT as documented by a similar mean rate of MCT sessions in both groups. A further interesting aspect is that our data indicate that the effect of MCT is independent of patient's age. Effect sizes were higher in younger patients, but still large in older patients, meaning that possible differences in depressive symptoms relating to age do not hinder the therapy progress.

Our data are in accordance with other studies demonstrating strong efficacy of MCT in depression. In recent meta-analyses, large controlled effect sizes were reported favoring MCT over wait-list control and CBT in depression and anxiety disorders (Normann et al., 2014; Normann and Morina, 2018). A recent randomized study with 48 depressed participants from New Zealand compared MCT to CBT and reported a similar reduction in depressive symptoms with both treatments (Jordan et al., 2014), but therapists were not trained in MCT. Effect sizes were *d* = 0.96 in the MCT group and *d* = 0.60 in the CBT group at week 4. In an analysis of the effects of these treatments on executive functioning, MCT appeared to produce superior outcomes than CBT (Groves et al., 2015). A three-armed depression study from Iran including 10 patients in a MCT group, 10 patients in a CBT group, and 13 patients in a comparison group with pharmacological treatment equally showed similar outcomes in MCT and CBT (Ashouri et al., 2013), but again therapists were not trained in MCT. Case series studies from England and one from Denmark showed significant improvements in depressive symptoms, rumination, and metacognitive beliefs after MCT (Wells et al., 2012; Callesen et al., 2014).

The novel aspect of the current study is the comparison of effects in patients with current MDD against those with persistent MDD. To date, two published studies have examined the effects of MCT in chronic and persistent depression (Wells et al., 2012; Papageorgiou and Wells, 2015) with results that are comparable with those in the current study. However, these uncontrolled earlier studies did not directly compare the effects in MDD with those in PDD.

In the current study, we observed that in both treated cohorts large and significant improvements in underlying thinking processes (rumination) measured, e.g., by the RRS and metacognitive beliefs were shown. The levels of improvement were similar in each case, which suggests that persistent depression is not associated with a lower level of change in hypothesized underlying causal mechanisms in patients undergoing MCT. This raises the question of why the PDD group had failed previous treatment attempts with antidepressant medication or CBT. One possibility is that the earlier treatments did not directly modify metacognitive beliefs and reduce the extent of rumination. In fact, it is possible that failed treatment attempts may strengthen unhelpful metacognitive beliefs about the uncontrollability of depressive thinking such that patients with PDD are more likely to search for a solution to their depression that relies less on using their own executive control processes to overcome the problem. This could contribute to a persistence of rumination and maladaptive thinking patterns.

Moreover, the study addresses the question of distinguishing between different types of depression becomes less relevant when applying MCT. MCT modifies underlying processes of depression that are not explicitly targeted by CBT or other psychotherapy methods. Changing the style with which a person deals with cognitions and modifying metacognitive beliefs may be more important in beating depression than dealing with the content of negative automatic thoughts or schemas, but this hypothesis needs to be investigated further.

A recent study by Timm et al. (2017) demonstrated that changes in repetitive negative thinking are important not only on a trait level (macro-level), but also on a micro-level of momentto-moment experiencing during daily life. They found that both trait and state processes of affective and cognitive processes impact the longer course of major depression (Timm et al., 2017). An important question arises whether MCT effectively changes cognitions on both levels, and whether this may result in longlasting effects for prevention of further depression. According to our results, positive and negative beliefs about rumination and rumination itself changed significantly in both patient groups, accompanied by lasting remission after 6 months follow-up.

Interestingly, we did not find an association between number of MCT sessions and clinical improvement, which may suggest that in some patients, fewer MCT sessions may be sufficient for therapeutic success. This effect may also underlie the relatively large standard deviation in the number of MCT sessions in both groups, since reaching therapeutic goals was a criterion to end treatment.

Since we did not find an outstanding remaining HamD item in remitted patients of both groups, one may conclude that MCT has a global effect on all dimensions of MDD or PDD, respectively. In summary, depressive symptoms as assessed by expert rating (HamD) decreased in both study groups during MCT with high effect sizes. Dysfunctional metacognition (PBRS, NBRS, and MCQ-30) decreased with high effect sizes as well as the style of responding to rumination (RRS). Gender, age, and family status had no effect on the treatment outcome as assessed by multiple measurements ANOVA with the respective variables as confounders.

# LIMITATIONS

Due to the small sample size and a missing independent control group, the study does not prove efficacy of the treatment. We cannot control for non-specific factors such as the effect of the therapeutic alliance and also for factors such as the passage of time and repeated testing. We therefore have no information on possible improvements without any intervention. The HamD rating was not completed by independent raters, but by therapists involved in the study. Although it was not necessarily the therapist who did the therapy with the assessed person, this may have resulted in an overestimation of change within therapy. Also inter-rater reliability was not measured and may limit the results. In addition, the Hamilton rating scale for depression itself may count as a limitation. Even though it is one of the most commonly used measures for depression, some of its quality criteria are poor (Bagby et al., 2004). Furthermore, the standardized post-evaluation of diagnosis is missing due to the naturalistic design. One further limitation has to be kept in mind, we analyzed data using the LOCF procedure, which may influence effect sizes. The use of LOCF can be criticized (Lachin, 2016) as it may overestimate or underestimate treatment effects. However, in our study, we consider this a conservative method that would most likely cause differences between the two groups (our null hypothesis) when one of the groups is considered more treatment resistant. In practice, there were few missing values and so the impact is in any case likely to be small.

# CONCLUSIONS

Independent of the subtype of depression, metacognitive therapy was associated with significant improvement of depressive symptoms, symptoms of comorbid disorders, and changes in thinking styles and metacognitions. This indicates that the underlying processes of persistent depression and current depression may be equally modifiable by MCT irrespective of whether or not depression is considered treatment-refractory.

# DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

# ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the ethics committee of the Hannover Medical School with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the ethics committee of the Hannover Medical School.

# AUTHOR CONTRIBUTIONS

LW, US, and KK designed the original concept. JG and JN recruited the patients. AW trained the therapists in MCT. LW, JG, and JN carried out the assessments and therapies. LW, AW, US, and KK conducted the data analysis and wrote the paper.

# REFERENCES


# ACKNOWLEDGMENTS

We wish to thank our patients for participating in this study.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2019 Winter, Gottschalk, Nielsen, Wells, Schweiger and Kahl. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Neurobiological Mechanisms of Metacognitive Therapy – An Experimental Paradigm

Lotta Winter<sup>1</sup>† , Mesbah Alam<sup>2</sup>† , Hans E. Heissler<sup>2</sup> , Assel Saryyeva<sup>2</sup> , Denny Milakara<sup>3</sup> , Xingxing Jin<sup>4</sup> , Ivo Heitland<sup>1</sup> , Kerstin Schwabe<sup>2</sup> , Joachim K. Krauss<sup>2</sup> and Kai G. Kahl<sup>1</sup> \*

<sup>1</sup> Department of Psychiatry, Social Psychiatry and Psychotherapy, Hannover Medical School, Hanover, Germany, <sup>2</sup> Department of Neurosurgery, Hannover Medical School, Hanover, Germany, <sup>3</sup> Center for Stroke Research Berlin, Charité – Berlin University of Medicine, Berlin, Germany, <sup>4</sup> Department of Neurosurgery, Zhongda Hospital, Southeast University, Nanjing, China

## Edited by:

Gerald Matthews, University of Central Florida, United States

#### Reviewed by:

Susana Ochoa, Parc Sanitari Sant Joan de Déu, Spain Karin Carter, Greater Manchester Mental Health NHS Foundation Trust, United Kingdom Bartosz Zurowski, University Medical Center Schleswig-Holstein, Germany

\*Correspondence:

Kai G. Kahl kahl.kai@mh-hannover.de †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 10 September 2018 Accepted: 11 March 2019 Published: 04 April 2019

#### Citation:

Winter L, Alam M, Heissler HE, Saryyeva A, Milakara D, Jin X, Heitland I, Schwabe K, Krauss JK and Kahl KG (2019) Neurobiological Mechanisms of Metacognitive Therapy – An Experimental Paradigm. Front. Psychol. 10:660. doi: 10.3389/fpsyg.2019.00660 Introduction: The neurobiological mechanisms underlying the clinical effects of psychotherapy are scarcely understood. In particular, the modifying effects of psychotherapy on neuronal activity are largely unknown. We here present data from an innovative experimental paradigm using the example of a patient with treatment resistant obsessive-compulsive disorder (trOCD) who underwent implantation of bilateral electrodes for deep brain stimulation (DBS). The aim of the paradigm was to examine the short term effect of metacognitive therapy (MCT) on neuronal local field potentials (LFP) before and after 5 MCT sessions.

Methods: DBS electrodes were implanted bilaterally with stereotactic guidance in the bed nucleus of the stria terminalis/ internal capsule (BNST/IC). The period between implantation of the electrodes and the pacemaker was used for the experimental paradigm. DBS electrodes were externalized via extension cables, yielding the opportunity to record LFP directly from the BNST/IC. The experimental paradigm was designed as follows: (a) baseline recording of LFP from the BNST/IC, (b) application of 5 MCT sessions over 3 days, (c) post-MCT recording from the BNST/IC. The Obsessive-Compulsive Disorder- scale (OCD-S) was used to evaluate OCD symptoms.

Results: OCD symptoms decreased after MCT. These reductions were accompanied by a decrease of the relative power of theta band activity, while alpha, beta, and gamma band activity was significantly increased after MCT. Further, analysis of BNST/IC LFP and frontal cortex EEG coherence showed that MCT decreased theta frequency band synchronization.

Discussion: Implantation of DBS electrodes for treating psychiatric disorders offers the opportunity to gather data from neuronal circuits, and to compare effects of therapeutic interventions. Here, we demonstrate direct effects of MCT on neuronal oscillatory behavior, which may give possible cues for the neurobiological changes associated with psychotherapy.

Keywords: metacognitive therapy, local field potential, deep brain stimulation, treatment resistance, BNST/IC

# INTRODUCTION

fpsyg-10-00660 April 2, 2019 Time: 19:53 # 2

With the introduction of neuroimaging techniques in psychotherapy research, the neurobiological correlates of psychotherapeutic interventions have been increasingly investigated (Yang et al., 2014). A number of studies suggest that the progress and outcome of psychotherapy can be associated with neurobiological changes (Messina et al., 2013; Barsaglini et al., 2014; Weingarten and Strauman, 2015). Neuroimaging studies about psychotherapy effects have, however, only roughly demonstrated that changes in cognition and behavior through psychotherapy (mainly cognitive behavioral therapy; CBT) and neuronal changes in the brain are somehow interrelated. Accordingly, the exact moderating and mediating effects of psychotherapy on neuronal substrates are largely unknown (Sakai et al., 2006; Yang et al., 2014).

Obsessive-compulsive disorder (OCD) is a severe psychiatric illness, which is treated by psychotherapy and pharmacotherapy in the first instance according to current guidelines (National Institute for Health and Care Excellence [NICE], 2005; Deutsche Gesellschaft fuer Psychiatrie und Psychotherapie, Psychosomatik und Nervenheilkund e.V [DGPPN], 2013), with response and remission rates between 20 and 70% depending on the kind of treatment and the measured criterion (Fisher and Wells, 2005; Grados and Riddle, 2008). Metacognitive therapy (MCT) is a modern development in psychotherapy standing out by comparably short treatment duration, high effect sizes and transdiagnostic effects considering comorbid disorders (Normann et al., 2014; Sadeghi et al., 2015; van der Heiden et al., 2016). MCT is a cognitive therapy derived from the Metacognitive theory of psychological disorders (Wells and Matthews, 1994). Referring to its distinctive theoretical origin MCT focusses on metacognitive processes and metacognitive beliefs as well as on regulating thinking styles. This is in contrast to traditional cognitive therapy where cognitive content is the target of psychotherapeutic intervention. For people suffering from OCD it is a promising treatment option (van der Heiden et al., 2016). The effects of MCT on neurophysiological mechanisms which lead to clinical improvement have not been elucidated so far.

Brain function in OCD has been investigated using functional magnetic resonance imaging, structural brain morphology, positron emission tomography and EEG methods (Linden, 2006; O'Neill et al., 2013; Dohrmann et al., 2017; Moody et al., 2017; Atmaca et al., 2018). In particular, hyperactivity of the corticostriato-thalamo-cortical (CSTC) circuit has been proposed as the neurobiological basis of OCD (Saxena et al., 2001). This concept achieved further support by studies demonstrating increased cerebral blood flow in the CSTC by symptom provocation (McGuire et al., 1994; Rauch et al., 1994; Adler et al., 2000), and decreased activation after treatment with selective serotoninreuptake inhibitors or psychotherapy (Brody et al., 1998).

However, the CSTC model does not take into consideration the role of the amygdala and its interaction with the frontal lobe in mediating fear and anxiety in OCD (Milad and Rauch, 2012). The amygdala and the associated bed nucleus of the stria terminalis (BNST, also called the extended amygdala) constitute an integrative center for emotions and emotional behavior, whose role in mediating fear and anxiety in OCD is a hotspot of current research (Lesting et al., 2011; Daldrup et al., 2016; Kohl et al., 2016).

Deep brain stimulation (DBS) targeting the BNST and the neighboring internal capsule (IC) is a novel therapeutic strategy in treatment resistant OCD (trOCD) that exerts its effects via electric stimulation, thereby possibly modulating the activity of pathological neuronal circuits (Naesstrom et al., 2016). In line with this, imaging and DBS studies suggest that the BNST and orbital frontal cortex are implicated in the pathophysiology of OCD (Luyten et al., 2016).

The DBS treatment approach provides a unique opportunity to study the neural activity of subcortical brain areas in patients. Further, postoperative recording via externalized leads of the electrodes provides the opportunity to gather data on brain activity in pathological disease states as well as changes of brain activity after psychotherapeutic intervention. We here present a new experimental paradigm to investigate the neuronal effects of psychotherapy, exemplified with MCT, in a patient with trOCD treated with DBS.

# MATERIALS AND METHODS

# Operative Procedure

The data reported in this study were recorded from a 51-yearold left-handed male with drug- and CBT- refractory OCD, who underwent implantation of DBS electrodes in the bed nucleus of the stria terminalis/ internal capsule (BNST/IC) bilaterally. This patient showed OCD symptoms mainly in the domains of checking, ordering and symmetry with an onset in the 1980s. In the pre-assessment prior to the first surgery he presented a sum score of 39 on the German version of the Yale Brown Obsessive Compulsive Scale (Y-Bocs), (Hand and Büttner-Westphal, 1991; Jacobsen et al., 2003). A current depressive episode could be excluded. The patient was drug free during the study procedure. Before DBS the patient was treated according to the German S3-guidelines for OCD (Deutsche Gesellschaft fuer Psychiatrie und Psychotherapie, Psychosomatik und Nervenheilkund e.V [DGPPN], 2013), and had received two qualified treatments using disorder specific cognitive-behavioral therapy including exposure and response prevention, combined with recommended drug treatments. Currently MCT is not part of this guideline, and was therefore not considered before DBS treatment. The study was approved by the Ethics Committee of Hannover Medical School and the patient gave written informed consent prior to the study onset.

The quadripolar DBS electrodes (model 3387, Medtronic, Minneapolis, MN, United States) had four platinum-iridium cylindrical contact surfaces (1.27 mm diameter and 1.5 mm length) and a contact-to-contact separation of 1.5 mm. DBS electrodes were implanted bilaterally with CT-stereotactic guidance, aided by magnetic resonance imaging, and microelectrode recording in the BNST/IC under local anesthesia. Microelectrode recording was used to define the trajectory within BNST and IC. Contact 0, the lowermost contact, was placed in the BNST, and the upper contacts were placed in the IC. Details of target localization during the intraoperative procedure and implantation of the neurostimulation system are described elsewhere (Winter et al., 2018). Appropriate electrode placement was confirmed by postoperative stereotactic CT. The implantable pulse generator was implanted under general anesthesia. Appropriate electrode placement was confirmed by postoperative stereotactic CT.

# Experimental Design

fpsyg-10-00660 April 2, 2019 Time: 19:53 # 3

The paradigm was part of a larger study on the effects of DBS of the BNST/IC in OCD (in preparation). The time period between implantation of DBS electrodes and the implantable pulse generator (IPG) was used for the experiments. During this period LFPs were obtained directly from the contacts in the BNST/IC. Elements of MCT were applied five times and neurophysiological oscillatory activity was recorded via the DBS electrodes before and after MCT. No stimulation was performed during this period. **Table 1** presents an overview of the protocol.

# Metacognitive Therapy

Metacognitive therapy is a theory-based development in modern psychotherapy. Founded on the Self-Regulatory Executive Function Model (S-REF) (Wells and Matthews, 1994), MCT postulates that psychiatric disorders are a result of disturbed information processing. Perseverative thinking styles and inflexible attention patterns are maintained by unhelpful metacognitions. The aim of the treatment is to help the patient develop new ways of controlling attention, relating to thoughts and inner events and modify underlying metacognitions. Part of the intervention strategies is to practice detached mindfulness and attention training. With detached mindfulness the patients

#### TABLE 1 | Study protocol.


Y-Bocs, Yale Brown Obsessive Compulsive Scale; HamD, Hamilton Depression Scale; BDI-II, Beck Depression Inventory II; OCD-S, Obsessive-Compulsive Disorder Scale; MCT, metacognitive therapy; ATT, Attention Training Technique; IPG, implantable pulse generator.

can develop the experience that one can step back from thoughts and other inner events and let these control themselves without doing anything actively. This experience can be presented and practiced using different metaphors and exercises described in the treatment manual (Wells, 2009). Attention training (ATT) aims to help strengthen the awareness of attentional control (Wells, 1990). To practice, a sound file can be used. The training consists of actively listening to several presented sounds. Instructions help to focus and regulate attention in three phases. The first phase is to practice selective attention. Here the task is to focus on individual sounds whilst trying not to get distracted by other sounds. The second phase involves rapid switching of attention between different sounds and spatial locations. The last phase practices dividing attention by trying to widen the attention to attend as many sounds as possible.

According to the manual average treatment duration is around 12 sessions. As the time between implantation of the electrodes and the stimulator is limited the paradigm gives time for five sessions only. Therefore the content of the sessions does not follow the manual. We chose to investigate two components of therapy which are detached mindfulness and ATT. They were both practiced with the patient. In the first session, he was provided with an individual case formulation of his OCD to socialize the concept. In this session, detached mindfulness was introduced. Detached mindfulness was also trained in session 2. In session 3, ATT was introduced and the audio file of the German version of ATT was provided for practice. The patient was asked to practice at least three times per day on that day and the next day. ATT and practice of detached mindfulness were repeated in session 4. Session 5 consisted of supervised ATT only. Each session lasted approximately 45 min.

# Psychometric Measures

The German version of the Obsessive-Compulsive Disorder Scale (OCD-S) (Wells, 2009) was used to evaluate the effect of MCT subjectively. The focus was to get information on the effects MCT may have even in this treatment resistant case. Instead of following the original instruction to rate the items considering the last week the patient was asked to refer to the time frame since he last answered the questionnaire. The OCD-S is a selfrating-scale used in MCT to evaluate the therapy progress. The questionnaire consists of four main questions and 22 sub-items. The patient was asked to rate items 1 to 3 on a scale ranging from 0 to 8. Item 4 asks for percentages (0–100%). As shown in **Table 1** the OCD-S was obtained before surgery (T0), after session 1 (T1), after session 3 (T2) and after the last session (T3).

# Local Field Potentials and EEG Recording

The electrophysiological recordings were undertaken 2 days after the implantation of the electrodes prior to the first MCT session and on day 6 after the last MCT session (see **Table 1**). DBS leads were still externalized during this time frame. The EEG and LFP recording was made in a resting condition. The patient sat in an arm chair in a relaxed and calm condition. We explicitly instructed the patient not to move the head or body and to

keep his eyes open. The recording was running for at least 300 s (**Figures 1A,B**).

The local field potentials (LFP) were obtained from adjacent bipolar contact pairs (0 to 1) in the left and right BNST from the implanted DBS electrodes. LFP signals were amplified 50.000 fold and filtered (bandwidth 0.5–100 Hz) using a D360 amplifier (Digitimer Ltd., Welwyn Garden City, Hertfordshire, United Kingdom) at a sampling rate of 512 Hz through a 1401 A-D converter (CED, Cambridge, United Kingdom) onto a computer using Spike2 software. Simultaneous surface EEG recordings were taken over frontal cortical areas (F3 and F4) according to the International 10–20 System using Ag–AgCl contact surface electrodes referenced to the mastoid and band pass filtered at (0.5–100 Hz) and the sampling rate was 512 Hz. Electrode impedances were kept below 2 k.

# Local Field Potentials and EEG Data Analysis

Due to an expected intrinsic non-stationarity in the LFP and EEG signals we segmented 300 s recorded data in to three epochs of equal length (3 × 100 s) for power of spectral analysis in different frequency bands e.g., theta, alpha, beta, and gamma. The analysis of spectral power or coherence of neural oscillatory activity measured in EEG and LFPs have provided a new insight into brain mechanisms of information processing in different neurological and neuropsychiatric disorders (Marceglia et al., 2007; Uhlhaas and Singer, 2010; Bowyer, 2016).

Three epochs of 100 s without major artifacts were used for frequency-domain signal processing from simultaneous recordings of BNST/IC LFP and frontal cortical EEG. After eliminating 50 Hz artifacts using a finite impulse response (FIR) notch filter, data were normalized by subtracting the mean amplitude and dividing the standard deviation, which allowed the frequency domain signals to be pooled and compared with less influences from individual/non-specific differences. Frequency domain transformation was applied by computing the Fast Fourier Transform (FFT) spectra from blocks of 512 samples, which resulted in a frequency resolution of 1.953 Hz. Hanning's window function was applied to overcome spectral leakage phenomena. For compa rison of power at different frequency bands, the areas under the computed power density spectrum in specified frequency ranges, i.e., theta (4–8 Hz), alpha (8–12 Hz), beta (12– 30 Hz), and gamma (30–100 Hz) were calculated and averaged. Further, power-spectra were normalized and expressed as percent of total power.

Functional relationships between the BNST/IC LFP and frontal cortical EEG were estimated by means of coherence using the methods described by Halliday et al. (1995). Coherence is one mathematical method of signal processing that can be used to determine the strength of oscillatory synchronizations across the brain networks in different neurological and neuropsychiatric disorders (Bowyer, 2016). Coherence of oscillatory signals provides a frequency-domain measure of the linear phase and amplitude relationships between signals (Alam et al.,

2017). In this finite measure of values from 0 to 1, 0 indicates no linear association and 1 indicates a perfect linear association. Coherence is defined as the normalized crossspectrum according to the formula "Coh x, y (f) = Sxy(f) divided by squared root of Sx(f) −Sy(f)," where x(t) and y(t) are two random, zero-mean processes and Sx(f), Sy(f), and Sxy(f) are the values of their auto-and cross-spectra at a given frequency (f). Representative epochs of 100 s without major artifacts were used for the signal processing. A finite impulse response (FIR) 50 Hz notch filter and 100 Hz lowpass filter was used. Fourier transformation with blocks of 512 samples using a Welch periodogram in a custom MATLAB (MathWorks, Inc.) resulted in a frequency resolution of 1.953 Hz. Hanning's window function was applied to overcome spectral leakage phenomenon. For comparison of power at different frequency bands, the power of the density spectrum in specified frequency ranges was calculated and the coherence was averaged (Kim et al., 2016).

# Statistics

The statistical procedure of a paired t-test was used to verify the difference of spectral power between pre-therapy and posttherapy in the subject. P-value < 0.05 was considered as statistically significant.

# RESULTS

# Obsessive-Compulsive Disorder Scale

The MCT sessions resulted in immediate symptomatic changes of the OCD-S items which were scored. **Table 2** shows the results of repeated OCD-S measurement during MCT. Only those items are presented that show the main changes.

# Electrophysiological Measures

Several measures in LFP and EEG recordings and their coherence give cues to possible impacts of MCT. In the following the main findings are demonstrated. All results are shown as mean ± standard error of the mean.

Prior to MCT therapy, the mean percentage of relative power of theta (4–8 Hz) band LFPs was higher on the left (80.26 ± 1.39%) and on the right (69.19 ± 3.27%) BNST/IC. Whereas, after MCT the relative power of theta band LFPs decreased on the left (61.55 ± 1.04; p < 0.01) and right (55.51 ± 1.12%; p < 0.04) BNST/IC region, respectively (**Figure 2A**).

Prior to MCT, the mean percentage of relative power of alpha (8–12 Hz) band LFPs was lower in the left (9.23 ± 1.02%) and on the right (12.81 ± 1.22%) BNST/IC. Whereas, after MCT the relative power of alpha band LFPs increased on the left (19.45 ± 0.77%; p < 0.001) and right BNST/IC region (21.07 ± 0.41%; p < 0.01; **Figure 2B**).

Prior to MCT, the mean percentage of relative power of beta (12–30 Hz) band LFPs on the left BNST/IC was lower (6.85 ± 0.22%), whereas, after MCT the relative power of beta LFPs increased in the left BNST/IC region (11.62 ± 0.42%; p < 0.01; **Figure 2C**).

Prior to MCT the mean percentage of relative power of gamma (30–100 Hz) band LFPs was lower on the left (3.64 ± 0.14%) and on the right (6.51 ± 0.38%) BNST/IC. After MCT the relative power of gamma LFPs increased on the left (7.37 ± 0.33%; p < 0.01) and right BNST/IC region (9.3 ± 0.5%; p < 0.01 and p < 0.05; **Figure 2D**).

The coherence of oscillatory activity in the frontal cortex and the BNST/IC LFP was analyzed before and after MCT to delineate differences in spectral peak amplitudes and phase locking strength of neuronal network synchronization. A decrease in the mean value of theta frequency band

TABLE 2 | Course of selected items of the Obsessive-Compulsive Disorder Scale (OCD-S) ratings: T0 represents the baseline intensity of OCD symptoms before the first MCT session.


Further measurements show the results after the first MCT session (T1), after the third MCT session (T2), and after the fifth/ last MCT session (T3).

coherence was observed on the left (p < 0.001) and right (p < 0.05) frontal cortical EEG and BNST/IC LFP after MCT (**Figure 3A**). No differences in alpha, beta and gamma coherency for the factor therapy were noted (**Figures 3B–D**).

# DISCUSSION

The present study describes symptom reduction after MCT in trOCD and a possible link between psychotherapeutic interventions and changes in neuronal activity of associated brain network.

Remarkably, OCD symptoms were reduced after only 5 sessions of MCT. Some symptoms remitted already after session 1. Fisher and Wells (Fisher and Wells, 2008) have shown previously that MCT might even be superior to CBT as it appears to be relatively time efficient and an easily delivered treatment. Further, MCT is a straightforward treatment that can be applied even in a laboratory setting and it may be particularly suited to investigate network activity via implanted DBS electrodes.

In our experimental setting, direct recordings from the DBS electrodes revealed a decrease of theta activity and an increase of alpha, beta and gamma-band oscillatory activity in the BNST/IC after MCT. Moreover, MCT was associated with suppression of theta band coherence of the frontal cortex and the BNST/IC.

Our results on basal activity are in line with previous studies who found relatively low alpha and beta power in OCD patients recorded via DBS electrodes in different targets (Guehl et al., 2008; Neumann et al., 2014). More remarkably, clinical and experimental studies have also shown that enhanced cortico-limbic network synchronization in the theta band is correlated with severity of symptoms in OCD, and reduction in such coupling strength may be correlated with clinical improvement (Cavanagh et al., 2011; Voon et al., 2017; Rappel et al., 2018). Enhanced neuronal synchronization in specific frequency bands has been linked to clinical symptoms in movement disorders and disturbed behavior, specifically in

theta and beta bands (Nini et al., 1995; Linkenkaer-Hansen et al., 2004; Wilson et al., 2004; Womelsdorf and Fries, 2007). Excessive synchronization therefore is considered pathological with secondary maladaptive signaling (Popovych and Tass, 2014). A recent study has shown an increase in theta activity in the frontal cortex of OCD (Kamaradova et al., 2018). Further, error-related negativity in OCD is thought to be associated with excessive theta synchronization (Luu et al., 2004; Trujillo and Allen, 2007).

Enhanced theta band synchronization, however, may not be specifically attributed to OCD because such a spectrum of synchronization has also been described in patients with dystonia, Tourette syndrome, and psychiatric disorders such as schizophrenia and attention deficit/hyperactivity disorder (Maling et al., 2012; Alam et al., 2015; Kim et al., 2016; Kohl et al., 2016; Neumann et al., 2017, 2018; Won et al., 2018).

This novel paradigm potentially shows much promise to be considered as a possible methodology in future treatment process studies and its main limitation is that the experimental setup was conducted in only one patient, thus far. Also, we cannot fully rule out that the surgical procedure itself had an influence on the initial oscillatory activity in the BNST/IC network, although we started recording of LFP activity only 24 h after electrode implantation to reduce the risk of artifacts, and to give time to the neuronal network to adapt. In our study design it was difficult to rule out the effect of DBS electrodes implantation induced changes to the neuronal activity. However, with regards to current knowledge of DBS in OCD it can be emphasized that the improvement of OCD symptoms most likely only appears after delivery of high frequency electric stimulation. Clinical studies of treatment refractory OCD patients have shown that post-surgery DBS electrodes implantation without current delivery i.e., sham stimulation did not show significant improvement. However, following 12 months of chronic DBS, 4 of 6 patients responded with a decrease of ≥35% in the YBOCS score from baseline (Goodman et al., 2010). Further, studies of DBS in OCD patients have shown altered LFPs before DBS and compensation of abnormal LFP after DBS (Neumann et al., 2014; Pearson et al., 2017).

In contrast to DBS therapy our results have shown compensation of altered oscillatory activity of LFPs after MCT in the OCD patient.

No information can be given addressing the question whether symptom reduction through MCT alone in this treatment resistant case would have lasted as DBS started once the IPG was implanted. The patient initially came to receive DBS. He additionally participated in the described paradigm, but was then treated and monitored according to DBS protocol.

# CONCLUSION

We here present an experimental paradigm to directly investigate neuronal oscillatory activity in patients with trOCD before and after application of MCT by recording LFP via implanted DBS electrodes. Our results suggest that a dominant decrease in the theta frequency band in the BNST/IC and in frontal cortical coherency, and an increase in the relative power of alpha, beta and broad band gamma frequency oscillatory activity in the BNST/IC may be associated with OCD symptom reduction by MCT. Our preliminary results may give possible cues for neuronal circuitry changes in OCD secondary to psychotherapy.

# REFERENCES


# ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the ethic committee of the Hannover Medical School. The subject received oral and written information about the study, participation was voluntary and he gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the ethic committee of the Hannover Medical School.

# AUTHOR CONTRIBUTIONS

LW, MA, HH, KS, JK, and KK planned the original concept. JK and AS performed surgery. MA and HH were in charge of data recordings and handling of all equipment needed. LW performed the intervention sessions. MA, DM, XJ, IH, and KS did the data analysis. All authors contributed to writing the paper and interpretation of the results.

# ACKNOWLEDGMENTS

We wish to thank Dr. med. Götz Lütjens for his support.



terminalis/internal capsule in obsessive-compulsive disorder. World Neurosurg. 111, e471–e477. doi: 10.1016/j.wneu.2017.12.084


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Winter, Alam, Heissler, Saryyeva, Milakara, Jin, Heitland, Schwabe, Krauss and Kahl. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Metacognitive Therapy Versus Cognitive Behavioral Therapy: A Network Approach

Sverre Urnes Johnson<sup>1</sup> \* and Asle Hoffart1,2

<sup>1</sup> Modum Bad Psychiatric Center, Vikersund, Norway, <sup>2</sup> Department of Psychology, University of Oslo, Oslo, Norway

A network perspective on mental problems represents a new alternative to the latent variable perspective. Diagnoses are assumed to refer to a causal network of observable mental problems or symptoms (observables). The observable symptoms that traditionally have been considered indicators of latent traits (disorders) are taken to be directly related causal entities. Few studies have investigated how different therapies affect a network-structure of symptoms and processes. In this study, three anxiety symptoms, three depression symptoms and mechanisms in the form of cognitions, metacognitions, worry and threat monitoring were selected. The network structure over the course of therapy for metacognitive therapy (MCT) and Cognitive behavioral therapy (CBT) was investigated. It was hypothesized that worry, attention, and metacognition would be important nodes in MCT and that cognitions would be important in CBT. The data used in the analysis are from a RCT where 74 patients with comorbid anxiety disorders were randomized to either transdiagnostic MCT or disorder-specific CBT. Symptoms and mechanisms were measured every week. The data was analyzed using the multilevel vector autoregressive (mlVAR) model, which is currently the most developed method to analyze multivariate time series in multiple subjects and construct networks. The results indicate that there were different networks of symptoms and mechanisms in MCT and CBT. Central nodes in both treatments are worry and attention, however, the node of negative metacognitive beliefs about uncontrollability was more central in the MCT treatment. The results are consistent with predictions from the S-REF model.

#### Edited by:

Lora Capobianco, Manchester Mental Health and Social Care Trust, United Kingdom

#### Reviewed by:

Roger Hagen, Norwegian University of Science and Technology, Norway Jennifer Jordan, University of Otago, Christchurch, New Zealand

\*Correspondence:

Sverre Urnes Johnson Sverre.johnson@modum-bad.no

#### Specialty section:

This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology

Received: 10 August 2018 Accepted: 12 November 2018 Published: 30 November 2018

#### Citation:

Johnson SU and Hoffart A (2018) Metacognitive Therapy Versus Cognitive Behavioral Therapy: A Network Approach. Front. Psychol. 9:2382. doi: 10.3389/fpsyg.2018.02382 Keywords: metacognitive therapy, CBT, mlVAR, network approach, mechanisms

# INTRODUCTION

Outcome in psychotherapy research is traditionally measured in relation to presence or absence of a disorder or severity of diagnostic symptoms. The disorder is assumed to be a latent entity, whereas the symptoms are viewed as indicators of this entity. The different indicators assumed to reflect the latent construct are rated and summarized in a total score. Mechanisms of change are treated as latent constructs and thus measured by total scores of relevant indicators. Thus, treatments are supposed to influence latent mechanisms that affect an underlying disorder manifested by specific symptoms (Borsboom and Cramer, 2013). However, there are several problems with this latent variable perspective (Borsboom, 2017; Hoffart and Johnson, 2017). First the latent variable

perspective does not permit that symptoms cause each other. The symptoms are supposed to be caused by the underlying latent variable. In psychopathology, however, it makes sense that for example lack of sleep could lead to increased nervousness, or that lack of activity could lead to low mood. Thus, symptoms clearly influence each other. A new statistical approach called the network-approach takes this interdependence between symptoms into consideration (Borsboom and Cramer, 2013). The network approach conceptualizes symptoms as mutually interacting, often reciprocally reinforcing, elements of a complex network (Borsboom and Cramer, 2013). Thus different anxiety disorders do not exist as latent entities, but exists in the network of the symptoms. Each symptom could cause the release of others symptoms, and the comorbidity between disorders is explained by so called bridge symptoms or overlapping symptoms in the networks (Fried et al., 2017). This new approach opens up new questions regarding which and what kind of symptoms are the most central, so called centrality. Centrality-indices provide information about what kind of symptoms are most closely related to other symptoms, thus a promising target for interventions. In contrast, a latent disorders approach provides a sum-score that indicates the degree of anxiety or degree of depression. There are several differences between latent disorders models and network-models described elsewhere (Fried and Cramer, 2017; Bringmann and Eronen, 2018), but – for our purposes – the critical aspects are that modeling the data with network analysis gives new specific information about the relationship between symptoms, and that different treatments may activate different networks in similar patients.

An important purpose of therapy-models is to describe what maintains different symptoms or nodes in the network. These mechanisms, derived from theory, could also be called micro networks (Hoffart and Johnson, 2017). Both metacognitive therapy (MCT; Wells, 2009) and Cognitive behavioral therapy (CBT; Beck, 1976) specify micro-networks, however, the included variables are different. The origin of MCT can be traced to the early publications of the self-regulatory executive function model (S-REF; Wells and Matthews, 1994, 1996). The S-REF model consists of three interacting levels: a level of automatic and reflexively driven processing units; a level of attentionally demanding, voluntary processing; and a level of stored knowledge or self-beliefs (Wells and Matthews, 1996). The level of stored knowledge, or metacognitions, is involved in the development and maintenance of anxiety and emotional distress because these metacognitions give rise to the specific use of transdiagnostic strategies in the form of worry, rumination, and threat-monitoring. CBT originally developed by Beck (1967, 1976), is today an umbrella term of different therapies. In it's traditional form schemas are thought to influence negative automatic thoughts that again drive specific symptoms. Thus, dysfunctional cognitions are thought to be crucial mechanism of change in CBT.

Several studies have investigated the role of metacognitions and cognitions in anxiety disorders (Smits et al., 2012; Johnson et al., 2018). Most of the studies have been conducted on at between-person level. Thus, it is investigated whether higher metacognitions or cognitions than the group mean predict anxiety. The reference-point is then the mean of all the patients. Therapists, however, are mainly interested in the within-person level, that is, if deviations from the persons own mean on a mechanism variable, are related to personal change on an outcome variable. Two studies have investigated if metacognitions predict anxiety on a within-person level, which requires repeated assessments points (Hoffart et al., 2018; Johnson et al., 2018). To our knowledge no studies have investigated MCT and CBT on a network of symptoms and mechanisms separating the within- and between person effects. Studies investigating network analysis have mainly used symptoms (Borsboom, 2017), even though there has been a call for networks that also include key mechanisms, such as cognition and metacognition (Jones et al., 2017).

The aim of this paper was to investigate the network structure of symptoms and mechanisms over the course of therapy in MCT and disorder-specific CBT separately. Since MCT and CBT emphasize different mechanisms it was hypothesized that worry, threat-monitoring and metacognition would be important nodes in MCT and that the cognitions would be important in CBT.

# MATERIALS AND METHODS

The materials of this paper come from a randomized controlled trial (RCT) comparing MCT and CBT and are thoroughly described in two other papers (Johnson et al., 2017, 2018).

# Participants

Participants were referred to treatment at the Department of Anxiety Disorder at Modum Bad Psychiatric Center in Norway. Modum Bad is a specialized hospital running an inpatient treatment program for treatment resistant patients with anxiety disorders. The patients were referred because they had not benefited sufficiently from outpatient treatment. Recruitment was designed to be liberal using the clinical criteria for treatment used at the department. To be eligible for participation in the study, participants had to meet criteria for a principal DSM-IV disorder, exceeding 4 on the clinical severity rating (CSR), of PTSD, social phobia (SAD) or panic disorder with and without agoraphobia (PD/A). The Anxiety Disorders Interview Schedule (IV) (ADIS; Brown et al., 1994) was used to diagnose the patients. Further, participants had to have experienced failure of at least one structured psychological treatment, be 18 years of age or older, Norwegian speaking, and provide informed consent. Following the procedures at the department of Anxiety Disorders at Modum Bad, patients were excluded if (a) in a clinical context they would have required immediate treatment or simultaneous treatment that could interact with the treatment in unknown ways, (b) had current DSM-IV diagnosis of organic mental disorders, (c) clear and current suicidal risk, or (d) current substance abuse. All participants had to terminate the use of psychotropic medications before treatment, and were contacted before treatment to ensure that they were medicationfree or had started discontinuation of medications. The study was approved by the Norwegian regional ethical committee

(2013/209/REK South-East). All subjects gave written informed consent in accordance with the Declaration of Helsinki.

Patients were randomized to MCT or CBT stratified on their principal disorder.

All patients that started treatment, 74 participants (n = 38 CBT, n = 36 MCT) were included in the sample analyzed. Seven participants did not complete the treatment program, leaving 67 who completed all the treatment sessions (n = 33 in CBT, n = 34 in MCT). The average age was 42 (SD = 12.8), and there were 45 female and 29 male patients. The patients had on average 3.7 (SD = 1.6) diagnoses at the start of treatment, 41 % of the patients had a personality disorder. The duration of their anxiety problem was M = 16.1, SD = 11.8. A majority of the patients (80.5%) were either out of work or on a disability allowance, which indicates a sample with chronicity and poor level of functioning.

# Treatments

The number of sessions for completers were equivalent in both conditions (M = 9.4, SD = 1.7). The sessions in CBT lasted longer, due to the protocols of SAD and PTSD, which lasts 90 min. All therapists were trained in MCT or CBT, and the adherence and competence ratings of every session were above 4 on a scale from 0 to 6 (Johnson et al., 2017).

# Metacognitive Therapy

The MCT treatment consisted of a manualized treatment protocol for the generic MCT model (Wells, 2009). MCT is a process-oriented therapy. The protocol deemphasizes disorderspecific aspects, and focuses instead on challenging positive and negative metacognitions that drive the use of worry, rumination, threat-monitoring and coping behaviors, called the cognitive attentional syndrome (CAS), to regulate emotions.

# Cognitive Behavioral Therapy

Treatments in the disorder-specific CBT condition were the most extensively documented cognitive treatments of PD/A (Clark, 1986; Wells, 1997), of social phobia (Clark and Wells, 1995; Wells, 1997), and of PTSD by using prolonged exposure (PE) therapy (Foa et al., 2007). CBT is a content-based psychotherapy where the focus is on challenging the content of thought's. Different catastrophic beliefs are thought to be central in different disorders. In PD/A thoughts about going crazy or loosing control are central, in social phobia thoughts about being embarrassed in front of others are key, and in PTSD thought that the world is dangerous and that the trauma is dangerous are central thoughts.

# Differences Between MCT and CBT

In MCT processes in the form of worry and the metacognitions that leads to the unhelpful thinking style is targeted. Thus, MCT also works with cognition, but on the level of metacognition. This can be exemplified with a patient who brings up a thought in session about being worthless. In CBT this thought could be taken for a possible schema about being worthless, and the reality of this belief could be tested. In MCT the statement about being worthless could be seen as either a trigger for rumination or an endpoint of rumination. The goal of the therapist is to challenge the dysfunctional metacognitions that drives the use of rumination. Further differences can be found in the use of exposure. In CBT, especially the PE-treatment, trauma-exposure is a critical component. In MCT exposure is not necessary, and reliving the trauma is not part of the treatment.

# Measures

In network analysis specific items are selected that captures the key processes that are under investigation. The two authors wanted to select central anxiety and depression symptoms as well as CBT-mechanisms and MCT-mechanisms. They independently selected the most appropriate items, and met to discuss whether there were any disagreements. There were none. The most relevant CBT-processes, MCT-processes and symptoms where then selected before the analysis. Three central anxiety symptoms were chosen from the Beck Anxiety Inventory (BAI; Beck et al., 1988), and three depression items from the Patient Health Questionnaire (PHQ-9; Kroenke et al., 2001). Two central cognitions were also chosen from the BAI, while three central processes in MCT where chosen from the CAS-1 (Wells, 2009). An overview of the items and measures can be found in **Table 1**.

# Statistical Analysis

The patient filled out the questionnaires every Monday during the course of therapy, giving a longitudinal dataset. The multilevel vector autoregressive (mlVAR) model is currently the most developed method to analyze multivariate time series in multiple subjects and construct networks (Epskamp et al., 2017). In time series data, consecutive responses are not likely to be independent (e.g., anxiety at one time point predicts anxiety at the next), thus violating a typical statistical assumption. The autoregressive (AR) part of mlVAR accounts for this time dependency within an individual by regressing a variable at time t on a lagged (measured at the previous time point, t-1) version of that same variable. The VAR model is a multivariate extension of the AR model. In VAR, variables are regressed on a lagged version of the same variable and all other variables of the multivariate set. Finally, the multilevel (ml) extension of the VAR allows the modeling of time dynamics across individuals. In mlVAR, each subject is assumed to have their own VAR model, and the VAR parameters vary randomly across individuals. In a mlVAR analysis three

TABLE 1 | Abbreviation and meaning of the different nodes in the analysis.


networks are estimated: a temporal network in which withinperson effect predicts different nodes on the next time-point (lag 1), a contemporaneous network in which a node predicts another node at the same time-point, and a between-person network in which the overall score over the course of therapy are associated with other variables. The three network structures generated from our data are visualized through the R-package qgraph (Epskamp et al., 2012). The networks were calculated separately for MCT and CBT.

Centrality indices were calculated (Opsahl et al., 2010). These parameters indicate how central a node is in a given network. Outward degree is the sum of all outgoing connections, while inward degree is the sum of all incoming connections. Betweenness centrality takes into account both the direct and indirect connections of a symptom. Thus, a node with high betweenness centrality is a node that is located on many paths between other symptoms. It is thus an important node for how the network develops. Node-strength is the sum of all incoming and outgoing connections for the node.

# Model Assumptions

There are three central assumptions in using the mlVAR model. The first assumption is that the time intervals between two consecutive measurements are approximately equal. In this study the measures were included a week a part, every Monday, thus the assumption was fulfilled. The second assumption concerns stationary, indicating that the mean and variance of the series must stay unchanged. Stationary is often a problem in longitudinal dataset in clinical psychology, since most of the variables of interest are expected to change as a consequence of treatment. The variables were detrended according to the procedure outlined by Curran and Bauer (2011), and new variables were constructed consisting of the person-mean of all the measurements points as wells as the residuals from the detrending procedure. We used the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test for the null hypothesis that a timeseries is level or trend stationary on the residuals from the detrending procedure (Kwiatkowski et al., 1992). The test was conducted separately for each of the patients and 11 variables per patient in each group using the R package tseries 0.10-43. The KPSS test indicated that the majority of time-series was trend (91%) and level (77%) stationary for MCT and CBT. The third assumption is the specific order of the model. We present only the results of the baseline models with lag-1 predictors included due to parsimony. In the network analysis a significance level of 0.05 for the individual effect was used. There was no correction for multiple testing, due to the exploratory nature of the study.

# RESULTS

Positive relationship between symptoms is marked with green lines, while negative relationships are marked with red. The strength of the relationship between symptoms is represented by the thickness of the arrows in the figures. The thicker the arrow

between two symptoms, and the closer the arrows are together in the figure, the stronger the relationship.

The temporal network shows the averaged within-person effects from 1 week to the next. In the MCT network (see **Figure 1**, right side), the belief about uncontrollability of thoughts predicts threat-monitoring. Threat-monitoring, is also predicted by fear of losing control. Worry predicts the degree of feeling shaky. Thus, worry is a central node in the network, which is shown in the centrality indices in **Figure 4**. The CBT-network is more densely connected (see **Figure 1**, left side). The anxiety symptom of the heart pounding and raising is a central node. It is negatively predicted by the symptoms of shaky/unsteady and little interest. So higher levels of shaky/unsteady and little interest leads to less heart pounding at the next time point. Furthermore, the cognition fear of losing control predicts the cognition fear of dying, which, in turn, is predicted by heart pounding and racing. As shown in the centrality indices in **Figure 5**, worry, sleep and threat-monitoring are also central nodes.

The contemporaneous network captures the averaged withinperson associations at the same measurement point, controlled for the lag-1 temporal effects. In the MCT network (**Figure 2**, right side), the belief about uncontrollability of thoughts is central as well as worry and threat-monitoring. The symptom of shakiness (sha) is also central in the network. This is evident in the centrality-plot in **Figure 6**. In the CBT plot (**Figure 2**, left side), worry and attention are still important nodes, but beliefs about uncontrollability of thoughts are less important. The centrality-plot is given in **Figure 7**.

The between-person network in **Figure 3** shows the partial correlation between person-means on the 11 variables. In MCT, worry is again a central node, and worry is connected to threat-monitoring and threat-monitoring to the belief about uncontrollability of thoughts. The red line from worry to interest indicates that higher degree of worry is associated with less interest. Furthermore, a central symptom in the MCTnetwork was the feeling of numbness and lack of interest (see **Supplementary Figure S1**). In the CBT network the network is less connected, numbness is not a central symptom, but worry, feeling down and belief about uncontrollability of thoughts is (see **Supplementary Figure S2**).

# DISCUSSION

The purpose of this study was to investigate the psychological networks in anxiety disorder patients receiving MCT versus CBT. The analysis indicated that the networks reflected the therapy form they received, especially with respect to the importance of nodes specified from the S-REF model. Across all three types of networks, worry and threat-monitoring were central nodes. Thus, worry and threat-monitoring are central in the maintenance of other symptoms or mechanisms in treatmentresistant anxiety disorder. It is previously argued that worry and threat-monitoring are important transdiagnostic mechanisms of change (Wells, 2009), but this results gives further empirical evidence for how these variables interact with other mechanisms and symptoms.

It was hypothesized that worry and metacognition would be important nodes in MCT. The networks for MCT indicated that for the three different networks metacognition, worry and attention to threat were densely connected. These results are consistent with the S-REF model, which predicts that metacognitions should affect the use of worry or threatmonitoring, as strategies for regulating low-level input or emotion (Wells and Matthews, 1996). It is also expected that the association between these variables should be strong, since these mechanisms are in the focus of treatment. Furthermore, lack of interest was a central symptom in MCT, indicating that targeting this symptom would also affect other symptoms. The clinical implications from the MCT-networks can be summarized by the centrality indices for the three different networks, especially the strength, which indicates how much change in a node will affect other nodes. On a between-person level, that is the overall means scores in therapy, reduction in worry and lack of interest was central in MCT. However, between-person relationships in longitudinal models can give limited information about how symptoms and processes develop over time (Bos et al., 2017). The temporal and contemporaneous networks, on the other hand, reflect within-person relationships and are therefore of particular relevance for therapeutic theories and the study of mechanisms of maintenance and change. This is because the mechanisms depicted in theories concern within-person relationships, that is, how change in a process variable in a given patient relates to change in an outcome variable during therapy. Consequently, it is also these two types of networks that provide clinical implications. In particular, nodes with high out-strength in the temporal network are targets for potentially effective interventions as changes in such nodes are likely to propagate through the network. It is evident in the MCT networks that worry, fear of losing control, and the meta-cognitive belief of uncontrollability of thoughts should be primary targets of treatment. These clinical implications are in accordance with MCT (Wells, 2009).

In the CBT networks the cognitions were associated on the temporal networks in association with specific symptoms. Thus, there is a relationship between catastrophic beliefs and symptoms, as would be expected from theory (Beck, 1976). Worry and attention were also central variables, which gives further support for the S-REF model. In the temporal network, there were also negative relationships between some bodily symptoms and between the depressive symptom disinterest and a bodily anxiety symptom (heart pounding/racing). These relationships probably reflect oscillation between reciprocally excluding emotional and bodily systems and are more a basis for therapeutic observation than for manipulation. The clinical implications from the CBT-networks can be summarized by the centrality indices. Threat-monitoring, worry, and sleep problems have high out-strength and should be targeted. Also heart pounding/racing has high out-strength. None of the cognitions have high out-strength, thus the clinical implications of which processes that should be targeted, is not in accordance with CBT-theory.

However, does the apparent influence of processes from the S-REF-model indicate that CBT therapists to a larger degree

should target MCT-processes? Targeting the content of cognition using verbal reattribution (CBT technique), and proposing to leave the thoughts alone with detached mindfulness and postpone worry (MCT technique), could create confusion for the patient. In many ways the goal of the therapist in MCT and CBT is also incompatible. In MCT the goal is to change how patients respond to thoughts by changing metacognitions that drive the CAS. In CBT the goal is to change the content of thoughts. The finding that core processes, specified from the S-REF model, is the central nodes, rather implies the importance of targeting these processes in a metacognitive framework. It is previously shown that MCT was more effective then CBT in this treatment sample (Johnson et al., 2017). It is also evident that the two treatments have different networks, which may indicate treatment specificity. Thus, one possible explanation for the results in the RCT (Johnson et al., 2017) could be that MCT to a larger extent activated the association between worry, attention and metacognition.

Lack of interest being an important symptom in both MCT and CBT may be a bit surprising since the sample consisted of anxiety disorder patients. However, the present sample had high degree of comorbidity, with an average of 3.7 diagnoses (Johnson et al., 2017). Thus, the high centrality of lack of interest may be due to the treatment resistant aspect of the sample. Overall, the network-analysis across CBT and MCT gives a clear message about the importance of targeting worry and threat-monitoring in therapy.

Using network analysis allows for a more specified understanding of which symptoms and mechanisms that are crucial for therapeutic interventions. Specific predictions from therapy theories can be tested using network analysis on longitudinal data. This paper gives further evidence for the MCT-model, with the edges between the variables in the S-REF model being significant in both treatments. Network analysis could also be implemented in routine care situations. By having patients answer several questions repeatedly during a specific time-frame before treatment, an individual network can be made. The clinicians can then start to work directly on the most central symptom. Future research should investigate whether this specific use of network analysis could lead to larger treatment-effects.

Even though the paper has several strengths in the form of novel analysis and a new way to investigate treatments effects, several limitations should be acknowledged. In this paper the

# REFERENCES


different networks were not tested against each other using significance tests, since that would likely be a power problem. The sample size is limited, even though normal for psychotherapy studies. To the authors knowledge there are no implemented packages in R to estimate stability and accuracy in longitudinal networks (Epskamp et al., 2018). In order to test if the results found with the mlVAR method could be replicated, the results should have been compared with a second validation dataset. However, no such data set was available at the time of the writing. Thus, future replications of the results are needed.

Analyzing psychotherapy data using a network approach is in its early stages, and it is therefore important to explore possible differences between treatments that could be tested in larger samples at a later stage. Furthermore in this study items from the BAI were used, thus fear of losing control and fear of dying might not be the most representative items for catastrophic beliefs. The items chosen for the concept of negative metacognitions about uncontrollability of thoughts such as, "I cannot control my thoughts," is not representative for all aspects of metacognitions. Other aspects of metacognition like positive metacognitions, cognitive confidence, need for control and cognitive self-consciousness should also be investigated. In our models we used a t-1 lag, representing a week. Other relationship between the nodes could exist on other timeframes and should be investigated.

# AUTHOR CONTRIBUTIONS

SJ performed the analysis and wrote the first draft of the manuscript. AH commented and assisted on the analysis and the writing of the manuscript.

# ACKNOWLEDGMENTS

We thank the Department of Anxiety Disorder at Modum Bad Psychiatric Center for the data collection.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg. 2018.02382/full#supplementary-material


Liebowitz, D. A. Hope, and F. R. Schneier (New York, NY: Guilford Press), 69–93.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Johnson and Hoffart. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Metacognitive Beliefs as Predictors of Return to Work After Intensive Return-to-Work Rehabilitation in Patients With Chronic Pain, Chronic Fatigue and Common Psychological Disorders: Results From a Prospective Trial

#### Edited by:

Lora Capobianco, Manchester Mental Health and Social Care Trust, United Kingdom

#### Reviewed by:

Marcantonio M. Spada, London South Bank University, United Kingdom Bjørn Lau, University of Oslo, Norway Calvin Heal, The University of Manchester, United Kingdom

\*Correspondence:

Henrik B. Jacobsen henrik.borsting@gmail.com

#### Specialty section:

This article was submitted to Psychology for Clinical Settings, a section of the journal Frontiers in Psychology

Received: 24 October 2018 Accepted: 10 January 2020 Published: 06 February 2020

#### Citation:

Jacobsen HB, Glette M, Hara KW and Stiles TC (2020) Metacognitive Beliefs as Predictors of Return to Work After Intensive Return-to-Work Rehabilitation in Patients With Chronic Pain, Chronic Fatigue and Common Psychological Disorders: Results From a Prospective Trial. Front. Psychol. 11:70. doi: 10.3389/fpsyg.2020.00070

#### Henrik B. Jacobsen1,2,3 \*, Mari Glette<sup>4</sup> , Karen W. Hara5,6 and Tore C. Stiles<sup>7</sup>

<sup>1</sup> Department of Pain Management and Research, Division of Emergencies and Critical Care, Oslo University Hospital, Oslo, Norway, <sup>2</sup> Department of Psychology, Faculty of Social Sciences, University of Oslo, Oslo, Norway, <sup>3</sup> CatoSenteret Rehabilitation Center, Son, Norway, <sup>4</sup> Department of Circulation and Medical Imaging, Faculty of Medicine and Health Sciences, Norwegian University of Science and Technology, Trondheim, Norway, <sup>5</sup> Department of Public Health and Nursing, Faculty of Medicine and Health Sciences, Norwegian University of Science and Technology, Trondheim, Norway, <sup>6</sup> Norwegian Labour and Welfare Administration, Oslo, Norway, <sup>7</sup> Department of Psychology, Faculty of Social and Educational Sciences, Norwegian University of Science and Technology, Trondheim, Norway

Background: Metacognitions are associated with work status, but no research has examined to what extent metacognitions before treatment and change in metacognitions following treatment predict return to work (RTW) prospectively. The present study aims to address these two gaps in knowledge.

Methods: 212 patients on long-term sick leave (>8 weeks) with extensive fatigue, chronic pain conditions and/or mental distress received 3.5 weeks of intensive rehabilitation treatment, aimed at returning them to work. Only part of the population (n = 137) had complete follow-up data on metacognitions. Metacognitions were measured with the Metacognitions Questionnaire 30 (MCQ-30), while RTW was measured using official registry data from the Norwegian Labor and Welfare Service. A registry record of participation in competitive work ≥2.5 days (50% work participation) per week, averaging over 14 weeks, was chosen as an outcome reflecting a successful RTW. The registry data spanned a total of 56 weeks per participant.

Results: Our results indicated that baseline MCQ scores was not associated with RTW. This was analyzed for the total MCQ score as well as for all subscales. We observed substantial changes in metacognitions following treatment, and a 1-point change in the total sum of metacognitive beliefs was associated with 5% greater odds for successful RTW at all time points (p = 0.040), while a 1-point change on the subscale of beliefs about the need to control thoughts gave 20% greater odds for successful RTW (p = 0.016).

Conclusion: Metacognitions concerning the need to control thoughts appear to have a significant influence on patients return to work. Here, we observed that a change in these beliefs following treatment substantially affected RTW over the course of 1 year.

Keywords: rehabilitation, return-to-work, metacognition, prospective, pain, fatigue syndromes

# INTRODUCTION

In recent years the metacognitive model has been associated with work participation and absence (Nordahl and Wells, 2017a,b), as well as changes in common mental disorders (Solem et al., 2009; Wells et al., 2012). Depression, anxiety, persistent pain, and fatigue are common justifications for long-term sick leave in Norway (Jacobsen et al., 2015), and most other western countries (Henderson et al., 2005).

However, clinical and epidemiological studies highlight that there is considerable comorbidity amongst anxiety, depression, chronic pain, and fatigue (Kessler et al., 2007; Reme et al., 2011; Jacobsen et al., 2015). This overlap is supported by recent data where the specific reasons justifying sick leave vary, but clinical symptomatology and disorders overlap significantly (Jacobsen et al., 2015; Hara et al., 2017b).

Return to work (RTW) rehabilitation using psychological interventions has been somewhat successful for both musculoskeletal disorders and common mental health disorders. A recent meta-analysis showed a small effect size when psychological rehabilitation is compared to a "treatment as usual" condition (g = 0.16) (Finnes et al., 2019). This effect size was similar regardless of diagnoses justifying sick leave and different psychological interventions (Finnes et al., 2019), lending support to interventions targeting transdiagnostic processes of change (Loisel and Anema, 2013; Hara et al., 2017b).

A transdiagnostic stance that may further our understanding of factors that may implicate individuals RTW is the metacognitive model (Wells and Matthews, 1994, 1996).

According to the metacognitive model, psychological distress and emotional disorders are maintained by the activation of a maladaptive thinking style called the cognitive attentional syndrome (CAS). The CAS is characterized by repetitive negative thinking in the form of rumination and worry, and is associated with increased self-focused attention and maladaptive coping behaviors. The CAS is maintained by individual's metacognitive beliefs, which can be broken down into positive and negative metacognitive beliefs. Positive metacognitive beliefs concern the usefulness of worry (e.g., If I worry I will be prepared), while negative metacognitive beliefs concern the uncontrollability and dangerousness of worry (i.e., worrying could make me lose control) (Wells and Matthews, 1994, 1996).

Recently, studies have begun to evaluate the influence of the metacognitive model on RTW (Nordahl and Wells, 2017a,b). Nordahl and Wells (2017b) investigated the crosssectional association of metacognitive beliefs and work status in individuals with social anxiety disorder. They found that greater negative metacognitive beliefs were associated with individuals being out of work. More specifically, beliefs regarding the need to control thoughts were greater in those who were out of work.

More broadly, Nordahl and Wells (2017a) evaluated if metacognitive beliefs could predict work status. After controlling for gender, presence of a diagnosed mental health disorder, and trait anxiety (vulnerability to emotional disorder), they found that metacognitive beliefs regarding the need for mental control was a significant predictor of work status over and above the presence of a mental health disorder, and emotional vulnerability. Nordahl and Wells (2017a) highlight that metacognitive beliefs regarding the need to control may lead to increased worrying, threat monitoring, and attempts to control thoughts, which likely decreases cognitive processing capacity for work and impact on individuals interpretations of their ability to work effectively (Nordahl and Wells, 2017a).

Coping strategies might play a significant role in terms of understanding the sick leave process over time. People suffering from depression and anxiety tend to improve symptoms or workrelated functioning in the short-term if pushed toward work, but they are vulnerable for falling out again due to anxiety (Knudsen et al., 2013; Oyeflaten et al., 2014). The development of the CAS might play a role in this cyclical pattern of stress and sick leave. Repetitive negative thinking has been shown to delay homeostatic recovery following recovery from induced stress (Capobianco et al., 2018). Similarly, Jacobsen et al. (2014) found that Norwegians on sick leave had a dysregulated stress response in response to an induced stressor. A dysregulated stress response when faced with psychosocial stressors has been associated with depression, anxiety and pain (Kudielka et al., 2007), and is considered by many as a hallmark of chronic fatigue (Wyller et al., 2009).

However, a controversial finding within the field of RTW is the lack of a substantial relationship between symptom levels and work participation (Henderson et al., 2005). However, strong associations have been found between a long duration of depression and work disability (Lagerveld et al., 2010), moreover lifestyle factors affected by symptom severity have also been documented, which again could affect work participation (Blank et al., 2008). Metacognitions can play a crucial role in the resurgence of symptoms, but their relation to RTW has only been investigated cross-sectionally (Nordahl and Wells, 2017a,b). Thus, longitudinal studies are highly warranted (Myhre et al., 2014).

This study aimed to investigate the influence of metacognitions on RTW in a population on long-term sick leave with chronic pain, chronic fatigue and common psychological disorders. RTW was measured over the course of 56 weeks following completion of a common, on-site occupational rehabilitation program. As such we aimed to

evaluate: (1) if self-reported metacognitions at baseline are associated RTW in a 12-month period in patients attending an occupational rehabilitation program, (2) if changes in selfreported metacognitions from baseline to time of discharge of the rehabilitation program are associated with long-term RTW.

# MATERIALS AND METHODS

The current study was an explorative analysis nested within a randomized controlled trial investigating the effectiveness of telephone-guided follow-up versus standard RTW follow-up after on-site occupational rehabilitation. The overarching study is registered in ClinicalTrials.gov (No. NCT01568970).

Project participants, design and flow have been detailed in previous publications (Hara et al., 2017a,b, 2018), as such the subsequent paragraphs provide a brief overview of the project.

# Participants

Participants were referred by general practitioners (GPs) or other medical specialists to a 3.5-week intensive, inpatient rehabilitation from January 2012 to June 2013. The RTW rehabilitation took place at Hysnes Rehabilitation Centre located in the county of Trøndelag, Norway. Upon inclusion participants were invited to take part in the aforementioned study of boosted follow-up. The boosted follow-up consisted of six phone calls from their RTW-coordinator where they discussed progression toward work. Prior to inclusion the participants were assessed by an interdisciplinary team consisting of a physician, psychologist and a physical therapist. Participants completed a comprehensive questionnaire at baseline prior to their first meeting with the assessment team, following which informed consent was obtained and the data from the baseline questionnaire was made available to the researchers.

# Eligibility and Exclusion Criteria

Participants were eligible for the study if referred for either/or persistent pain, fatigue, depression or anxiety to inpatient rehabilitation, participants had to be between 18 and 59 years of age and had to have a clearly stated goal of wanting to RTW. In addition, they had to receive temporary medical benefits due to work incapacity (duration over 8 weeks, partial or fulltime). In Norway this involves being on one of two benefits that both require sickness certification; either sickness benefit (compensates for loss of income for employees or others with equivalent rights earned through previous participation in paid work) or work assessment allowance (for those who have either already received sickness benefits for the maximum period of 52 weeks, or have not earned the right to sickness benefits through previous employment).

Participants were to state a self-defined goal of increasing participation in competitive work, be adequately treated for health problems demanding acute care, be able to communicate in Norwegian and to maintain basic daily care for themselves during a stay at the rehabilitation centre. Participants were excluded from the study if they suffered from ongoing mania, psychosis or suicidal ideation, active substance abuse and addiction. Or if they reported pregnancy, planning to enter/return to studies rather than competitive work, incomplete study registration procedure, not registered as receiving temporary medical benefits, or not completing the rehabilitation program due to acute injury/disease or personal/family reasons.

# Study Setting

The 3.5-week inpatient occupational rehabilitation program consisted of individual and group sessions of mental and physical training and work-related problem solving. Pairs of RTW coordinators were in charge of coordinating and executing the on-site program for groups of maximum eight participants. Activities were organized around 6–7 h "workdays" with weekends free. Collaboration with GPs, participant work place and the social security office was initiated on-site, and participants had prepared their own action plan for RTW with guidance from on-site RTW coordinators and community stakeholders. The on-site program is described in detail elsewhere (Fimland et al., 2014).

# Primary Outcome

The primary outcome was (re)entry to the ordinary work force analyzed from baseline and up to 1 year (56 weeks) after discharge. The primary outcome variable was dichotomous and defined as participation in competitive work ≥2.5 day (18.75 h) per week, using four different time periods with 14 weeks between each time point.

# Independent Variable

The Metacognitions Questionnaire-30 (MCQ-30; Wells and Cartwright-Hatton, 2004) is a 30-item measure evaluating metacognitive believes across give subscales: (1) positive metacognitive beliefs about the usefulness of worry (e.g., Worrying helps me cope); (2) Negative metacognitive beliefs regarding the uncontrollability and dangerousness of worry (e.g., when I start worrying I cannot stop); (3) Beliefs about cognitive confidence (e.g., "I have a poor memory"); (4) Beliefs about the need to control thoughts (e.g., "Not being able to control my thoughts is a sign of weakness"); (5) Beliefs about cognitive self-consciousness (e.g., "I pay close attention to the way my mind works"). Items are scored from 1 to 4 ("do not agree," "agree slightly," "agree moderately," "agree very much"). Cronbach's alpha coefficients for these subscales range from 0.72 to 0.93, with test-retest correlations of: 0.75 (total score), 0.79 (positive beliefs), 0.59 (uncontrollability/danger), 0.69 (cognitive confidence), 0.74 (need for control), and 0.87 (cognitive selfconsciousness) (Wells and Cartwright-Hatton, 2004).

# Covariates

The Hospital Anxiety and Depression Scale [HADS (Zigmond and Snaith, 1983)] evaluates symptoms of anxiety and depression. The scale includes 14 items with two subscales: anxiety and depression. Items are scored using a four-point Likert scale ranging from 0 to 3. In a review of HADS in Norwegian adults the correlations between the two subscales varied from 0.40 to 0.74 (mean 0.56). Cronbach's alpha for HADS-A varied from

0.68 to 0.93 (mean 0.83) and for HADS-D from 0.67 to 0.90 (mean 0.82) (Zigmond and Snaith, 1983; Bjelland et al., 2002). When investigated in the current sample, Cronbach's alpha for the total sum score of the HADS scale had an average of 0.86, with the HADS-D having a mean Cronbach's alpha of 0.82 and the HADS-A having a mean Cronbach's alpha of 0.90.

The Chalder Fatigue Questionnaire [CFQ (Chalder et al., 1993)] consists of eleven questions asking about physical and mental fatigue and is frequently used to measure symptoms in chronic fatigue patients. Each item has four response categories (0–4), which are scored bi-modally 0-0-1-1. When scored, the 11 items are summed and gives each participant a score on a scale of 0–11. This eleven-item scale has been validated for a Norwegian adult population with a cut-off on symptom intensity ≥4. Cronbach's alpha has been calculated for all items (range 0.88–0.90). Split half reliability has also been calculated (0.86 and 0.85, respectively) (Chalder et al., 1993; Loge et al., 1998). When investigated in the current sample, Cronbach's alpha for this CFQ scale had an average of 0.86.

Chronic pain was measured with an item from Short Form-8 (SF-8) asking "How much bodily pain have you had the last week?" (None, very mild, mild, moderate, severe, and very severe). This scale has been validated as a self-report measure of chronic pain in Norwegian populations. As this is a one-item measurement, alpha values are not applicable. The item has been shown to have an intra-class correlation coefficient of 0.66 (95% CI 0.65–0.67) (Ware et al., 2001; Landmark et al., 2012).

# Statistical Methods

Descriptive statistics are used to report the participants' baseline socio-demographic, health, psychological and work-related characteristics. t-tests of change on the MCQ-30 sum score are investigated as well as its subscales pre-post intervention.

Generalized estimated equations (GEE) was performed to analyze the dichotomous outcome variable (≥2.5 days of competitive work per week) using repeated measurements (RTW per 14-week period) and an unstructured working correlation structure. The variable time was treated as a categorical variable. A GEE analysis was used as it allows for the association between MCQ-30 and RTW to be estimated across several timepoints while considering the correlation between timepoints.

The first 14-week period immediately after occupational rehabilitation was used as reference category. Each 14-week follow-up period was added to the model as a as a dummy variable (i.e., post rehabilitation weeks 1–14, weeks 15–28, weeks 29–42, weeks 43–56). Precision was measured with 95% confidence intervals (CI).

To investigate the associations between RTW and change in metacognitions, each participant's MCQ-30 total score was calculated at baseline as well as immediately after the participants completed rehabilitation, and a change score was calculated subtracting the post from the pre-value. These change scores were then used to analyze the association between change in metacognitions and probability for RTW over the four different follow-up periods.

As a sensitivity analysis to evaluate whether the observed patterns differed at different time points, interaction terms between the studied variable and each registration time-point were included in the model. Odds ratios (OR) are reported. Every GEE model was adjusted for age, gender and the underlying intervention of the randomized controlled trial. Precision was measured with 95% CI and p < 0.05 was considered statistically significant. Data analysis was performed using STATA version 15 (StataCorp. 2015. College Station, Texas, United States).

# RESULTS

To be eligible for participation in the current study you had to have registry data for outcome of RTW at preintervention and over 56 weeks, as well as baseline (preintervention) data on the MCQ-30, HADS, CFQ, and SF-8. This resulted in 212 eligible participants, however, as there was a software problem during data collection, only 137 participants completed the MCQ-30 at post intervention. Our final study population consisted predominantly of females on work assessment allowance (n = 137). Most of the participants reported a combination of chronic pain (76.6% SF-8 > 3), fatigue (89.0% CFQ ≥ 4), and also reported mental distress (61.3% HADS > 8). Further demographics reported at baseline are presented in **Table 1**.

In order to report the absolute number of participants reaching successful outcome criteria at all the four follow-up time points, we calculated the number of participants registered as working at least 50%, averaged over a 14-week period, at the four selected follow-up time points. The raw RTW data showed that n = 15 (10,3%) met criteria at the first time point (14 weeks after rehabilitation), n = 23 (16,5%) at the second time point (28 weeks), n = 33 (23,7%) at the third time point (42 weeks), and n = 37 at the fourth time point (27,1%) (56 weeks).

In **Table 2**, dividing the participants into those who achieved at least 50% RTW (n = 39) and those who did not (n = 98), the

TABLE 1 | Baseline characteristics of the study population (n = 137) are either presented as percentages of total N, or as mean and standard deviation (SD).


baseline scores on MCQ-30 and its subscales, as well as changes from baseline to immediately after completing rehabilitation on MCQ-30 are presented.

t-tests of change and absolute change is reported. Paired t-tests indicated significant changes on metacognitions in those who returned to work, but also in the larger group not achieving RTW. On both the total sum the MCQ-30 and the subscales of cognitive confidence and beliefs about the need to control thoughts the mean change was greater in the group achieving at least 50% work. In the subscale reporting beliefs about thoughts concerning danger and uncontrollability those achieving RTW had a significant change from baseline to immediately after rehabilitation, those not achieving RTW did not (**Table 2**).

# Associations From the MCQ-30 Scores at Baseline

Baseline scores on the MCQ-30 were analyzed for association with RTW at all follow-up measurements spanning a year (56 weeks). None of the MCQ-30 subscales at baseline were associated with RTW at the four time points when adjusted for age, gender and the underlying intervention of the randomized controlled trial. Further details are presented in **Table 3**.

# Associations From the MCQ-30 Change Scores

Substantially higher work participation was observed for participants that reported change on the total sum of MCQ-30 from pre to post treatment. There was an association of 5% greater odds for successful RTW at all time points (p = 0.04) per 1-point change on the total sum of MCQ-30. On the subscale of need to control thoughts there was a 20% increase in the OR of reaching the successful outcome per 1-point change, when looking at the association over all time points (see **Table 3**). None of the other metacognition subscales reached statistical significance.

Sensitivity analysis: The interaction between total MCQ score and time was not statistically significant at any timepoint with reference to the first 14-week time period following rehabilitation. This was also the case for all subscales measured at baseline. Change in the subscale of beliefs about the need to control thoughts showed a significant interaction with time for the second time period 15–28 weeks (OR 0.78, CI 0.65–0.94, p = 0.01) and the third time period 29–42 weeks (OR 0.78, CI 0.62–0.98, p = 0.03) with reference to the first 14-week time period following rehabilitation.

# DISCUSSION

The current study evaluated the prospective association between baseline metacognitions, changes in these beliefs after multidisciplinary rehabilitation, and sustainable return-to-work over the course of 56 weeks (RTW). We did not find an association between the subscales of metacognitions or the total score of metacognitive beliefs at baseline and subsequent RTW.

However, when investigating changes in metacognitions, both a change in the total sum of metacognitions and metacognitions about the need to control thoughts substantially affected RTW. None of the interaction effects with time changed the results in a significant way, indicating that the effect from MCQ-30 on RTW is stable over time.

The results indicate that metacognitions about the need to control thoughts could be of particular interest in the work rehabilitation context. Previously published data on metacognitions and work status have shown that the need to control thoughts is significantly different in those that are working and not working when suffering from social

TABLE 2 | Averaged change on metacognitive beliefs reported by participants by those returning to work at least 50% (n = 39) within the 56-week period indicated as group 1, and those not meeting this criterion (n = 98), indicated as group 0.


The MCQ-30 scores are pre and post intervention. All variables were significance tested with a paired t-test and degree of change was described as absolute change and as a Hedges g effect size.

TABLE 3 | Predictive associations presented as odds ratios (OR) for achieving successful 50% return to work (RTW) given metacognitions reported by participants at baseline and change in these metacognitions pre to post intervention.


anxiety (Nordahl and Wells, 2017b). The same substrate of metacognitions has been associated with work status above and beyond the existence of a mental disorder and trait anxiety in the same study population (Nordahl and Wells, 2017a). Thus, in combination, this lends support to the idea that metacognitive beliefs regarding the need for mental control (i.e., "not being in control of my thoughts is a weakness") could be implicated in the ability to sustain work participation.

These specific metacognitions are thought to intensify worry about having certain thoughts, leading to enhanced monitoring or searching for threatening thoughts, which are then coupled with attempts to control metacognitive processes. It is theorized that this could increase frequency and duration of rumination and worry. If this happens, the process is debilitating for coping strategies over time, as the activation of CAS may happen as a consequence. Thus, beliefs about the need for controlling thoughts as a coping strategy is likely to have paradoxical effects such as increasing awareness of thought intrusions and using up mental capacity. This could lead to a subjective experience of cognitive dysfunction, given that a level of processing capacity is preoccupied with threat monitoring and searching interoception (Jacobsen et al., 2016). Moreover, according to the metacognitive model this could affect work capacity and drive perceptions of increased work load, ultimately enhancing negative interpretations of one's ability to work effectively (Nordahl and Wells, 2017a).

The total score on the MCQ-30 at baseline did not predict RTW at baseline. However, the data showed that it was participants with a higher total score on MCQ-30 at baseline who subsequently reported the greatest change on the MCQ-30, and had higher odds of reaching the chosen success criteria within the follow-up period. This is an indication that participants with higher potential for change i.e., higher score on the MCQ-30, and who experience the largest change in metacognitions, are those who achieve RTW to a larger extent. This observation is in line with changes in the sum total of metacognitions predicting RTW. The observed data was supported by the GEE analysis showing that those achieving the greatest reduction in metacognitions during rehabilitation significantly increased in odds of RTW.

Previously, metacognitions about thoughts concerning danger and uncontrollability have been associated with work status (Nordahl and Wells, 2017b), and in the current study there was a trend indicating that a reduction in these metacognitions following treatment could influence RTW. When investigating RTW the deciding factor often lies in the chosen success criteria which is challenging when using longitudinal measures. In previous studies on metacognitions, the design has been cross-sectional and participant work status has been subjectively reported, which always gives a potential for misrepresentation and misunderstandings (Andersen et al., 2012). Future studies on work disability prevention programs should attempt to assess metacognitions as this may be relevant for most interventions. A larger sample size might have yielded a significant odds ratio in this study.

Another important point is that the participants' in this study were not selected for a particular diagnosis or diagnostic category. Rather, they reflect the Norwegian population on long-term sick leave and in need of specialized occupational rehabilitation. In this population, the rule rather than the exception is comorbidity and several mental as well as physical obstacles and symptoms. A selected group of participants with common mental disorders and only mental disorders might have yielded different results. A previous publication from our group has showed the contribution of several factors when looking at prediction and facilitation of RTW and how these arguably describe different pieces of a complex puzzle (Hara et al., 2018).

The current results generate hypotheses on which factors should be designed when targeting mental obstacles when attempting to facilitate RTW. Recently there have been systematic reviews showing that adding traditional CBT in concert with RTW programs does not increase the effect of such programs above the control condition (Salomonsson et al., 2017; Cullen et al., 2018). This could in part be due to the lack of focus on metacognitions, especially those concerned with the beliefs about the need to control thoughts. We here propose that a future trial should use a randomized controlled design to evaluate an intensive RTW rehabilitation based on the metacognitive model, alongside physical therapy and RTW coordination, comparing this to an active arm using either a traditional CBT or ACT intervention.

# Limitations

A limitation to this study is the potential selection bias given the number of participants with follow-up data. There was a software problem during data collection, and only 137 participants completed the MCQ-30 at post intervention. Missing data was treated as missing completely at random (MCAR) due to no systematic drop-out. In addition, drop-out analysis demonstrated that there was an overlap between periods of non-response with reports of software and Wi-Fi-malfunctioning from the software

developer. Therefore, non-response was assumed to be due to factors beyond the control of the participants. However, the use of registry data somewhat counteracts the low number of participants. It is also a limitation that the intervention used was not MCT, thus we cannot know whether an MCT targeted rehabilitation would be more adequate and yielded larger results. However, this was a secondary analysis of an RCT trial and the intervention was a result of the overarching trial.

# CONCLUSION

We here conclude that in participants on long-term sick leave due to chronic fatigue, pain and/or mental distress, metacognitions concerning the beliefs about the need to control thoughts appear to have a significant influence on their RTW life. Our data indicate that subtle differences in the need to control thoughts when entering rehabilitation can affect RTW. Moreover, that a reduction in the total score on MCQ-30 as well as a reduction in the need to control thoughts subscale following treatment gives significantly better odds of returning to work. We therefore recommend future studies to include these measures in RTWrehabilitation, and propose an RCT to examine the potential

# REFERENCES


effect of adding techniques from the metacognitive model to existing rehabilitation programs.

# ETHICS STATEMENT

All procedures performed in the study involving human participants were in accordance with the ethical standards of the National Research Committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Written informed consent was obtained from all individual participants included in the study. The study has been approved by the Regional Committee for Medical and Health Research Ethics in Central Norway (No. 2010/2404).

# AUTHOR CONTRIBUTIONS

HJ composed the manuscript. KH and MG performed statistical analysis. TS authored parts of the manuscript. All authors read through and commented on the manuscript.

with mental and somatic disorders: a cohort study. BMC Public Health 18:1014. doi: 10.1186/s12889-018-5803-0



**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer BL declared a shared affiliation, with no collaboration, with one of the authors, HJ, to the handling editor at time of review.

Copyright © 2020 Jacobsen, Glette, Hara and Stiles. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Innovation in Psychotherapy, Challenges, and Opportunities: An Opinion Paper

*Janina Isabel Schweiger1 , Kai G. Kahl2 , Jan Philipp Klein3 , Valerija Sipos3 and Ulrich Schweiger3 \**

*1Department of Psychiatry and Psychotherapy, Medical Faculty Mannheim, Central Institute of Mental Health, University of Heidelberg, Mannheim, Germany, 2Klinik für Psychiatrie, Sozialpsychiatrie und Psychotherapie, Medizinische Hochschule Hannover, Hannover, Germany, 3Klinik für Psychiatrie und Psychotherapie, Universität zu Lübeck, Lübeck, Germany*

# *Edited by:*

*Adrian Wells, University of Manchester, United Kingdom*

#### *Reviewed by:*

*Karin Carter, Greater Manchester Mental Health NHS Foundation Trust, United Kingdom Peter Myhr, MCT-Stockholm, Private Practice, Stockholm, Sweden*

#### *\*Correspondence:*

*Ulrich Schweiger ulrich.schweiger@uksh.de*

#### *Specialty section:*

*This article was submitted to Clinical and Health Psychology, a section of the journal Frontiers in Psychology*

*Received: 07 August 2018 Accepted: 20 February 2019 Published: 19 March 2019*

#### *Citation:*

*Schweiger JI, Kahl KG, Klein JP, Sipos V and Schweiger U (2019) Innovation in Psychotherapy, Challenges, and Opportunities: An Opinion Paper. Front. Psychol. 10:495. doi: 10.3389/fpsyg.2019.00495*

Psychotherapy as a field tends toward conservativism, and the rate of innovation and development of new evidence-based effective treatments has been slow. The paper explores important barriers to innovation like the dodo bird verdict and the habit of starting the development of therapeutic methods from techniques. The paper looks at the opportunities for translating basic science in psychology into psychotherapeutic techniques. Metacognitive therapy stands out from other psychotherapies by its development from basic science. The paper describes the development of the techniques detached mindfulness and attention training, how they were derived from basic science and tested for their suitability in the therapy of patients with anxiety disorders. By this process, metacognitive therapy may be an important model for the innovation process in psychotherapy.

Keywords: psychotherapy development, psychotherapy innovation, randomized controlled clinical trails, metacognitive therapy (MCT), scientific base of psychotherapy

# INTRODUCTION

The implementation of psychotherapy in general healthcare has been one of the significant innovations of the twentieth century and has revolutionized how the health care system deals with mental disorders. Psychotherapy is an essential focus of training in clinical psychology and physicians aiming for board certification in psychiatry or psychosomatics in many countries. Despite this transformative impact, the rate of innovation and development of new evidencebased effective treatments has been slow, and it has been noted that compared with medication psychotherapy use is on the decline in the US (Gaudiano and Miller, 2013). This opinion paper examines some of the barriers to innovation that we believe have slowed progress. It discusses alternative ways of fostering innovation and uses the development of metacognitive therapy by Wells and colleagues as an example of a strategy that overcomes barriers and discusses how MCT fits into current assumptions about innovation.

# BARRIERS

# The Therapeutic Relationship and the Dodo Bird Verdict

One of the widespread assumptions in psychotherapy is that a good therapeutic relationship is the critical mechanism of successful psychotherapeutic treatment (Wampold, 2015). It is assumed that the relationship is more significant than the underlying model of causality and the manipulation of its causal variables and is the universal change mechanism uniting all psychotherapy approaches. This way of thinking postulates that creating expectations through explanations of the disorder and the treatment involved and the enactment of healthpromoting actions are further common factors. The presumed equivalence of all therapies after correction for the therapeutic relationship has resulted in the dodo bird verdict (Luborsky et al., 2002). Based on the finding in meta-analyses that a broad spectrum of psychotherapeutic treatments in depression is similarly effective, Cuijpers has claimed that there is a possibility to minimize the number of existing therapies (Cuijpers, 1998). However, results of meta-analyses support differences between psychotherapies (Budd and Hughes, 2009; Tolin, 2010).

While the patient-rated quality of the therapeutic alliance is a good predictor of outcome in therapy (Cameron et al., 2018), a meta-analysis of the relationship between therapeutic alliance and treatment outcome in eating disorders showed that the association between alliance and outcome is weaker than the association between early symptom improvement and later alliance (Graves et al., 2017). Thus, it would seem that early symptom improvement affects the later alliance. We might presume that the most effective treatments give rise to the strongest alliances. What is lacking are experimental studies that actively manipulate therapeutic alliance, and so the evidence remains restricted to longitudinal predictor analyses that can do little more than implying causal relations (Fluckiger et al., 2018). Despite the lack of experimental evidence, the prevalent assumption is that a good working alliance is "a thing" that resides in the interpersonal harmony between two persons, providing a patient with a healing experience that appears to be part of a stable, benign relationship. Related to this idea is the presupposition that some therapists "have it" while others do not, meaning that there are good and bad therapists, as categories. Unfortunately, this explanation falls short of the alternative but little-tested assumption that a good therapeutic relationship is an emergent phenomenon produced by professionalism, plausible models, and experience of change already early in therapy.

Consistent with the assumption that the alliance is, in fact, an emergent factor of effective therapy, the working alliance in pure Internet therapy is remarkably good (Heim et al., 2018). The continued perception of the therapeutic relationship as the primary underlying factor of psychotherapy effectiveness is a barrier because it reduces the necessity of developing innovative theories and techniques since new techniques only make a marginal difference. Assigning the therapeutic relationship to the role of the critical cause of change, instead of modeling it as an emergent phenomenon of change creates inertia in research on psychopathological mechanisms and complacency in therapists.

# Starting the Development of Therapeutic Methods From Techniques

New approaches have most often been devised based on techniques, that is on the basis of assembling combinations of treatment techniques that appear to work. Such approaches are often only loosely grounded in theoretical models, and the models of treatment mechanisms may develop after the treatments themselves.

A top-down approach in the design of technology starts with an overview of the relevant system (e.g., dysfunctional beliefs) but does not specify subsystems in sufficient detail or elucidate how they impact on functioning. For instance, negative automatic thoughts and beliefs are purported to cause or maintain disorder in the cognitive model. However, as pointed out by Wells and Matthews (Wells and Matthews, 1996), this approach does not consider broader aspects of cognition that are known to be associated with the disorder such as biases in the regulation of attention and levels of control of cognition. The cognitive-behavioral model has not advanced along with recent developments in cognitive psychology and theory such that the practice of therapy is only loosely tied to an understanding of mechanisms. Beck based CBT on the description of problematic thought content and processes of cognitive distortion in patients (Beck, 1963, 1964). The primary intervention derived from this observational approach and comprised of correcting cognitive distortions and deficiencies in schema content using Socratic dialogue. This fundamental change technique of cognitive therapy (CT) is derived from philosophy and is not rooted in or supported by experimental psychology. To the contrary, research shows that trying to replace dysfunctional thought by more appropriate thinking may result in thought suppression and have adverse paradoxical effects (Longmore and Worrell, 2007; Magee et al., 2012). Subsequently, more techniques used initially in behavioral activation, assertiveness training, anxiety management or mindfulness meditation have been incorporated to form a more eclectic cognitive behavioral therapy (CBT).

A second notable example of technique-driven development is dialectical behavior therapy (DBT). It is based on the assumption that patients with borderline personality disorder have skills deficits in emotion regulation (Linehan et al., 1991; Linehan, 2014). At the core of the interventions are approximately 50 skills that are taught to patients to improve emotion regulation. Again, learning theory informed the selection of these skills, but none was derived from experimental psychology nor were they individually tested. As packages, both CBT (Beck and Dozois, 2011) and DBT (Stoffers et al., 2012) can be considered as well supported by evidence. There were a few studies involving component analysis (Jacobson et al., 1996) showing that in the case of CBT challenging thoughts on the content level, the primary and elemental technique may not be the essential ingredient. The introduction of disorderspecific treatment methods for depression, anxiety disorders, and personality disorders beginning in the 1960s was a big step forward for psychotherapy. These new methods led to a considerable extension of the field of activities of psychotherapy toward groups that are severely ill and were traditionally underserved.

While there is evidence that these treatments offer innovation and can work, it is important to question whether the techniquedriven approach of combining a range of techniques is the most effective means of treatment development. In particular, multi-component and highly eclectic treatment packages may hide detrimental effects of specific components of a treatment method (Castonguay et al., 1996). In summary, these examples show that in psychotherapy, the dominant technique-driven approach (as in other fields) has advantages but also creates serious problems.

# OPPORTUNITIES

# Starting From Basic Science

All methods of modern behavior therapy refer to general learning theory (behaviorism, cognitivism, constructivism, and social cognitive theory) or information processing theory. Only two refer to a specific psychological theory derived from general psychology: metacognitive therapy (MCT) (Wells, 2009) draws on and develops the concept of metacognition as described by Flavell (Flavell, 1979). It is grounded in the self-regulatory executive function (S-REF) model, a detailed information processing model of human cognitive and affective regulation (Wells and Matthews, 1996). Acceptance and commitment therapy (ACT) refers to relational frame theory (RFT) (Hayes et al., 2001). The exact nature of the interaction between RFT and the techniques proposed by ACT is an ongoing point of discussion. For specific information, see (Zettle et al., 2016).

An essential aspect of starting from basic science is to direct therapeutic techniques at psychological mechanisms or processes and not at mental disorders which are broad concepts summarizing symptom clusters. Focusing on a specific mechanism necessarily results in a reductionistic approach. For example, MCT assumes that worry, rumination, and threat monitoring are part of a cognitive attentional syndrome (CAS) which is a core psychological process and a transdiagnostic factor across most disorders. Putting worry, rumination, and attention to threat in the center implies that psychological dysfunction such as anxiety is a product of this mechanism, and there is no need to directly address the emotion anxiety if a technique can limit the CAS.

MCT seems to be quite unusual as it exclusively developed and uses techniques that can be directly related to the parent theory, and it was developed by systematically testing the assumptions derived from this theory. MCT started with case studies demonstrating the effects of manipulating attention focus, through the attention training technique (ATT) as a means of enhancing cognitive control and disrupting the CAS (Wells, 1990), and later on the effects of attention enhancements on exposure (Wells and Papageorgiou, 1998). There is now a significant database supporting the probable efficacy of ATT (Knowles et al., 2016; Fergus and Wheless, 2018) and full MCT (Normann et al., 2014). While there are a variety of techniques intended to modify attentional focus and attentional processes in behavior therapy literature, these were often focused on reducing anxiety through distraction, rather than based on a theory linking attention to psychological causal or maintenance mechanisms. An exception is presented by work in the area of attention bias modification (ABM) based on the finding that anxiety is associated with "automatic processing" of threat-related information and in principle, such bias might be retrained (MacLeod, 2015). However, these examples of ATT and ABM appear to be among the few exceptions in the field.

# Theory-Driven Construction of Psychotherapeutic Methods

In the case of MCT and of its individual techniques such as ATT, we see a paradigmatic shift with a predominant theorydriven development of therapeutic techniques. Furthermore, the theory is firmly grounded in objective psychological science of attention (Wells and Matthews, 1996). However, we need an awareness of the potential risks involved in this system of therapy development, and we require an ongoing process of refining psychotherapy from a basic science perspective. Helpful tools may be qualitative studies examining the effects of specific psychotherapeutic techniques, and single case studies that focus on testing-isolated techniques. Essential principles of psychotherapy like "doing a few things well" or "less is more" (low complexity results in better skill acquisition, focus on key information results in better decisions) may show their advantages in further enhancing the theory-driven approach to therapy development.

Starting the construction of psychotherapeutic methods from basic science is an exception rather than a rule. However, this is not related to a lack of progress in general psychology. Actually, there is a substantial amount of new knowledge in the field with obvious relevance that awaits translation into psychotherapy techniques, e.g., knowledge about decision making (Morewedge and Kahneman, 2010), human cooperation (Rand and Nowak, 2013), heuristics (Raab and Gigerenzer, 2015), or the theory of constructed emotions (Barrett, 2017). The development of MCT presents an example of a systematic approach to theory and testing that could be emulated in developing the full potential of other psychological discoveries.

# CONCLUSION

Our opinion paper points to the necessity of rethinking innovation processes in psychotherapy. Psychotherapy is a significant achievement in modern health care. It needs further evolution. To this end, it still needs to overcome barriers and might benefit from a more rigorous theory-driven approach that is informed by discoveries in psychological science. Metacognitive therapy is an example of this type of approach in which an interface between cognitive psychology and applied psychology has been developed and exploited with good effect.

# AUTHOR'S NOTE

The idea for this opinion paper arose during a lengthy discussion after a presentation by Adrian Wells at the World Congress of Psychiatry in Berlin in October 2017.

# REFERENCES


# AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

# ACKNOWLEDGMENTS

We thank Prof. Dr. Heike Tost for her critical comments to the manuscript.


Wells, A. (2009). *Metacognitive therapy for anxiety and depression*. New York: Guilford.

Wells, A., and Matthews, G. (1996). Modelling cognition in emotional disorder: the S-REF model. *Behav. Res. Ther.* 34, 881–888.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2019 Schweiger, Kahl, Klein, Sipos and Schweiger. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*