# ADAPTIVE GAIT AND POSTURAL CONTROL: FROM PHYSIOLOGICAL TO PATHOLOGICAL MECHANISMS, TOWARDS PREVENTION AND REHABILITATION

EDITED BY : Helena Blumen, Paolo Cavallari, France Mourey and Eric Yiou PUBLISHED IN : Frontiers in Aging Neuroscience, Frontiers in Neurology, Frontiers in Medicine and Frontiers in Bioengineering and Biotechnology

#### 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 978-2-88963-626-6 DOI 10.3389/978-2-88963-626-6

#### 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

# ADAPTIVE GAIT AND POSTURAL CONTROL: FROM PHYSIOLOGICAL TO PATHOLOGICAL MECHANISMS, TOWARDS PREVENTION AND REHABILITATION

Topic Editors: Helena Blumen, Albert Einstein College of Medicine, United States Paolo Cavallari, University of Milan, Italy France Mourey, Université de Bourgogne, France Eric Yiou, Université Paris-Saclay, France

Citation: Blumen, H., Cavallari, P., Mourey, F., Yiou, E., eds. (2020). Adaptive Gait and Postural Control: From Physiological to Pathological Mechanisms, Towards Prevention and Rehabilitation. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-626-6

# Table of Contents


Jeannette R. Mahoney and Joe Verghese

*42 Effects of Aging on Postural Responses to Visual Perturbations During Fast Pointing*

Yajie Zhang, Eli Brenner, Jacques Duysens, Sabine Verschueren and Jeroen B. J. Smeets


*149 Walking Along Curved Trajectories. Changes With Age and Parkinson's Disease. Hints to Rehabilitation*

Marco Godi, Marica Giardini and Marco Schieppati

*160 Long-Term Effects of Whole-Body Vibration on Human Gait: A Systematic Review and Meta-Analysis*

Matthieu Fischer, Thomas Vialleron, Guillaume Laffaye, Paul Fourcade, Tarek Hussein, Laurence Chèze, Paul-André Deleu, Jean-Louis Honeine, Eric Yiou and Arnaud Delafontaine


Katherine M. Martinez, Mark W. Rogers, Mary T. Blackinton, M. Samuel Cheng and Marie-Laure Mille

*214 Effects of Virtual Reality-Based Physical and Cognitive Training on Executive Function and Dual-Task Gait Performance in Older Adults With Mild Cognitive Impairment: A Randomized Control Trial* Ying-Yi Liao, I-Hsuan Chen, Yi-Jia Lin, Yue Chen and Wei-Chun Hsu

*224 Motor-Cognitive Neural Network Communication Underlies Walking Speed in Community-Dwelling Older Adults* Victoria N. Poole, On-Yee Lo, Thomas Wooten, Ikechukwu Iloputaife, Lewis A. Lipsitz and Michael Esterman

*232 Age-Related Adaptations of Lower Limb Intersegmental Coordination During Walking*

Mathieu Gueugnon, Paul J. Stapley, Anais Gouteron, Cécile Lecland, Claire Morisset, Jean-Marie Casillas, Paul Ornetti and Davy Laroche


Luigi Tesio and Viviana Rota

*286 Motor Adaptation in Parkinson's Disease During Prolonged Walking in Response to Corrective Acoustic Messages* Mattia Corzani, Alberto Ferrari, Pieter Ginis, Alice Nieuwboer and Lorenzo Chiari

# *298 Acute Effects of Whole-Body Vibration on the Postural Organization of Gait Initiation in Young Adults and Elderly: A Randomized Sham Intervention Study*

Arnaud Delafontaine, Thomas Vialleron, Matthieu Fischer, Guillaume Laffaye, Laurence Chèze, Romain Artico, François Genêt, Paul Christian Fourcade and Eric Yiou


Chae-gil Lim

# Editorial: Adaptive Gait and Postural Control: from Physiological to Pathological Mechanisms, Towards Prevention and Rehabilitation

Helena M. Blumen<sup>1</sup> , Paolo Cavallari <sup>2</sup> , France Mourey <sup>3</sup> and Eric Yiou4,5 \*

<sup>1</sup> Departments of Medicine and Neurology, Albert Einstein College of Medicine, The Bronx, NY, United States, <sup>2</sup> Human Physiology Section of the DePT, Università degli Studi, Milan, Italy, <sup>3</sup> Laboratory "Cognition, Action et Plasticité Sensorimotrice," INSERM U1093, Burgundy University, Dijon, France, <sup>4</sup> CIAMS, Université Paris-Saclay, Orsay, France, <sup>5</sup> CIAMS, Université d'Orléans, Orléans, France

Keywords: posture, gait, aging, prevention, rehabilitation

**Editorial on the Research Topic**

**Adaptive Gait and Postural Control: from Physiological to Pathological Mechanisms, Towards Prevention and Rehabilitation**

# INTRODUCTION

Gait and postural control are affected by aging, and in neurological, and musculoskeletal disorders. Poor gait and postural control are associated with disability, falls, increased morbidity and mortality; therefore representing major public health issues. The aims of this Research Topic was two-fold. First, it aimed to promote a better understanding of the patho-psychophysiological mechanisms affecting posture and gait in normal and pathological aging. Second, it aimed to provide an up-to date picture of motor and cognitive interventions directed to restore posture and gait in different aging populations. This Research Topic includes 29 contributions (16 original articles, 2 reviews, 3 systematic reviews, 5 clinical trials, 1 perspective, 1 methods paper, and 1 brief report) which, as a whole, report investigations related to posture and gait in several different populations, including healthy young and older adults, individuals that fall or have preclinical stages of dementia, and patients with stroke or Parkinson's disease—through the use of multidisciplinary concepts and techniques including Biomechanics (Neuro)physiology, Neuroimaging and Psychology. These contributions were subdivided into three key sections: (1) posture and gait changes during normal and pathological aging; (2) motor and cognitive preventive/rehabilitative interventions to restore posture and gait; (3) evaluation of posture and gait in normal and pathological aging.

Edited and reviewed by: Thomas Wisniewski, New York University, United States

> \*Correspondence: Eric Yiou eric.yiou@u-psud.fr

Received: 01 February 2020 Accepted: 10 February 2020 Published: 25 February 2020

#### Citation:

Blumen HM, Cavallari P, Mourey F and Yiou E (2020) Editorial: Adaptive Gait and Postural Control: from Physiological to Pathological Mechanisms, Towards Prevention and Rehabilitation. Front. Aging Neurosci. 12:45. doi: 10.3389/fnagi.2020.00045

# POSTURE AND GAIT CHANGES DURING NORMAL AND PATHOLOGICAL AGING

# Changes in Sensory-Motor Integration and Postural Control During Normal and Pathological Aging

The ability to combine information across sensory modalities is an integral aspect of mobility, and has been examined by analyzing the role/use of proprioceptive and visual information. Wholebody vibration (WBV) is a training method used by clinicians to improve specific motor outcomes in various populations, such as old and young adults. In this regard, Delafontaine et al. showed that WBV applied prior to gait initiation increases stance leg stiffness, an effect known to be detrimental to stability. This negative effect, however, was compensated for by an increase in the amplitude of "anticipatory postural adjustments," resulting in improved stability. These changes were observed in young adults, but not old adults. These findings are consistent with the hypothesis that balance control mechanisms within the postural system are interdependent (i.e., they may compensate each other if one component is perturbed) and vary with age.

Asymmetrical sensory-motor function after stroke creates unique challenges for bipedal tasks such as walking or perturbation-induced reactive stepping. Martinez et al. examined perturbation-induced stepping between legs, in stroke survivors and non-impaired controls. Participants were given an anterior perturbation from three stance positions: symmetrical distribution of the body weight or asymmetrical distribution on the preferred and the non-preferred leg. They found that stepping with the more-involved leg can be facilitated by un-weighting the leg. Post-stroke participants had earlier anticipatory postural adjustments, and always took more steps than controls to regain balance. These differences point to the simultaneous bipedal role (support and stepping) both legs have in reactive stepping.

When it comes to visual information, people can quickly adjust goal-directed hand movements to an unexpected perturbation, such as a target jump, or background motion. Zhang et al. examined whether this ability correlates with age. Although young and old adults were equally accurate, reaction time, and movement times were longer in old adults, and the responses were less vigorous. Moreover, old adults responded more strongly to the motion of the background than to the target jump. Thus, old adults showed delayed responses to visual perturbations and relied more on visual surroundings to adjust goal-directed movements. Given the observed overlap in neural circuitry necessary for both multisensory integration and goal-directed locomotion, it has also been shown that visual-somatosensory integration is significantly associated with gait pace but not rhythm, which is a more automatic process controlled mainly through brainstem and spinal networks. Actually, Mahoney and Verghese demonstrated that in old people stride length variability, but not swing time, is associated with magnitude of visual-somatosensory integration, concluding that worse visual-somatosensory integration in aging is associated with worse spatial, but not temporal, components of gait.

Lower-limb intersegmental coordination is a complex component of human walking, and aging may result in impairment of motor control and coordination contributing to the decline in mobility, and inducing loss of autonomy. In this regard, two reviews, by Tesio and Rota and by Delafontaine et al. were devoted to the analysis of center of mass (CoM) during walking and to anticipatory postural adjustments during gait initiation in stroke patients. Tesio and Rota state that the trajectory of CoM is a promising summary index of both balance and the neural maturation of walking. In fact, alterations in CoM motion could reveal motor impairments that are not detectable by clinical observation. In asymmetric gaits, the affected lower limb avoids muscle work by oscillating almost passively, but extra work is required from the unaffected side and the average work across a stride remains normal. In more demanding conditions, the affected limb can provide more work; however, the unaffected limb does also, and asymmetry between the steps persists. This "learned" asymmetry is a formerly unsuspected challenge to rehabilitation attempts to restore symmetry. The study of the three-dimensional trajectory of the CoM motion, which assumes a figure-of-eight shape, also represents a clinical frontier. Shape and size of the "figure-of-eight" seems not to change from child to adulthood. Delafontaine et al. make clear that stroke patients undergo a decrease in the tibialis anterior activity associated with difficulties in silencing soleus muscle of the paretic leg, display a decreased shift of center of pressure, and have lower propulsive anterior forces and suffer of a longer preparatory phase. Regarding possible gait-rehabilitation strategies, this systematic review suggests that the use of the non-paretic as the leading leg can be a useful exercise to stimulate the paretic postural muscles. Moreover, Gueugnon et al. evaluated the impact of aging on the coordination of lower limb kinematics and kinetics during walking at a comfortable speed. They showed that lower-limb coordination modifies with age, thus influencing ankle motion and power. They also hypothesized that this modification of coordination constitutes a neuromuscular adaptation of gait control in order to maintain gait efficiency. Foot-shank coordination might thus represent a valid outcome measure to estimate the efficacy of rehabilitative strategies and to evaluate their efficiency in restoring lower-limb synergies during walking.

Tripping over an obstacle may end with a fall, but a fall can also be caused by a spontaneous loss of balance, which are common during normal, and pathological aging. Thus, fall prevention is an important topic. A study by Jayakody et al. examined changes in gait variability—a known predictor of falls—factors (medical, sensory-motor, cognitive, and demographic) that may predict this variability across time (close to 5 years). They found that variability in different gait measures do not undergo uniform changes. Furthermore, each variability measure was associated with different factors such as presence of cardiovascular disease, arthritis and body mass index, quadriceps strength, postural sway, processing speed, and lower education. These results provide useful information on potential targets for future trials to maintain mobility and independence in aging. The risk of falling, and the changes in balance control strategy, was further analyzed by Li et al. More specifically, they compared the differences in the kinematic characteristics of crossing obstacles of different heights between stroke survivors and age-matched healthy controls. The authors suggested that, because of reduced muscle strength, stroke survivors use a conservative strategy to negotiate the obstacles and control balance. As a result, abnormal patterns during obstacle crossing increase the risk of falling. They propose that future rehabilitation training program should be designed to enhance body stability, reduce energy cost, and improve motion efficiency.

A better understanding of the pathophysiology of steering, which requires an interaction between posture, balance and progression along non-linear trajectories, is reviewed by Godi et al. Their review focused on the fundamental processes (physical laws, changes in muscle/brain activation, role of proprioception) allowing for changes in locomotion trajectory along a curved path in old adults, and people with Parkinson's Disease (PD). They also propose rehabilitative approaches to improve locomotion in order to reduce the risk of falling in both healthy old adults and patients. Finally, Lu et al. evaluated the timing of avoidance of a virtual obstacle in old patients with PD in the off-medication state and in young and old controls, while walking on a treadmill. They found that the probability of successful obstacle avoidance was associated with the timing of obstacle appearance, and that age was positively correlated with the time required to successfully avoid obstacles. Nonetheless, the PD group required significantly more time. They concluded that slowing of gait adaptability could contribute to high fall risk in old adults and PD because of disturbances in motor planning, movement execution, or disordered response inhibition.

# Relationship Between Psycho-Cognitive Conditions and Motor Control of Gait in Normal and Pathological Aging

Walking is a highly automated process, but as we age the demands for controlled processes increases. Walking is also influenced by psychological factors such as fear of falling (FOF). Several articles in this topic focuses on psychocognitive conditions and control of gait during normal and pathological aging.

Dual-tasking (DT; walking while performing a secondary task) paradigms are often used to examine the cognitive demands of walking, as it involves attention switching and task prioritization. However, studies using this paradigm to explain age-related performance decline sometimes reveal contradictory results. To clarify this point, Wollesen and Voelcker-Rehage investigated whether differences in demographics, physical functioning, FOF, and other psychological factors could explain reduced DT performance. Their results confirm a large interindividual variability in old adults, suggesting that factors causing performance differences in DT costs need to be reassessed. Specifically, functional age may be a better predictor of DT costs than chronological age, and psychological factors have a negative effect on DT performance.

The neural correlate of DT costs in old adults was investigated by Poole et al. who showed that faster walking was associated with increased functional connectivity between motor and cognitive networks, and decreased functional connectivity between limbic and cognitive networks. Moreover, smaller DT costs was associated with increased connectivity within the motor network, and increased connectivity between ventral attention and executive networks. These findings support the importance of both motor network integrity as well as internetwork connectivity amongst higher-level cognitive networks in healthy old adults' ability to maintain mobility, particularly under DT conditions. Hermand et al. focused on examining prefrontal cortex (PFC) activity of subacute stroke patients in DT situations. Their results showed a "ceiling" effect in PFC oxyhemoglobin levels while walking, leaving no available resources for simultaneous cognitive tasks during the early recovery period following stroke. In these patients, cognitive, but not motor performance declined with a higher cognitive load, i.e., prioritization was given to maintain walking performance. Interestingly, a similar finding was reported in healthy old adults by Vervoort et al. who examined the effects of aging and task prioritization on split-belt gait adaptation with or without a cognitive load. They showed that young and old adults modify, respectively, the perturbed leg's timing and gait speed to adapt to split-belt gait perturbations. Results also showed a decline in cognitive task performance during early gait adaptation, which suggests task prioritization, especially in older adults. They concluded that healthy old adults retain the ability to adapt gait to split-belt perturbations, but through a different adaptation strategy.

The ability to adapt gait to environmental constraints may also change in PD, as shown in the study of Caetano et al. These authors investigated cognitive, physical, and psychological factors associated with gait adaptation to obstacles and stepping target negotiation. They showed that executive function and reactive balance capacity are important for precise foot placements; and that cognitive capacity is important for step length adjustments associated with obstacle avoidance. The role of psychological factors on gait is also stressed by Santos Bueno et al., who investigate gait patterns in healthy old women with and without fall-history, and with high and low FOF. They found that after exposure to a fictional disturbing factor (psychological and non-motor agent), all participants had shorter step length, stride length, and slower walking speed. Those without fall-history and with high FOF, however, showed greater changes and lower Gait Profile Score. They concluded that these gait changes in the presence of a FOF causing agent led to a cautious gait pattern in an attempt to increase protection.

# MOTOR AND COGNITIVE PREVENTIVE/REHABILITATIVE INTERVENTIONS TO RESTORE POSTURE AND GAIT

A better understanding of the pathophysiological mechanisms associated with gait and posture can ultimately be used to develop successful interventions for maintaining gait and posture in healthy and pathological aging. This Research Topic incorporates articles describing interventions that aim to maintain gait performance in healthy old adults, in individuals with chronic and subacute stroke, and in old adults with mild cognitive impairment (MCI).

In a study of healthy old adults, Eckardt and Rosenblatt investigated how different resistance training modalities affect kinematic synergies related to the medio-lateral trajectory of the swing-foot during normal and perturbed gait. Researchers found that training using unstable conditions had a significant effect on kinematic synergy. They concluded that unstable training conditions promote neuromuscular adaptations and should be incorporated into training programs.

The relative efficacy of 4 weeks of underwater gait training with water jet resistance and underwater gait training with ankle weights was explored in 22 middle-aged to old adults with chronic stroke by Lim. Results indicated that underwater gait training with water jet resistance led to greater improvements in dynamic balance than underwater gait training with ankle weight, as well as a number quantitative gait measures (e.g., gait velocity, cadence, and step length). In another pilot RCT of 37 individuals with subacute stroke, Lim examined the relative efficacy of multi-sensory training and treadmill gait training on proprioception and balance. After 8 weeks of training, conventional rehabilitation, and multi-sensorimotor training generated more improvements than conventional rehabilitation and treadmill gait training on both proprioception and balance encouraging results for designing a future large-scale RCT. Another contribution to this topic (Giannouli et al.) presents a conceptual framework that is ripe for exploring the efficacy of a novel square-stepping program for falls prevention in older adults.

The benefits of a suite of computer games that involves physical exercise and demands different cognitive functions was explored in old adults by de Bruin et al. Quantifying EMG coherence and gait during single and dual-task walking conditions, they found that 6 weeks of exercise improved corticospinal transmission to the tibialis anterior muscle, and increased minimum toe clearance during dual-task walking conditions. These preliminary findings are encouraging, but need to be more systematically examined in a randomized controlled trial (RCT). The relative efficacy of virtual-reality based and traditional physical and cognitive training was explored in an RCT of 34 older adults with MCI by Liao et al. Results suggest that virtual-reality based cognitive and physical training leads to greater improvements in dual-task gait performance (cadence) and a measure of executive function (trail making test) in MCI. Others (Pedroli et al.) propose a virtual-reality based motor rehabilitation program for falls prevention in frailty that takes place in a CAVE system (4-screen room with a stationary bike) and focuses on improving balance.

A final contribution to this section is a systematic review by Fischer et al. of WBV training for balance control in healthy and pathological participants over the lifespan (16 studies evaluated the effects in old adults). They concluded that there was a positive effect of WBV on gait speed and the timed-up-and-go test in old adults, but a consensus on specific WBV training was not reached.

# EVALUATION OF POSTURE AND GAIT IN NORMAL AND PATHOLOGICAL AGING

A better understanding of the pathophysiological mechanisms associated with gait and posture is also dependent on appropriate measures and characterization of at-risk individuals. In this Research Topic, several articles address these two related issues. Mandigout et al. compared step counts of wrist-worn and hipwork actigraphs in 22 young and 22 old adults over a 24-h period. Their key finding was that step counts are greater from wrist-worn actigraphs than hip-worn actigraphs in both young and old adults. In a systematic review of young adults, old adults and PD patients, Siragy and Nantel found that quantifying dynamic balance can be challenging due to varying measures and definitions, reflects neuromuscular stability mechanisms, and depends on whether walking conditions are perturbed or unperturbed. Thus, caution is recommended when comparing step counts from different actigraphs and different measures of dynamic balance across different studies.

Caution is also recommended when characterizing old adults with the motoric cognitive risk (MCR) syndrome. The motoric cognitive risk syndrome is a preclinical stage of dementia that is typically characterized as slow gait speed (for their age and sex) and a subjective cognitive complaint (Verghese et al., 2013). In this Research Topic, Sekhon et al. examined if increased fivetimes-sit-to-stand time rather than slow gait speed could be used to characterize individual with MCR. They found that the prevalence of this new characterization of MCR is lower than the typical characterization of MCR and that it does not predict non-amnestic MCI—thus, may not be tapping in to the same preclinical stage of dementia.

Finally, Corzani et al. characterized motor adaptation in response to acoustic stimuli during walking in individuals with PD. In order to improve gait rhythmicity their interdisciplinary team developed wearable technologies to stimulate gait rhythm and increase or decrease cadence. Their results suggest that verbal cuing is more effective at restoring normal cadence than metronome beats.

# AUTHOR CONTRIBUTIONS

HB, PC, FM, and EY drafted, revised, and approved the final version of manuscript.

# REFERENCES

Verghese, J., Wang, C., Lipton, R. B., Holtzer, R. (2013). Motoric cognitive risk syndrome and the risk of dementia. J. Gerontol. A Biol. Sci. Med. Sci. 68, 412–418. doi: 10.1093/gerona/gls191

**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 Blumen, Cavallari, Mourey and Yiou. 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.

# Quantifying Dynamic Balance in Young, Elderly and Parkinson's Individuals: A Systematic Review

Tarique Siragy and Julie Nantel\*

School of Human Kinetics, University of Ottawa, Ottawa, ON, Canada

Introduction: Falling is one of the primary concerns for people with Parkinson's Disease and occurs predominately during dynamic movements, such as walking. Several methods have been proposed to quantify dynamic balance and to assess fall risk. However, no consensus has been reached concerning which method is most appropriate for examining walking balance during unperturbed and perturbed conditions, particularly in Parkinson's Disease individuals. Therefore, this systematic review aimed to assess the current literature on quantifying dynamic balance in healthy young, elderly and Parkinson's individuals during unperturbed and perturbed walking.

Methods: The PubMed database was searched by title and abstract for publications quantifying dynamic balance during unperturbed and mechanically perturbed walking conditions in elderly adults and PD. Inclusion criteria required publications to be published in English, be available in full-text, and implement a dynamic balance quantification method. Exclusion criteria included clinical dynamic balance measures, non-mechanical perturbations, pathologies other than PD, and dual-tasking conditions. The initial database search yielded 280 articles, however, only 81 articles were included after title, abstract and full-text screening. Methodological quality and data were extracted from publications included in the final synthesis.

Results: The dynamic balance articles included 26 Coefficient of Variation of Spatiotemporal Variability, 10 Detrended Fluctuation Analysis, 20 Lyapunov Exponent, 7 Maximum Floquet Multipliers, 17 Extrapolated Center of Mass, 11 Harmonic Ratios, 4 Center of Mass-Center of Pressure Separation, 2 Gait Stability Ratio, 1 Entropy, 3 Spatiotemporal Variables, 2 Center of Gravity and Center of Pressure, and 2 Root Mean Square in the final synthesis. Assessment of methodological quality determined that 58 articles had a low methodological rating, a 22 moderate rating, and 1 having a high rating.

Conclusion: Careful consideration must be given when selecting a method to quantify dynamic balance because each method defines balance differently, reflects a unique aspect of neuromuscular stability mechanisms, and is dependent on the walking condition (unperturbed vs. perturbed). Therefore, each method provides distinct information into stability impairment in elderly and PD individuals.

Keywords: dynamic balance, Parkinson's disease, falls, walking stability, perturbations

#### Edited by:

France Mourey, Université de Bourgogne, France

#### Reviewed by:

Alexandre Kubicki, Université de Bourgogne, France Bianca Callegari, Universidade Federal do Pará, Brazil Jeannine Bergmann, Schön Klinik, Germany

> \*Correspondence: Julie Nantel jnantel@uottawa.ca

Received: 31 July 2018 Accepted: 05 November 2018 Published: 22 November 2018

#### Citation:

Siragy T and Nantel J (2018) Quantifying Dynamic Balance in Young, Elderly and Parkinson's Individuals: A Systematic Review. Front. Aging Neurosci. 10:387. doi: 10.3389/fnagi.2018.00387

# INTRODUCTION

# Parkinson's Epidemiology

Parkinson's Disease (PD) is the second most common neurodegenerative disease in the world, with an increasing incidence within elderly individuals over the age of 60 (De Lau and Breteler, 2006; Hausdorff, 2009). It is well-established that people with PD are at an increased risk of falling due to gait instability with fall rates are as high as 70% (Hausdorff et al., 1998; Plotnik et al., 2007, 2008; Plotnik and Hausdorff, 2008; Hausdorff, 2009; Kerr et al., 2010; Nantel et al., 2011). Additionally, previous reports indicate that individuals with a fall history have an increased likelihood for subsequent falls thereby creating a self-perpetuating cycle (Hamacher et al., 2011; Francis et al., 2015). Therefore, accurately identifying individuals with unstable gait is crucial to determining fall risk as well as to assessing therapeutic intervention effectiveness for PD individuals.

# PD Balance and Gait Issues

Parkinson's disease is a progressive neurodegenerative disease that causes loss of dopaminergic neurons that begins in the substantia nigra pars compacta propagating further into additional structures of the basal ganglia (Poewe et al., 2017). The basal ganglia is composed of several midbrain structures that rely on dopamine as a vital neurotransmitter to regulate movement (Blandini et al., 2000; Poewe et al., 2017). Dopamine reduction in PD causes impairment to the basal ganglia's function that results in PD's cardinal symptoms (bradykinesia, rigidity and tremor), which ultimately result in a multitude of motor impairments (Blandini et al., 2000; Poewe et al., 2017). Amongst these impairments is the disrupted gait pattern commonly observed in PD individuals. Indeed, previous research demonstrated that PD individuals ambulate with a reduced velocity, shorter stride lengths, increased double support time, reduced cadence and reduced interlimb coordination (Hausdorff et al., 2010).

Furthermore, parkinsonian gait is characterized by an increase in spatiotemporal variability that progressively worsens during the course of the disease (Hausdorff et al., 1998, 2010; Bloem et al., 2004). This has been attributed to the impaired basal ganglia in generating internal cues for rhythmic motor outputs (Hausdorff et al., 1998). Due its severity and implications for ensuing fall risk, increased spatiotemporal variability is considered a hallmark feature of Parkinsonian gait (Hausdorff et al., 2003; Plotnik and Hausdorff, 2008; Hausdorff, 2009). In addition, walking stability in PD is further threatened due to the paroxysmal phenomenon known as Freezing of Gait (FOG) (Nanhoe-Mahabier et al., 2011, 2013). Freezers, exhibit increased gait asymmetry and spatiotemporal variability, as well as reduced interlimb coordination when compared to non-freezers, even during optimally medicated states (Nieuwboer et al., 2001; Hausdorff et al., 2003; Bloem et al., 2004; Nantel et al., 2011).

# Quantifying Dynamic Balance

Dynamic Balance during steady-state gait is defined as the ability to stabilize an individual's COM within a series of alternating unilateral stances. Typically, clinicians utilize motor performance tests (e.g., Berg scale, Timed Up and Go, POMA, etc.) to assess dynamic balance and patient fall risk. However, previous research demonstrated several limitations in their ability to fully assess dynamic stability and predict fall likelihood in both healthy and clinical populations (Bhatt et al., 2011; Hubble et al., 2015). Therefore, several quantitative dynamic balance measures have been proposed as alternatives to assess fall risk (Hausdorff, 2009). However, despite these developments, ambiguity exists concerning which method is more suited to quantify dynamic balance for a particular demographic during both unperturbed and perturbed walking. This is primarily due to each test reflecting different properties of the neuromuscular system necessary for successful walking (Hausdorff, 2009).

With the widespan of dynamic balance measures, there is an apparent lack of uniformity in regards to which measure is most suitable for assessing dynamic stability, in both PD individuals and healthy adults, and in which specific scenario (perturbed or unperturbed walking; Hubble et al., 2015). This ambiguity is further perpetuated when one considers the various environmental conditions that may result in external perturbations.

# Quantifying Dynamic Balance Recovery From Mechanical Perturbations

As mechanical perturbations (trips, slips, and surface conditions) can also disrupt an individual's balance, previous researchers suggested that dynamic balance measures must adequately evaluate the perturbations' destabilizing effects and their recovery therefrom. Current evidence on mechanical perturbations demonstrate that the neuromuscular system employs active recovery strategies to return the perturbed COM to a dynamic stability state (Marigold, 2002; Marigold and Misiaszek, 2009). Indeed, Marigold and Patla (2002) suggested that the neuromuscular system's ability to adapt to multiple balance conditions is rudimentary to maintain stability during walking (Marigold and Patla, 2002). However, during the course of natural aging, the ability to execute these recovery strategies in a timely manner becomes impaired. As PD is a neurodegenerative disease that predominately affects elderly individuals, this demographic is at increased risk to external perturbations debilitating effects. By quantifying an individual's dynamic stability level before and after perturbations, researchers can assess the effectiveness, or the lack thereof, of implemented recovery strategies. To accomplish this, it is necessary to determine the advantages and limitations of current quantitative dynamic stability methods in assessing balance and balance recovery from external mechanical perturbations.

# Systematic Review Purpose

Currently, there are a few literature reviews that examine and compare the various methods for quantifying dynamic stability (Hamacher et al., 2011; Bruijn et al., 2013). Existing literature reviews are limited in that they either only examine kinematic measures or do not conduct the review systematically. As such, the aims of this systematic literature review are (1) to examine the various methods of quantifying dynamic stability in PD individuals and healthy controls during steady-state walking and (2) to examine dynamic stability measures in recovering to a state of dynamic stability from external mechanical perturbations.

# METHODS

# Identification of Materials

This literature review was conducted in accordance with the PRISMA 2015 guidelines and protocol (Moher et al., 2009). The PubMed and Medline electronic databases were searched for publications assessing dynamic balance in healthy young, healthy older and people with Parkinson's Disease. The key search terms used included the following combinations:

	- AND

Furthermore, reference lists of articles were scanned for any additionally relevant articles. If relevant articles were found, the PubMed and Medline databases were searched for full-text accessibility for inclusion in the article screening. If an article was not full-text accessible from the databases, the authors were not contacted. Articles searched were from 1990 to 2017.

# Inclusion/Exclusion Criteria

After removing duplicates, potential articles that were identified by the database search were independently screened by two researchers for relevancy. Discrepancies between screeners were resolved through discussion and comparison. Articles meeting the following inclusion criteria proceeded to data extraction:


Article exclusion criteria were as follows:


# Data Extraction and Quality Assessment

The same two researchers then independently extracted information from articles that passed both title and abstract screening as well as for methodological quality. Discrepancies were resolved through comparison and discussion. Extracted information from articles included: (1) author, (2) publication date, (3) dynamic balance quantification method, (4) study-design, (5) sample size, (6) sampling method, (7) demographic description, (8) inclusion/exclusion criteria, (9) unperturbed or perturbed walking, (10) type perturbation, (11) over-ground or treadmill walking, (12) number of strides, (13) walking duration, (14) Faller or Non-faller, (15) Freezer or non-Freezer, (15) Instrumentation.

To assess methodological quality, publications that proceeded to data extraction were assessed with a modified version of the Downs and Black Quality (Hubble et al., 2015). This checklist consists of 27-item that evaluates publications in regard to reporting, external validity, internal validity-bias, internal validity-selection bias, and power. General methodological quality was determined on 25 of the items with each having a corresponding single point. If publications met an item's criterion then one point was assigned to its score, if not then no point was assigned for that item. Additionally, no point was assigned if it was deemed unreasonable to determine if the item's criterion was met based on the publication's provided information. One item on the checklist assessed publications on a two-point scale based on reporting of principal cofounders. Specifically, twopoints were given if all cofounders were listed, one-point if cofounders were only partially described, and zero points if none were described. The final item on the checklist assessed the publication's statistical power, which was measured on a fivepoint scale and carried more weight in the final total score. The higher the publications statistical power, the higher the score that was provided.

# RESULTS

# Study Selection

The initial database search yielded 280 articles based on the defined search criteria (**Figure 1**) (Moher et al., 2009). Of the identified articles, 15 articles were removed as duplicates leaving 265 remaining publications for title and abstract screening. After title and abstract screening, 216 publications were deemed eligible for full-text screening based on their relevance to quantifying dynamic stability during walking. Full-text screening further excluded an additional 140 articles due to not meeting the predetermined inclusion criteria of this systematic review. The remaining 76 publications were included in the final synthesis and an additional 5 publications were included from manually scanning reference lists, thereby yielding a total of 81 publications for inclusion.

# Study Characteristics

Several methods for quantifying dynamic stability during unperturbed and perturbed walking were identified with some articles quantifying balance with multiple measures: 26 Coefficient of Variation for Spatiotemporal Variability, 10 Detrended Fluctuation Analysis, 20 Lyapunov Exponent, 7 Maximum Floquet Multipliers, 16 Extrapolated Center of Mass, 11 Harmonic Ratios, 4 Center of Mass-Center of Pressure Separation, 2 Gait Stability Ratio, 1 Entropy, 3 Spatiotemporal Variables, 2 COG or COP, 2 Root Mean Square.

Of the unperturbed articles: 18 examined dynamic balance in healthy young adults, 30 examined elderly adults, and 15 examined PD individuals. The age range for young adults was from 22 to 35 years old, elderly adult age range was 60.6–84 years old, and age range for PD individuals was 60.2 to 71.9. The PD severity was commonly assessed by the Hoehn & Yahr (H&Y) scale and ranged between stages I-III. Additionally, scores on the Unified Parkinson's Disease Rating Scale (UPDRS) ranged from 13.8 to 36.1. Within the PD articles, two examined differences between ON and OFF. All studies reported details on age and generally provided an appropriate age-matched control group. Fourteen articles examined differences between fallers and nonfallers distributed between 12 articles in elderly adults and 2 in PD individuals. Definitions of fallers varied substantially between articles, however, only two articles examined falls prospectively. Two articles examined differences between freezers and nonfreezers. In the perturbed walking articles 13 examined healthy young adults and 5 in elderly adults. No articles examined perturbed dynamic balance in PD individuals. The types and magnitude of the perturbations varied across studies. However, perturbed walking articles included 3 compliant surface (foam mat), 4 platform oscillations, 1 ML foot translation, 1 trunk pull, 5 slip, 2 trip, and 2 examined differences between both trips and slips. The articles were summarized into unperturbed and perturbed walking conditions then further divided by age group (young, elderly, and PD) and dynamic stability measure. All data extracted into the final synthesis are listed in **Tables 1**–**3** for unperturbed walking in young, elderly and PD individuals, respectively and **Tables 4**, **5** for perturbed walking conditions in young and elderly individuals, respectively.

# Methodological Quality

Seventy-nine of the assessed studies had a cross-sectional design while two studies used a randomized control protocol. After assessing methodological quality, 1 article was rated as having a very low methodological quality (range = 0–25% ), 72 having a low methodological quality (range = 25.51–50%), and 7 having a moderate methodological quality (range = 50.1–75%). In general, the majority of the articles scored poorly on criteria for internal validity (selection-bias), external validity and on reporting the statistical power for their paradigms. **Figure 2** displays the average percentage of each scoring for articles per category on the Downs and Blacks checklist sorted by study design. Articles scoring in the very low to low quality ranking are listed in **Table A** while moderately ranked articles are in **Table B** of the Supplementary Material.

# DISCUSSION

The purpose of this literature review was to examine the various methods for quantifying dynamic stability during unperturbed and perturbed walking in elderly and PD individuals. As falls in these demographics occur primarily during walking, accurately assessing gait instability is crucial to determine those with increased fall risk. After full-text screening, 81 publications were assessed for methodological quality and included in the final article synthesis. From this total 63 publications examined dynamic stability during unperturbed walking (**Table 1**) conditions and 18 publications during perturbed walking conditions (**Table 2**). Of these articles, 1 article had a very low methodological quality, 72 articles had a low methodological quality and 7 articles had a moderate quality. Articles are grouped by methodological quality in **Tables A, B** in the Supplementary Material. The low score in the majority of the articles was due to many neglecting to report on threats to internal validity and a priori power **Figure 2**. Indeed, based on the information provided, we were unable to determine if the samples were representative of their respective population. According to Downs and Black, a poor rating on internal validity items introduces an inherent amount of bias into the study as these items were designed to assess if a study's sample truly represents its population, which is crucial as statistical measures are based on the assumption that an unbiased sample was drawn from the population. By reporting specifications on participant recruitment, researchers would ensure that they are minimizing the risk of unsubstantiated conclusions regarding their sample and the inferences drawn for the associated target population. Furthermore, none of the articles reported an a priori power analysis to determine if their sample size was sufficiently large to achieve statistical significance. Item 27 on the Downs and Black Checklist scores up to five points based on statistical power. Overall, the lack of reporting may be due to researchers being unaware of their necessity in assessments of methodological quality as both components are generally implied as criteria for quality research.

The following discussion will elaborate on the main quantification methods, as reported in the literature, in

succession from unperturbed to perturbed conditions. Within each section, the method will be discussed in terms of the aspect of dynamic stability the respective method aims to quantify, how these measures are similar or different between demographics (young adults, older adults, and Parkinson Disease Individuals), the mechanisms of motor output associated with the method, the method's predictive ability in fall-risk assessment, and its limitations.

# Unperturbed Walking Methods

### Spatiotemporal Variability

Spatiotemporal variability is an overarching term that generally encompasses variability of stride time and length as well as step width. The variability in each of these parameters reflects a distinct aspect of motor control that contributes to stable walking. For instance, temporal parameters reflects the internal timing of the lower extremity (Hausdorff et al., 1998, 2003; Hausdorff, 2005). While spatial parameters, on the other hand, reflect an individual's ability to consistently spatially orient the lower extremity (Maki, 1997; Brach et al., 2005). However, spatial parameters can be further divided as current evidence indicates that separate motor processes control the lower extremity in the AP and ML directions (Kuo, 1999; Bauby and Kuo, 2000). Indeed, Bauby and Kuo (2000) demonstrated that lower extremity placement in the AP direction (stride and step length) is governed by automatic passive mechanisms while active mechanisms control ML placement (step width; Bauby and Kuo, 2000). Thus, each spatial parameter reflects a distinct aspect of motor control that contributes to stable walking. As the COM's trajectory dictates lower extremity placement, the neuromuscular system must integrate these multiple spatiotemporal parameters to successfully predict the COM's future dynamic state to avoid an uncontrolled fall (Brenière and Do, 1991; MacKinnon and Winter, 1993; Winter, 1995). Therefore, erratic spatiotemporal values reflect an inability to control extraneous COM movement within a rhythmic base of support. Previous research demonstrated that both aging and PD contribute to neuromuscular deterioration thereby threatening walking stability (Hausdorff et al., 1997, 2001; Brach et al., 2005; Baltadjieva et al., 2006; Nanhoe-Mahabier et al., 2011; Barbe et al., 2014; Kirchner et al., 2014). However, how aging and PD affects each spatiotemporal parameter specifically is inconsistently reported in the literature.

For instance, Malatesta et al. (2003) and Chien et al. (2015) reported increased stride time variability in healthy elderly adults compared to younger adults (Malatesta et al., 2003; Chien et al., 2015). In contrast, Hausdorff et al. (1997) demonstrated no differences in stride time variability between healthy elderly and young adults (Hausdorff et al., 1997). Additionally, aging's effect on spatial variability measures also yielded mixed results. When


examining differences in step length variability between healthy elderly and young adults, Ihlen et al. (2012b) found increased variability in the elderly group (Ihlen et al., 2012b). The authors suggested that the increased variability reflects a reduced ability of elderly individuals to redirect their COM at the beginning of the double support phase (Ihlen et al., 2012b). Contrastingly, however, Beauchet et al. (2009) demonstrated that the increased stride length variability in their sample was due to elderly adults walking at a slower velocity and not due to aging effects (Beauchet et al., 2009). Instead, the authors suggested that aging directly accounts for increases in step width variability and thus is a more accurate characteristic parameter of elderly gait. Three additional studies demonstrated significant differences in step width variability between elderly and young adults. However, these differences were not consistently due to increased step width variability in elderly participants (Helbostad and Moe-Nilssen, 2003; Brach et al., 2005; Hurt et al., 2010). Indeed, Hurt et al. (2010) found reduced step width variability in elderly adults compared to their younger counterparts (Hurt et al., 2010). The authors explained the differences as elderly participants increasing voluntary control of their trunk in the ML direction thereby reducing stride width variability (Hurt et al., 2010). However, Brach et al. (2005) demonstrated that fall likelihood was closely associated with both low and high step width variability in elderly adults (Brach et al., 2005).

The disparity in the literature may be due to methodological issues in the samples collected. Indeed, the articles discussed thus far either did not account for participants' fall history or did so retrospectively. In two seminal articles, Maki (1997) and Hausdorff et al. (2001) examined spatiotemporal gait variability in elderly adults and their ability to prospectively identify fallers from non-fallers (Maki, 1997; Hausdorff et al., 2001). Maki (1997) demonstrated that fallers walked with increased stride length variability while Hausdorff et al. (2001) found increased stride time variability characterized elderly fallers (Maki, 1997; Hausdorff et al., 2001). A plausible explanation for the differences between fallers and non-fallers is that aging does not uniformly affect the motor control mechanisms responsible for rhythmic walking in all elderly adults thereby resulting in only some elderly falling (Hausdorff et al., 1997). It is possible that neurodegeneration in some elderly adults (fallers) is physiologically more akin to individuals with pathological conditions (Hausdorff et al., 1997). Indeed, four studies in this literature review reported increased spatiotemporal variability in PD individuals compared to healthy age matched controls. Three studies reported increased stride time variability and one study reported increased stride length variability in PD individuals compared to elderly adults (Blin et al., 1990; Frenkel-Toledo et al., 2005a; Plotnik et al., 2007; Nanhoe-Mahabier et al., 2011).

Gait is predominately considered an automatic motor process that is generated in subcortical structures with only limited supraspinal input to provide environmental information (Hausdorff, 2005, 2009). Increased stride time and length variability in PD and elderly fallers suggests

#### TABLE 2 | Elderly adults unperturbed walking.



that neurodegeneration, although stemming from separate pathological origins, affects the motor pathways involved in gait's passive and internal timing mechanisms (Hausdorff et al., 2001; Plotnik et al., 2007; Nanhoe-Mahabier et al., 2011; Kirchner et al., 2014). Additionally, based on the present evidence, aging seems to have a greater impact on the active mechanisms controlling walking stability in the ML direction as both fallers and nonfallers have altered step width variability compared to young adults (Brach et al., 2005; Beauchet et al., 2009; Hurt et al., 2010). However, there is currently a lack of literature that examined step width variability in PD individuals. As frontal plane walking stability is dependent on active sensory integration, an ability impaired in PD, examining step width variability would reflect ML foot placement strategies in this demographic (Bauby and Kuo, 2000). Overall though, it is interesting to note that some amount of variability is consistently reported across studies in healthy young and elderly subjects (stride time <2%, stride length <3%, stride width <25%; Hausdorff et al., 1997; Maki, 1997; Brach et al., 2005). Brach et al. (2005) discussed the possibility that a moderate amount of variability is necessary for an individual to adapt to their environment (Brach et al., 2005). Thus, variability that is outside of this threshold may indicate both impairment for environmental adaptability as well as walking imbalance.

However, when quantifying spatiotemporal variability through the Coefficient of Variation, it is important to recognize this method's limitations. The Coefficient of Variation quantifies how much variability's magnitude contributes to the mean's value in any given gait parameter. While this method has the advantage of quantifying large spatiotemporal gait inconsistencies, it does not account for the variability's structure, which may provide additional information to the neuromuscular system's control of walking stability.

### Detrended Fluctuation Analysis

In his seminal article, Hausdorff et al. (1995) demonstrated that variations in a gait time series are not random but exhibit long-range correlations, where one stride influences subsequent strides (Hausdorff et al., 1995). To quantify these correlations, Hausdorff et al. (1995) implemented the Detrended Fluctuation TABLE 3 | Parkinson's disease unperturbed walking.


Hoehn & Yahr 3


#### TABLE 4 | Young adults perturbed walking articles.

Analysis (DFA) method from Dynamical Systems Theory to quantify the amount of persistence (correlation) in a time series data (Hausdorff et al., 1995). When applied to gait, the DFA assesses the amount of correlation between stride intervals in a walking session along a spectrum, with lower values (approaching 0) indicating strides are uncorrelated and larger values (approaching 1) indicating greater stride correlation (Hausdorff et al., 1995; Frenkel-Toledo et al., 2005b; Jordan et al., 2007).

However, the degree to which stride intervals are correlated with one another in young, elderly and PD individuals varies considerably in the literature. Despite these discrepancies, an emerging pattern does appear demonstrating that long-range correlations naturally occur in healthy adults (Hausdorff et al., 1995; Jordan et al., 2007; Dingwell and Cusumano, 2010; Terrier and Dériaz, 2011; Chien et al., 2015). Indeed, six studies examined long-range correlations in healthy young adults and reported DFA values ranging from 0.72 to 0.81 (Hausdorff et al., 1995; Malatesta et al., 2003; Jordan et al., 2007; Dingwell and Cusumano, 2010; Terrier and Dériaz, 2011; Chien et al., 2015). However, it remains unclear as to how aging affects the presence of these long-range correlations. For instance, Malatesta et al. (2003) reported no differences in DFA values between young and elderly adults (Malatesta et al., 2003). Contrastingly, Chien et al. (2015) demonstrated that long-range correlations begin to deteriorate when adults reach middle-age (DFA values of 0.76 in young adults and 0.64 in middle-aged adults), although no further DFA break downs were demonstrated between middleaged and elderly adults (Chien et al., 2015). As such, it remains unclear as to what effects aging have on the motor pathways responsible for stride interval correlations. Hausdorff et al. (1996) suggested that the presence of long-range correlations in healthy adults potentially reflect gait rhythmicity and automaticity generated by Central Pattern Generators (CPGs) in human subcortical structures (Hausdorff et al., 1996; Hausdorff, 2007, 2009). The contribution of CPGs to long-range correlations would explain the results found in PD. Three studies that investigated DFA in PD demonstrated reduced correlations in PD stride intervals compared to age-matched controls (Frenkel-Toledo et al., 2005b; Bartsch et al., 2007; Kirchner et al., 2014). For instance, Bartsch et al. (2007) reported DFA values of 0.72 in PD individuals and 0.80 in age-matched controls (Bartsch et al., 2007). Similarly, Kirchner et al. (2014) demonstrated lower DFA values in PD (0.76) compared to healthy elderly



adults (0.93) (Kirchner et al., 2014). This suggests that each step, in PD individuals, is more independent and unrelated to previous steps indicative of gait fluidity impairment (Hausdorff, 2009). Therefore, previous research proposed that PD individuals continuously restart the motor process that controls stepping instead of building off of the lower extremity's previous stepping states (Hausdorff, 2009). Interestingly, however, Bartsch et al. (2007) reported no differences between de novo PD individuals and age-matched controls (Bartsch et al., 2007). Hausdorff (2009) suggested that deterioration in the long-range correlation motor pathways may not be an early symptom of the disease but stated that it remains unclear whether this is due to compensatory mechanisms or if damage to the basal ganglia is not yet severe enough to impact this parameter (Hausdorff, 2009).

Although the exact mechanisms responsible for long-range correlations remain inconclusive, their presence in human gait is well-substantiated. However, the studies included in this literature review also demonstrate that a certain amount of anti-correlation is present in healthy human gait (Hausdorff et al., 1996; Jordan et al., 2007; Dingwell and Cusumano, 2010; Terrier and Dériaz, 2011). Hausdorff (2009) proposed that this may be to reduce the risk of perturbations leading to "mode locking" or resonance and may reflect an individual's capacity to adapt environmentally (Hausdorff, 2009). Thus, suggesting that an optimal correlation threshold may naturally exist in healthy gait. In support of this theory, Hausdorff (2007, 2009) borrowed evidence from research examining heart beat signals that demonstrated when signals exceeded or receded from the correlation threshold observed in healthy individuals, it was indicative of cardiovascular disease (Hausdorff, 2007, 2009).

Overall, DFA results are currently difficult to interpret and there is limited research that demonstrates its predictive ability to quantify fall risk. Indeed, of all the studies included in this literature review, none of the authors examined differences between fallers and non-fallers. Therefore, it is difficult to draw a conclusion as to the DFA's ability to identify fallers or predict future fall risk. Furthermore, a comparison between results of different studies is difficult due to the lack of standardization in both the DFA's formulaic computation and the number of strides quantified, a value that directly affects the DFA's calculation.

### Lyapunov Exponent

Another common method that quantifies gait stability from a dynamical systems approach is the maximum Lyapunov Exponent. This method quantifies a system's average logarithmic rate of divergence after infinitesimal perturbations (Dingwell and Marin, 2006; Bruijn et al., 2009a, 2010b, 2013). The underlying notion of the maximum Lyapunov Exponent is that if a system's current state is altered from that of its previous state, then either state is deemed as perturbed from the other (Bruijn et al., 2013). When applied to human walking, these perturbations are considered to arise from "noise" in either the neuromuscular system or from the environment. Calculating the Lyapunov Exponent is dependent on the construction of a proper and comprehensive state-space that adequately defines the state of the system in any point in time (Dingwell et al., 2001). In the literature, the predominate method for state-space construction is derived from an anatomical segment's kinematic data, such as velocity, acceleration, position and jerk (Dingwell et al., 2001; Dingwell and Marin, 2006; Bruijn et al., 2009a, 2010a). When computed, the Lyapunov Exponent quantifies the rate of convergence (<0) or divergence (>0) of the system's trajectory to its nearest neighboring trajectory in the reconstructed state space over the course of 0–1 strides (short-term Lyapunov Exponent) and 4–10 strides (long-term Lyapunov Exponent; Bruijn et al., 2009a). When the trajectories converge, the observed system is considered to have Local Dynamic stability while divergence indicates Local Dynamic Instability (Bruijn et al.,

2009a, 2013). However, as only a link between the short-term Lyapunov Exponent and gait stability has been established, only this component will be discussed and referred to Bruijn et al. (2013).

Currently, results from studies using the Lyapunov Exponent are difficult to interpret due to several methodological discrepancies. For instance, it is unclear which anatomical segment is most appropriate for calculating Local Dynamic Stability (LDS) and the literature is currently divided between the trunk and lower extremity joints. Indeed, three studies calculated the Lyapunov Exponent on lower extremity joints while twelve studies based their calculations on trunk kinematic parameters (Dingwell et al., 2001; Dingwell and Marin, 2006; Kang and Dingwell, 2006, 2009; England and Granata, 2007; Bruijn et al., 2009a, 2010b; Terrier and Dériaz, 2011; Ihlen et al., 2012a,b; Toebes et al., 2012, 2015; Stenum et al., 2014; Terrier and Reynard, 2015; Wu et al., 2016). In a comparison between the lower and upper extremities, Kang and Dingwell (2009) demonstrated greater local dynamic instability in the lower extremity in healthy young adults (Kang and Dingwell, 2009). The authors explained their results as the greater LDS of the upper extremity, compared to the lower, is plausibly due to greater trunk inertia (Kang and Dingwell, 2009). Furthermore, the authors suggested that, due to the different anatomical properties of each extremity, different motor control mechanisms may be responsible for maintaining Local Dynamic Stability (Kang and Dingwell, 2009). As stable motion of each extremity is necessary for successful locomotion, comparison between them may not be feasible as they would reflect different aspects of neuromuscular control and dynamic balance.

However, two commonalities emerge in the literature regardless of the extremity assessed. First, it is apparent that a certain amount of local dynamic instability exists in healthy individuals. Of the publications that examined local dynamic stability in healthy young adults, all authors reported Lyapunov Exponents that were greater than zero thereby confirming divergence rates of neighboring trajectories in this demographic (Dingwell and Marin, 2006; England and Granata, 2007; Bruijn et al., 2009a, 2010a,b; Kang and Dingwell, 2009; Ihlen et al., 2012a; Stenum et al., 2014; Wu et al., 2016). This may be due to inherent biological noise of the neuromuscular system and may additionally reflect an individual's attempt to attenuate unintended trajectories to maintain dynamic balance. Secondly, the literature converges on the notion that this divergence increases due to aging thus demonstrating that elderly adults have reduced local dynamic stability compared to younger controls. Indeed, both Ihlen et al. (2012b) and Kang and Dingwell (2009) demonstrated that lower extremity local dynamic stability was significantly reduced in elderly compared to healthy young adults (Kang and Dingwell, 2009; Ihlen et al., 2012b). Ihlen et al. (2012b) suggested that this may indicate an inability of elderly adults in controlling their COM's direction (Ihlen et al., 2012b). Similarly, in regard to the upper extremity, the literature demonstrates that aging reduces the trunk's local dynamic stability (Kang and Dingwell, 2009; Terrier and Reynard, 2015). Terrier and Reynard (2015) examined Local Dynamic Stability between young, middle-aged, and elderly adults and demonstrated that the trunk's mediolateral (ML) dynamic stability decreased as a function of aging (Terrier and Reynard, 2015). Current evidence demonstrates that trunk ML stability is achieved by "active" motor control mechanisms to maintain the COM within the base of support's frontal plane boundaries (Bauby and Kuo, 2000). Therefore, the reduced ML trunk stability in elderly adults may indicate an impairment in this demographic's ability to active mechanisms to maintain frontal plane dynamic balance, which potentially may result in falling. Indeed, Toebes et al. (2012) demonstrated that trunk Local Dynamic Stability in the ML direction was effective at retrospectively identifying elderly fallers from non-fallers.

Additional evidence examining dynamic stability differences between fallers and non-fallers is currently limited, and it is unclear whether the reduced LDS in elderly fallers causes falls or develops as a result of already falling. Additionally, the current methods for state-space reconstruction, the number of strides examined, and the Lyapunov Exponent's formulaic computation are to-date unstandardized, making comparison between studies difficult to interpret. Finally, current evidence lacks a direct link that elaborates on the exact mechanisms of neuromuscular output involved in the control of dynamic stability and how these mechanisms are affected in the presence of PD. Indeed, during the course of this literature review, no studies were found that met the inclusion criteria that examined local dynamic stability in PD individuals.

### Floquet Multipliers

The last commonly employed method from dynamical systems theory is Maximum Floquet Multipliers (FM), which measures a system's orbital stability. This method quantifies how the current state of a system diverges or converges away from a nominal periodic cycle at a discrete point (Dingwell, 2006; Bruijn et al., 2013). When applied to gait data, the nominal period is calculated as the average gait cycle in a time normalized state space (reconstructed from trunk kinematic parameters; Dingwell, 2006; Granata and Lockhart, 2008; Kang and Dingwell, 2009; Bruijn et al., 2010b). Afterward, a gait cycle is compared to the nominal period (average gait cycle) at a fixed discrete point along a Poincare Section, to assess if the cycle converges or diverges away from the nominal (average) period. When Floquet Multipliers are below the value of one, the system is considered orbitally stable, however, if the value approaches or exceeds one, then the system is considered to be diverging from the nominal period thereby threatening its orbital stability (Dingwell, 2006).

Similar to the Lyapunov Exponent, the results from Floquet Multipliers are difficult to compare and contrast due to a lack of standardization in the reconstructed state-space, the number of strides investigated and in their formulaic computation. Therefore, conflicting evidence exists regarding its ability to determine stability and differentiate between different demographics. For instance, Granata and Lockhart (2008) found that elderly fallers had reduced orbital stability compared to elderly non-fallers and young adults (Granata and Lockhart, 2008). However, the authors reported no additional differences between healthy elderly and young adults (Granata and Lockhart, 2008). In contrast, however, Kang and Dingwell (2009) demonstrated that healthy elderly adults were less orbitally stable compared to healthy young adults (Kang and Dingwell, 2009).

The motor control implications from FM remain largely inconclusive. Nevertheless, based on the literature, it appears that healthy young and elderly adults are capable of preserving orbital stability by minimizing deviations from their nominal limit cycle. Specifically, of the evidence included, all publications reported FMs of less than one (Dingwell, 2006; Granata and Lockhart, 2008; Kang and Dingwell, 2009; Bruijn et al., 2010b). Previous research suggests that when values exceed the value of one, then orbital stability is considered lost, which leads to falling (Dingwell, 2006). Granata and Lockhart (2008) discussed the possibility that FM quantify the neuromuscular system's ability to return to the limit cycle by attenuating arising deviations, which if left unmodulated would continue to expand (Granata and Lockhart, 2008). In turn this would threaten an individual's orbital stability, thus theoretically increasing fall likelihood. However, it is unclear whether this ability deteriorates uniformly in an aging population or if certain individuals (elderly fallers) are more impaired. Furthermore, it is unclear whether this impairment is a consequence from prior falls or is a factor that leads to falling.

Additionally, several limitations to FM need to be considered when applying this method to gait research. It is important to note that FM are based exclusively on the assumption that the system being quantified is strictly periodic (Dingwell, 2006; Bruijn et al., 2010b). Biological systems, such as human walking, are considered stochastic in nature thus drawing into question FM's applicability to quantify gait stability (Ashkenazy et al., 2002; Bruijn et al., 2013). Furthermore, this method analyzes orbital stability at discrete time points to an average value and does not examine differences between neighboring trajectories (Bruijn et al., 2013). Finally, during the course of this literature review, no publications were found that examined orbital stability in PD individuals. Thus, in addition to the aforementioned limitations, further investigation of FM is necessary to determine its ability in identifying fall risk, uncover the motor control mechanisms that contribute to orbital stability, and whether PD affects these mechanisms.

### Harmonic Ratios

Based on harmonic theory, Harmonic Ratios (HR) quantify walking balance by examining the periodicity from an acceleration signal (Lowry et al., 2009, 2012). Since control of the COM is crucial for maintaining walking balance, harmonic ratios are typically applied to trunk acceleration data (MacKinnon and Winter, 1993; Winter, 1995; Auvinet et al., 2002; Lowry et al., 2012). When examining the anteroposterior and vertical trunk accelerations, harmonic ratios assume that continuous walking consists of regular stride patterns with each stride consisting of two steps (Auvinet et al., 2002; Lowry et al., 2009, 2012). Therefore, rhythmic and stable acceleration signals should repeat in even-numbered multiples to be considered "inphase" with the stepping actions and result in a larger HR value (Auvinet et al., 2002). If, however, acceleration signals repeat as multiples of odd numbers then they are considered irregular and "out of phase" resulting in a smaller HR value (Auvinet et al., 2002). Overall, stability of an individual's walking pattern is then determined as a ratio of the summed amplitudes of the even to odd harmonics. In contrast to the AP and VT direction, rhythmic and stable acceleration signals in the mediolateral direction are characterized by multiples of odd numbers (Yack and Berger, 1993; Auvinet et al., 2002; Lowry et al., 2009, 2012). This is due to the fact that during heel strike, the COM is shifted in the frontal plane to the contralateral limb during stepping causing a monophasic, as opposed to biphasic, acceleration pattern during weight transfer in the double support phase (Auvinet et al., 2002).

Harmonic ratios, although grounded in a logical framework, is still an emerging method for quantifying walking stability. Of the publications found, conflicting evidence exists regarding HR's ability to differentiate trunk accelerations between demographics. For instance, Auvinet et al. (2002) and Lowry et al. (2012) demonstrated no differences between healthy young and elderly adults when participants walked at preferred walking speeds (Auvinet et al., 2002; Lowry et al., 2012). However, Bisi and Stagni (2016) reported a lower HR in the trunk's vertical accelerations for elderly adults over the age of 80 years old (Bisi and Stagni, 2016). Similarly, Yack and Berger (1993) found differences in the HR's of the anteroposterior and vertical accelerations in their "unstable" elderly participants compared to "stable" elderly and young adults (Yack and Berger, 1993). In opposition to the aforementioned studies, Kavanagh et al. (2005) demonstrated that elderly adults over the age of 70 years old exhibited lower HR's only in the trunk's mediolateral accelerations (Kavanagh et al., 2005). Furthermore, although still limited in the amount of research available, current evidence indicates a similar disparity in the trunk HR's for PD individuals. Indeed, two studies demonstrated lower trunk HRs for PD individuals compared to age-matched controls (Latt et al., 2009; Lowry et al., 2009). However, Lowry et al. (2009) reported lower HRs only in the mediolateral direction while PD participants in the study by Latt et al. (2009) had lower trunk HRs in all three movement planes (Latt et al., 2009; Lowry et al., 2009). Additionally, the authors reported differences between PD fallers and non-fallers in the anteroposterior and vertical planes (Latt et al., 2009).

Theoretically, rhythmic trunk accelerations indicate that the COM progresses between stance limbs along a smooth controlled trajectory (Auvinet et al., 2002). Control of the COM along the AP direction is considered largely "passive" as the COM's forward and downward momentum begins the stepping action with guiding input derived from the somatosensory system and subcortical cortical structures (Bauby and Kuo, 2000). Contrastingly, ML control is considered "active" as supraspinal input is required to determine lateral lower limb placement to stabilize the COM in this plane (Bauby and Kuo, 2000). Irregular COM acceleration may indicate an impairment in either the "passive" or "active" mechanisms responsible for COM movement. However, the disparity in the current literature makes it difficult to establish if and where impairments in trunk accelerations occur in elderly non-fallers, fallers and PD individuals. This disparity likely arises from the various methodologies used in each study. For instance, one issue arises from quantifying accelerations at different areas of the trunk. For instance, Yack and Berger (1993) as well as Mazzà et al. (2008) calculated HRs based on upper trunk accelerations (Yack and Berger, 1993; Mazzà et al., 2008). Contrastingly, Latt et al. (2009) and Brach et al. (2010) based their calculations on lower trunk accelerations (Latt et al., 2009; Brach et al., 2010). In a literature review, Winter (1995) collected evidence demonstrating that accelerations decreased in amplitude if measured on the upper trunk instead of the lower trunk (Winter, 1995). The author proposed that this is a stability mechanism whereby the neuromuscular system reduces perturbation amplitudes as they propagate toward the head in order to stabilize the visual field. As such, acceleration profiles between studies may not be comparable depending on whether the authors examined upper or lower trunk accelerations. A second cause for the lack of uniformity is caused by the different criteria used for participants. Auvinet et al. (2002) suggested that the inclusion criteria for participants, even within a specific demographic, will affect their gait performance (Auvinet et al., 2002). Future research should consider for example recording participants' activity level, fall history, and fear of falling level as these have been demonstrated to affect gait parameters and may potentially influence an individual's trunk acceleration patterns (Hausdorff et al., 1997; Toebes et al., 2012, 2015). Finally, it is important to consider that Harmonic Ratios do not directly quantify the COM in relation to the stability limits of the base of support (the lower extremity). As such, it is difficult to discern when altered trunk accelerations cause the COM to approach the base of support's stability limits.

### Extrapolated Center of Mass

In classical biomechanics, walking stability is defined as maintaining the COM within a series of unilateral stances (Pai and Patton, 1997). Therefore, the COM's position in relation to the stability regions of the stance limb (range of the Center of Pressure) has been used as a method for measuring stability (Pai and Patton, 1997). However, Hof et al. (2005) suggested that quantification of COM-COP position alone is insufficient for assessing dynamic balance (Hof et al., 2005). The authors demonstrated that although the COM may lie outside the base of support, stability can be achieved if the COM velocity is directed toward the COP (Hof et al., 2005). Therefore, to accurately quantify dynamic balance, the Extrapolated Center of Mass (xCOM) was proposed as a single parameter that assesses both COM position and velocity in relation to the base of support (COP) (Hof et al., 2005). Stability is then determined by quantifying the distance between the xCOM and the base of support. Within the literature, a shorter xCOM-BOS distance indicates greater stability as the COM's dynamic state lies closer to the Base of Support's boundaries (Lugade et al., 2011; Süptitz et al., 2012; Mademli and Arampatzis, 2014; Fujimoto and Chou, 2016; Yang and King, 2016). Typically, the xCOM is quantified at heel-strike and toe-off as these points are considered more unstable due to the transfer of the COM between limbs over a smaller base of support (Lugade et al., 2011; Fujimoto and Chou, 2016; Yang and King, 2016).

When examining this method, a consistent pattern emerges when elderly fallers are compared to elderly non-fallers and healthy young adults. Indeed, both Lugade et al. (2011) as well as Fujimoto and Chou (2016) demonstrated that elderly fallers ambulate with a reduced xCOM-BOS distance compared to the other two age demographics in the sagittal plane (Lugade et al., 2011; Fujimoto and Chou, 2016). Less consistent differences were reported between elderly non-fallers and young adults (Lugade et al., 2011; Mademli and Arampatzis, 2014; Fujimoto and Chou, 2016). Indeed, Lugade et al. (2011) reported no differences between elderly non-fallers and young adults, however, both Fujimoto and Chou (2016) as well as Mademli and Arampatzis (2014) found that elderly adults walked with a reduced xCOM-BOS distance compared to younger controls (Lugade et al., 2011; Mademli and Arampatzis, 2014; Fujimoto and Chou, 2016). Additionally, Fujimoto and Chou (2016) demonstrated that when the xCOM was derived from the COM's acceleration data elderly non-fallers had even greater reductions in xCOM-BOS distances compared to young adults(Fujimoto and Chou, 2016). The authors suggested that acceleration may be more sensitive to age-related differences and that the altered acceleration profiles indicate an inability to control the COM's momentum to preserve balance (Fujimoto and Chou, 2016). Interestingly, out of all the included publications, only Lugade et al. (2011) examined mediolateral xCOM-BOS, despite instability in this plane being closely associated with increased fall risk, and reported no differences between elderly fallers, non-fallers and young adults (Lugade et al., 2011).

In general, findings in the AP direction indicate that elderly fallers employ a strategy to keep the xCOM closer to their base of support's boundaries(Lugade et al., 2011; Fujimoto and Chou, 2016). Current evidence demonstrates that elderly adults, particularly fallers, modify their gait, compared to young adults, by reducing their walking velocity and stride length, and increasing step width (Maki, 1997; Herman et al., 2005). The combination of these modifications has been termed as the "cautious gait strategy" (Maki, 1997; Herman et al., 2005). Yang and King (2016) proposed that individuals implement this strategy to control the COM's dynamic state in order to more readily return the COM within the Base of Support in a potential balance loss (Yang and King, 2016). However, "cautious gait" does not explain the lack of differences found in the mediolateral direction. Although increasing step width would increase an individual's lateral base of support, previous research demonstrates that this action simultaneously increases trunk acceleration and velocity (Rosenblatt and Grabiner, 2010). This in turn only maintains an individual's already existing frontal plane balance instead of enhancing it (Rosenblatt and Grabiner, 2010).

When quantifying walking stability with the xCOM several limitations should be considered. For instance, a paradox exists within the xCOM theoretical definition and what is reported in the literature. According to the evidence, an individual is considered more stable when the xCOM is closer to their BOS (Lugade et al., 2011; Süptitz et al., 2012; Mademli and Arampatzis, 2014; Fujimoto and Chou, 2016; Yang and King, 2016). However, based on the publications found, the demographic that displayed the closest xCOM-BOS distance were elderly fallers (Lugade et al., 2011; Fujimoto and Chou, 2016). In contrast, healthy young individuals consistently exhibited the largest xCOM-BOS distance compared to both elderly non-fallers and fallers (Lugade et al., 2011; Mademli and Arampatzis, 2014; Fujimoto and Chou, 2016). As such, the current definition of stability for the xCOM is counterintuitive as elderly individuals are established to have a substantially increased fall risk, particularly if a history of falling exists (Hausdorff et al., 1997, 2001). Lugade et al. (2011) proposed the possibility that an increased xCOM-BOS may indicate increased stability as an individual can handle more dynamical states of the COM (Lugade et al., 2011). In addition to this paradox, the xCOM is only capable of assessing COM velocity and position in relation to the base of support at discrete time points and is therefore incapable of determining temporal effects on these variables. Furthermore, this method is based on the inverted pendulum theory and does not account for the effects of segments that are not represented in this model (Hof et al., 2005). Lastly, there appears to be an inconsistency in how authors define the base of support in their studies. When Hof et al. (2005) proposed the xCOM the authors defined the BOS as the range of the COP (Hof et al., 2005). However, several publications neglected to report how they defined the BOS or provided alternatives to this method. A clear and standardized definition of the BOS would be beneficial when examining xCOM-BOS distance.

# Unperturbed Walking Summary

Falling is one of the primary concerns for individuals with Parkinson's Disease and their caregivers. Therefore, there is a need for methods that can identify both individuals with fall risk and provide information on the neuromuscular system's impaired stability mechanisms. However, when examining the different quantification methods, it is important to consider how one defines stability. Indeed, this definition is crucial as it affects which anatomical structures are quantified (the lower or upper extremity), the quantification method itself, and the associated implications for motor control.

During walking, each extremity has a unique role that is controlled by different neuromuscular parameters. For instance, previous research proposed that CPG's control lower limb movement during walking (Hausdorff, 2009). Thus, lower extremity based methods may reflect the CPG's ability to control the stride-to-stride sequence. Evidence from spatiotemporal variability and detrended fluctuation analysis suggest that healthy adults are capable of stepping consistently (spatial variability), rhythmically (temporal variability) and in a correlated (DFA) manner culminating in regular stepping. In PD, however, these stepping processes are impaired to a degree beyond aging's effects. Indeed, compared to elderly adults, PD individuals walk with increased spatiotemporal variability and reduced long-range **correlations**. Plotnik and Hausdorff (2008) suggested that this increased variability is caused by impaired CPG output in this demographic (Plotnik and Hausdorff, 2008).

However, upper extremity stability may be controlled by alternative mechanisms. To accurately control the trunk during walking the neuromuscular system integrates multiple sensory systems to provide feedback information (Horak, 2006). As such, methods that quantify stability based on upper extremity parameters (Lyapunov Exponent, Floquet Multipliers, Harmonic Ratios) may reflect trunk stability mechanisms. The evidence from these various methods indicate that elderly adults are less capable of maintaining trunk stability compared to younger adults. Although still limited in research, these methods may hold direct implications for PD individuals due to their impairment in processing sensorimotor feedback.

Additionally, each quantification method reflects a unique aspect of neuromuscular control. For instance, both the Coefficient of Variation and DFA quantify variability in the lower extremity. However, the former quantifies variability magnitude while the latter quantifies variability in terms of its temporal structure. Hausdorff (2009) noted the magnitude and the temporal structure are two distinct characteristics and the value of each is independent of the other (Hausdorff, 2009). Similarly, Kang and Dingwell (2009) suggested that FM and the Lyapunov Exponent quantify different aspects of neuromuscular control as the upper extremity was more orbitally unstable and locally stable in both healthy young and older adults (Kang and Dingwell, 2009). Both methods quantify the divergence or convergence of variability in its own manner but are each based on specific assumptions about the investigated system. Finally, HRs and the xCOM both quantify the COM's kinematic state in relation to the base of support. However, HRs quantify stability through the trunk's cyclical movement, while the xCOM quantifies the trunk's motion state to the stability range of the base of support. As such, HRs represent an individual's ability to synchronize changes in trunk mechanics with that of the lower extremity. On the other hand, the xCOM indicates one's ability to withstand more dynamic conditions of the COM.

Overall, it appears that a certain amount of variability exists in the neuromuscular system and alterations outside this range increases fall likelihood. It is clear that aging and PD affect variability in some form indicating an altered capacity to process sensorimotor input, maintain rhythmical movement and execute corrective strategies. As walking is a complex skill requiring multiple neuromuscular control aspects and sensory input, implementing a single method to quantify stability is insufficient to reflect overarching balance impairment issues. Although all quantification methods have a grounded theoretical framework, additional research is necessary to determine their ability to predict fall risk in elderly and PD individuals. Indeed, only the Coefficient of Variation included studies that examined the differences between fallers and non-fallers prospectively thereby demonstrating its robustness in future fall prediction. Although methods, such as the Lyapunov Exponent, xCOM, and HR's examined differences between fallers and non-fallers, they did so retrospectively. Therefore, it is unclear whether these differences lead and caused the fall or if they arose only after fall onset. Thus, future research should consider prospectively examining potential differences within these demographics to determine each method's ability to predict falls. As elderly and PD individuals that have sustained a fall are at the greatest risk for future falls, quantifying dynamic stability with the Coefficient of Variation would assist clinicians in the early identification of individuals with unstable gait prior to fall onset. Additionally, implementing the Coefficient of Variation would assist clinicians in determining which aspect (passive or active mechanisms) are contributing to individuals' unstable gait. This subsequently would aid in the development of tailored gait therapy programs.

# Perturbed Walking Methods Extrapolated Center of Mass

The xCOM was proposed as a single parameter that accounts for both the COM's position and velocity together (Hof et al., 2005). Previous research demonstrated that the Central Nervous System is capable of proactively and reactively adapting the COM's motion state (position and velocity) in relation to the BOS before and after perturbations (McAndrew Young et al., 2012; Wang et al., 2012; Yang and Pai, 2013). These adaptations are theorized to reflect an individual's feedforward (proactive) and feedback (reactive) mechanisms to maintain and return the COM's to a stable motion state (Wang et al., 2012; Yang and Pai, 2013).

Wang et al. (2012) demonstrated that in response to tripping, young adults reactively reduce their COM velocity while simultaneously shifting it posteriorly (Wang et al., 2012). This adaptive response would bring the perturbed COM's motion state closer toward the BOS and would help neutralize the trunk's forward angular momentum induced by the trip. Additionally, the authors demonstrated that after repeated trip exposure, young adults proactively reduced their COM velocity in anticipation of the upcoming trip (Wang et al., 2012). Similarly, Yang et al. (2014) demonstrated that young adults proactively and reactively shift their COM forward and reduced its velocity in response to induced slips (Yang et al., 2014). However, the aging process appears to have detrimental effects on an individual's ability to engage in adaptive responses. Indeed, McCrum et al. (2016) found a reduced rate of adaptation in elderly adults, compared to young and middle-aged, during the onset of initial perturbations (McCrum et al., 2016). The authors suggested that the reduced adaptation rate increases fall risk in this demographic when exposed to continuous perturbations, such as uneven walking surfaces. However, the authors further demonstrated that after multiple perturbation exposure, elderly adults exhibited the same adaptation magnitude as the young and middle-aged groups. This finding indicates that aging affects the feedback mechanisms responsible for perturbation onset recognition, which in turn delays their response in executing adaptation strategies (McCrum et al., 2016). In contrast, it appears that feedforward mechanisms remain largely intact over the course of aging as no differences were found between age groups after repeated perturbation exposure (McCrum et al., 2016). Current evidence suggests that reactive adaptations differ from proactive ones in that they are rapidly executed in response to afferent input (Yang et al., 2014). However, despite the unpredictability of perturbations outside a laboratory setting, previous research demonstrated that feedforward mechanisms can facilitate feedback-controlled recovery mechanisms (Yang et al., 2014; McCrum et al., 2016). Yang et al. (2014) demonstrated that in anticipation to upcoming slips, young adults proactively implemented a "cautious gait" strategy that causes their COM position to be shifted anteriorly and reduce their COM velocity (Yang et al., 2014). Furthermore, Yang and Pai (2013) demonstrated that elderly adults exhibited similar proactive adaptations after exposure to repetitive slips that reduced post-training the percentage of falls in their sample during a novel slip (Yang and Pai, 2013). It is well-established in the literature that this "cautious gait" strategy is a characteristic feature in elderly adults during unperturbed walking conditions (Maki, 1997; Herman et al., 2005). Thus, it is plausible that this may be an implemented strategy that attempts to facilitate feedback recovery mechanisms, through proactive adaptations, in the event of a potential environmental perturbation.

One limitation that must be considered when interpreting xCOM results from perturbation studies is the lack of uniformity as to how the COM's motion state is defined. Several researchers reported the COM's position and velocity in relation to the BOS separately as opposed to a single measure. We chose to include these authors in this section as both position and velocity, in relation to the BOS, are the two necessary components for the xCOM and still provide an adequate representation of an individual's balance recovery mechanisms. However, differences between individuals may be more distinct when both COM position and velocity are combined as indicated by previous work (Fujimoto and Chou, 2016).

#### Lyapunov Exponent

Compared to unperturbed walking, relatively few articles were found that met this systematic review's inclusion criteria when examining the effect of perturbations on the Lyapunov Exponent's calculation of Local Dynamic Stability. Indeed, of the studies found, three examined the effect of minor surface perturbations and one examined an induced slip on participants' Local Dynamic Stability (Chang et al., 2010; McAndrew et al., 2011; Sinitksi et al., 2012; Yang and Pai, 2014). However, similar to unperturbed conditions, results from perturbation publications yield conflicting evidence.

For instance, McAndrew et al. (2011) reported that young adults had reduced Local Dynamic Stability (LDS), depicted by an increased Lyapunov Exponent, in all three cardinal planes (AP, ML, and VT) when perturbed by surface oscillations during walking (McAndrew et al., 2011). Additionally, the authors demonstrated that reductions in LDS were greatest when the Lyapunov Exponent was calculated in the same plane that the oscillation occurred in McAndrew et al. (2011). Similarly, Sinitksi et al. (2012) examined the effect between increasing amplitudes in ML surface oscillations and overground walking, and demonstrated that young adults were more unstable during perturbed trials than unperturbed but maintained LDS levels between oscillation amplitudes (Sinitksi et al., 2012). In contrast to these findings, no differences between surface conditions were found by Chang et al. (2010) when examining compliant foam surface and overground conditions (Chang et al., 2010). Furthermore, Yang and Pai (2014) demonstrated that LDS had a low ability in predicting falls from an induced slip in elderly adults (Yang and Pai, 2014).

Several methodological differences, in addition to the limitations mentioned previously, may account for the variable findings between studies. For instance, induced perturbations varied in type, intensity, and in their cardinal direction. In a literature review, Marigold and Misiaszek (2009) provided evidence that the neuromuscular system employs biomechanical recovery strategies that are specific to the encountered perturbation to maintain walking balance (Marigold and Misiaszek, 2009). Therefore, differences between Lyapunov Exponent values may reflect the various balance responses elicited by these specific perturbations. Additionally, these findings may indicate that certain walking conditions are more destabilizing to human walking than others. Indeed, Sinitksi et al. (2012) discussed the possibility that perturbation type has greater implications for walking stability than changes in perturbation amplitude (Sinitksi et al., 2012). However, previous research suggests that the Lyapunov Exponent has limited sensitivity in detecting local dynamic stability changes between various walking conditions due to the unstandardized methodology in its calculation (Bruijn et al., 2009b; Stenum et al., 2014). Furthermore, of the included publications, the Lyapunov Exponent was calculated based on trunk accelerometer data (Chang et al., 2010; McAndrew et al., 2011; Sinitksi et al., 2012; Yang and Pai, 2014). However, placement of the accelerometer varied between the Lumbar and Cervical regions amongst studies. As current evidence demonstrates that acceleration magnitudes diminish in superior trunk segments, this would affect the rate of divergence quantified by the Lyapunov Exponent thereby diminishing comparability between studies (Winter, 1995). Finally, the Lyapunov Exponent is only capable of quantifying a trajectory's divergence rate after infinitesimal perturbations (Bruijn et al., 2013). Thus, larger perturbations (trips or slips), may not be quantifiable with this method as they destabilize an individual globally (Bruijn et al., 2010a; Yang and Pai, 2014).

### Floquet Multipliers

Maximum Floquet Multipliers measures a system's convergence toward or divergence from a nominal (average) gait cycle due to neuromuscular noise or small environmental perturbations (Bruijn et al., 2013). Therefore, the majority of perturbation research on FM concentrates on the effect small surface perturbations, as opposed to trips or slips, has on an individual's orbital stability (McAndrew et al., 2011; Sinitksi et al., 2012; Yang and Pai, 2014). However, it is important to note that although a substantial amount of perturbation research was conducted on FM through modeling and robotics studies, relatively few experimental articles exist by comparison.

Of the studies included, two examined the effect small surface oscillations had on the trunk's orbital stability while one article examined FM's fall-predictive ability after an induced slip (McAndrew et al., 2011; Sinitksi et al., 2012; Yang and Pai, 2014). In a study of AP and ML surface oscillations on trunk orbital stability, McAndrew et al. demonstrated that heathy young adults became less orbitally stable in the direction of the induced oscillation (McAndrew et al., 2011). Additionally, the authors reported that participants were more sensitive to ML oscillations as the direction specific effects on trunk orbital stability were greatest in this direction (McAndrew et al., 2011). Indeed, Kuo (1999) demonstrated that humans are more unstable in the ML direction during walking (Kuo, 1999). Additionally, current literature has established that increased ML instability is closely linked with fall risk (Porter and Nantel, 2015). However, it is unclear how strong a destabilizing effect an individual can withstand before orbital stability is lost and a fall ensues. Sinitksi et al. (2012) reported that increasing surface oscillation amplitudes in the ML direction reduced orbital stability in the same direction in healthy young adults (Sinitksi et al., 2012). Additionally, the authors further stated decreases in orbital stability were marginal despite being statistically significant, therefore perturbation type may be more critical than magnitude (Sinitksi et al., 2012). This may explain findings by Yang and Pai (2014) who demonstrated that FM had a low predictive ability in differentiating fallers from non-fallers from an induced slip (Yang and Pai, 2014). However, an alternative explanation may account for these findings. As previously stated, FM are a measure of a systems convergence of divergence from a nominal period (average gait cycle; Bruijn et al., 2013). If FM quantify the neuromuscular system's capacity to modulate the continuous rise in deviations, as suggested by Granata and Lockhart (2008), it may be unable to detect large instantaneous stability threats, such as trips or slips (Granata and Lockhart, 2008). Overall, experimental perturbation research on FM are limited and further research is necessary to determine its suitability in quantifying perturbations effects.

### Perturbed Walking Summary

Slips, trips, and uneven terrain pose a substantial threat to walking stability and account for more than a quarter of all falls in elderly adults. In response to perturbations, current evidence demonstrates that the neuromuscular system employs several proactive and reactive strategies to return the individual to a state of stability (Yang et al., 2014, 2016). However, the effectiveness of these strategies may deteriorate over the course of aging and in the presence of PD (McCrum et al., 2016). Therefore, determining appropriate methods that can quantify an individual's stability when encountering a perturbation are necessary to provide information into the capability of impaired neuromuscular functioning (aging and PD) in returning to a state of stability.

Previous research indicates that the neuromuscular system employs biomechanical recovery strategies that are specific to the encountered perturbation (Marigold and Misiaszek, 2009). As such, the types of perturbations, along with their associated recovery strategies, likely elicit different effects on dynamic stability measures and are likely non-comparable. Additionally, based on their theoretically framework, each quantification method may measure a distinct aspect of neuromuscular recovery. For instance, Floquet Multipliers and the Lyapunov Exponent measure convergence/divergence of a system's trajectory after infinitesimal perturbations and, therefore, may reflect smaller and more fine-tuned neuromuscular strategies to attenuate these miniscule deviations to a system's trajectory (Bruijn et al., 2013). Contrastingly, the xCOM may reflect larger strategies that preserve global stability through the aforementioned feedforward and feedback mechanisms (Yang et al., 2014).

Overall, it is clear that the perturbation type, quantification method utilized, and an individual's stability state prior to perturbation onset are critical factors when assessing perturbations' threat to walking balance. Interestingly, during the course of this literature review, a limited amount of research met our inclusion criteria when examining perturbations on PD individuals. Furthermore, no studies included examined differences in recovering dynamic stability in fallers compared to non-fallers. Previous research suggests that aging does not affect all individuals uniformly and fallers have altered motor control output that is similar to that of clinical populations (Hausdorff et al., 1997). Therefore, future research should consider examining recovery strategies in both fallers and PD individuals as their perturbation responses are plausibly altered compared to age matched controls. This is particularly important as both groups are continuously reported to have a high fall-risk (Bloem et al., 2004; Canning et al., 2014). Additionally, based on the included articles, it appears that methods stemming from Dynamical Systems Theory (Lyapunov Exponent and Floquet Multipliers) are limited in assessing the effects of perturbations. However, Bruijn et al. suggested that the stability state of the system (the individual) prior to a perturbation may affect the perturbation's destabilizing effects (Bruijn et al., 2010a). As such, future research should consider examining how much dynamical stability a system must exhibit to minimize destabilizing forces and whether this is altered in fallers and PD individuals. Furthermore, it is unclear how long it takes an individual (young, elderly, and PD) to return to a stability state post-perturbation. Thus, current clinicians should consider implementing the xCOM to quantify individuals' responses to perturbations. In doing so clinicians could not only determine potentially impaired feedback responses but also develop programs that facilitate the training of feedforward mechanism to reduce falls from perturbations.

# LIMITATIONS

Several limitations should be considered when interpreting the results of this review. For instance, we did not examine differences between overground and treadmill walking, which is demonstrated to affect several dynamic balance measurements. Additionally, we did not examine dual-tasking or feedback paradigms which provide valuable information into gait's cognitive and motor control processes. Furthermore, we did not examine differences between freezer and non-freezers or between ON and OFF medications both of which have been demonstrated to affect dynamic balance in this demographic. Additionally, findings from perturbed evidence in this literature should be considered with some limitations. First, relatively few articles that examined perturbation effects on Floquet Multipliers and the Lyapunov Exponent were found, due to the relatively new application of these methods to perturbation paradigms. As such, both methods warrant further investigation in the field for more in-depth conclusions. Secondly, we only examined mechanical perturbation literature and did not include sensory perturbations in our synthesis. Furthermore, limited research was present that examined mechanical perturbations' effects on fallers and PD individuals. As such, additional research is required to examine perturbation response strategies in both demographics. Additionally, as most of the articles had a low score on the Downs and Black checklist, this may have introduced unintended bias into our assessment when interpreting the results.

# CONCLUSION

After examination of the evidence, it appears that each quantification method provides unique information into dynamic stability control, or lack thereof, in young, elderly and PD adults. Therefore, several considerations should be given when selecting a quantification method as they each appear to reflect a unique aspect of neuromuscular control, which when impaired may contribute to falling in elderly and PD individuals. Considerations, such as the walking condition, perturbation type and magnitude, as well as if the upper or lower extremity should be quantified. Based on the evidence, future clinicians and researchers should consider the method of quantification carefully as to reflect the aspect of neuromuscular control that they wish to examine. Additionally, clinicians should consider using the Coefficient of Variation and the Extrapolated Center of Mass when, respectively, examining unperturbed and perturbed walking conditions in their clients. The articles examined indicate that the Coefficient of Variation is the most supported method in predicting future falls during unperturbed walking. While the Extrapolated Center of Mass provides a robust indication to the effectiveness of perturbation response strategies. As such, both methods can provide not only a robust quantitative assessment for fall risk but also insight into impairments to gait's motor control mechanisms.

# REFERENCES


# AUTHOR CONTRIBUTIONS

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

# FUNDING

Funding for this project was provided by the Natural Sciences and Engineering Research Council (NSERC) Discovery grant RGPIN-2016-04928 and NSERC Accelerator supplement RGPAS-493045-2016.

# ACKNOWLEDGMENTS

We would like to acknowledge Allen Hill for his assistance in article screening and data extraction.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnagi. 2018.00387/full#supplementary-material


**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 Siragy and Nantel. 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.

# Visual-Somatosensory Integration and Quantitative Gait Performance in Aging

Jeannette R. Mahoney <sup>1</sup> \* and Joe Verghese1,2

<sup>1</sup> Division of Cognitive and Motor Aging, Department of Neurology, Albert Einstein College of Medicine, Bronx, NY, United States, <sup>2</sup> Division of Geriatrics, Department of Medicine, Albert Einstein College of Medicine, Bronx, NY, United States

Background: The ability to integrate information across sensory modalities is an integral aspect of mobility. Yet, the association between visual-somatosensory (VS) integration and gait performance has not been well-established in aging.

Methods: A total of 333 healthy older adults (mean age 76.53 ± 6.22; 53% female) participated in a visual-somatosensory simple reaction time task and underwent quantitative gait assessment using an instrumented walkway. Magnitude of VS integration was assessed using probability models, and then categorized into four integration classifications (superior, good, poor, or deficient). Associations of VS integration with three independent gait factors (Pace, Rhythm, and Variability derived by factor analysis method) were tested at cross-section using linear regression analyses. Given overlaps in neural circuitry necessary for both multisensory integration and goal-directed locomotion, we hypothesized that VS integration would be significantly associated with pace but not rhythm which is a more automatic process controlled mainly through brainstem and spinal networks.

#### Edited by:

Paolo Cavallari, Università degli Studi di Milano, Italy

#### Reviewed by:

Matthieu P. Boisgontier, University of British Columbia, Canada Enrico Mossello, Università degli Studi di Firenze, Italy

\*Correspondence: Jeannette R. Mahoney jeannette.mahoney@einstein.yu.edu

Received: 24 September 2018 Accepted: 30 October 2018 Published: 27 November 2018

#### Citation:

Mahoney JR and Verghese J (2018) Visual-Somatosensory Integration and Quantitative Gait Performance in Aging. Front. Aging Neurosci. 10:377. doi: 10.3389/fnagi.2018.00377 Results: In keeping with our hypothesis, magnitude of VS integration was a strong predictor of pace (β = 0.12, p < 0.05) but not rhythm (β = −0.01, p = 0.83) in fully-adjusted models. While there was a trend for the association of magnitude of VS integration with variability (β = −0.11, p = 0.051), post-hoc testing of individual gait variables that loaded highest on the variability factor revealed that stride length variability (β = −0.13, p = 0.03) and not swing time variability (β = −0.08, p = 0.15) was significantly associated with magnitude of VS integration. Of the cohort, 29% had superior, 26% had good, 29% had poor, and 16% had deficient VS integration effects.

Conclusions: Worse VS integration in aging is associated with worse spatial but not temporal aspects of gait performance.

Keywords: multisensory processing, sensorimotor integration, gait, falls, mobility

# INTRODUCTION

Gait, a complex sensorimotor behavior involving coordination of neural networks, bones, muscles and joints, requires sensory information to aid in control of movement and to influence gait adaptation (Barbieri and Vitório, 2017). Effective integration of concurrent sensory stimulation is crucial for successful mobility. In our previous work, we demonstrate a protective effect of multisensory integration in aging whereby greater ability to integrate visual and somatosensory information was associated with increased balance performance, and decreased likelihood of falls (Mahoney et al., 2014, 2018).

To our knowledge, the association between visualsomatosensory (VS) integration and gait performance has not been established. Verghese and colleagues identified three independent gait domains using a factor analysis approach (Pace, Rhythm, and Variability; see Verghese et al., 2007b). The pace factor encompasses spatial parameters including gait speed, stride length, and percentage of gait cycle spent in double support (i.e., immobilized with two feet on the ground), whereas the rhythm factor includes temporal parameters such as cadence (number of steps per minute), swing time and stance time. The variability factor quantifies inconsistencies (measured in standard deviation units) of the highest loading gait variables of both spatial (stride length) and temporal (swing time) domains. Noteworthy, these gait domains have been verified by other investigators (Hollman et al., 2011; Lord et al., 2013; Verlinden et al., 2014).

Neuroimaging of the brain during walking has not been perfected yet; however existing models of locomotion reveal associations of neural activation in cortical (frontal/supplementary motor/parietal), subcortical (basal ganglia/thalamus), cerebellar, and brainstem regions with mobility outcomes (Holtzer et al., 2014). Noteworthy, multisensory integration effects have also been linked to cortical [frontal/motor/primary sensory areas/superior temporal sulcus (STS)] and subcortical (superior colliculus/thalamus) regions in cats, primates, and humans (Meredith and Stein, 1986; Stein et al., 2002; Calvert et al., 2004; Schroeder and Foxe, 2004). Given noticeable overlaps in the neural circuitry necessary for both sensory integration and goal-directed locomotion through space (sensory/motor regions, basal ganglia, and thalamus to name a few), we hypothesize that VS integration will be significantly associated with spatial aspects of gait (pace) and not with temporal aspects of gait (rhythm) which appear to be more automatic processes, influenced less by sensory inputs, and controlled mainly through brainstem and spinal networks (Kandel et al., 2013). The variability factor encompasses aspects of both pace and rhythm that could collectively be associated with VS integration; however, if our above hypothesis is supported, then VS integration should only be associated with stride length variability and not swing time variability.

# MATERIAL AND METHODS

# Participants

Three-hundred-ninety-five participants enrolled in the Central Control of Mobility in Aging (CCMA) study at the Albert Einstein College of Medicine in Bronx, New York completed a multisensory simple reaction time (RT) experiment between June 2011 and June 2018. CCMA eligibility criteria required that participants be 65 years of age and older, reside in lower Westchester county, and speak English. Exclusion criteria for the CCMA included inability to independently ambulate, presence of dementia, significant bilateral vision, and/or hearing loss, active neurological or psychiatric disorders that would interfere with evaluations, recent or anticipated medical procedures that would affect mobility, and/or receiving hemodialysis treatment (see also Holtzer et al., 2013a,b). Presence of dementia was excluded using reliable cut scores from the AD8 Dementia Screening Interview (cutoff score ≥2; Galvin et al., 2005, 2006) and the Memory Impairment Screen (MIS; cutoff score < 5; Buschke et al., 1999); and later confirmed using consensus clinical case conference.

Additional exclusion criteria included history of severe unilateral vision (n = 5) and/or hearing loss (n = 4). All participants were required to successfully complete a sensory screening exam, where visual, auditory, and somatosensory acuity were formally tested to ensure appropriateness for the study. All CCMA participants were required to have bilateral visual acuity that was better or equal to 20/100 as measured by the Snellen eye chart. Individuals that were unable to hear a 2,000 Hz tone at 25 dB in both ears were not included in the CCMA study. As in our previous studies, presence or absence of neuropathy was diagnosed by the study clinician, and participants with severe neuropathy (unable to feel somatosensory stimulation) were not included (Mahoney et al., 2014, 2015, 2018). Additional exclusion criteria included inadequate multisensory performance (n = 40; see below) and prevalent dementia at study enrollment (n = 13).

After exclusions, the total study sample consisted of 333 older adults (mean age 76.53 ± 6.22 years; 53% female). All participants provided written informed consent to the experimental procedures, which were approved by the institutional review board of the Albert Einstein College of Medicine, Bronx, NY in accordance with the Declaration of Helsinki.

# Stimuli, Task, and Responses

Visual, somatosensory (vibratory pulses), and simultaneous VS stimuli were delivered through a custom-built stimulus generator (Zenometrics, LLC; Peekskill, NY, USA) that consisted of two control boxes, each housing a 15.88 cm diameter blue light emitting diodes (LEDs) and a 30.48 × 20.32 × 12.70 mm plastic housing containing a vibrator motor with 0.8 G vibration amplitude (Mahoney et al., 2015, 2018; Dumas et al., 2016). The devices were connected to a network control center, which allowed direct control for each device through the testing computer's parallel port. The devices were cycled on and off at precise predetermined intervals in any combination. A TTL (transistor-transistor-logic, 5 V, duration 100 ms) pulse was used to trigger the visual and somatosensory stimuli through E-Prime 2.0 software.

The control boxes were mounted to an experimental apparatus, which participants comfortably rested their hands upon, with their index fingers strategically placed over the vibratory motors on the back of the box and their thumb on the front of the box, under the LED (see **Figure 1**). A third dummy control box was placed in the center of the actual control boxes, at an equidistant length (28 cm) and contained a bull's eye sticker with a central circle of 0.4 cm diameter which served as the fixation point. To ensure that the somatosensory stimuli were inaudible, each participant was provided with headphones over which continuous white noise was played.

The three conditions were presented randomly with equal frequency and consisted of three blocks of 45 trials, for a total of 135 stimuli. Each block was separated by a 20 s break in order to reduce fatigue and facilitate concentration, and each subsequent block commenced immediately after the conclusion of the break. Participants were instructed to respond to all stimuli by pressing a stationary pedal located under their right foot as quickly as possible. Performance accuracy was defined as the number of accurate stimulus detections divided by 45 trials per condition. To prevent anticipatory effects, the inter-stimulus-interval varied randomly from 1 to 3 s. The duration of the entire experiment was approximately 7 min.

As in our previous multisensory studies, a 70% performance accuracy cutoff for all conditions was implemented to exclude participants with unreliable responses (n = 40; Mahoney et al., 2014, 2015, 2018). To be consistent with (Mahoney et al., 2018), data trimming procedures were purposefully avoided so as to not bias the distribution of the RT data (see Gondan and Minakata, 2016). If the participant failed to respond to any given stimulus, then that trial was considered inaccurate (omitted) and the corresponding RT was set to infinity rather than excluded from the analysis (Gondan and Minakata, 2016; Mahoney et al., 2018). To facilitate comparisons to other multisensory studies, the overall RT (average of all RTs regardless of condition) and overall RT facilitation effect (i.e., RT difference between the multisensory VS condition and the fastest unisensory condition) is included in **Table 1**.

# Quantification of Multisensory Integration using the Race Model Inequality

When two sources of sensory information are presented concurrently, they offer synergistic information that gives rise to faster responses, namely a redundant signals effect (Kinchla, 1974). Race models, commonly implemented to examine multisensory effects, are robust probability (P) models that compare the cumulative distribution function (CDF) of combined unisensory visual (V) and unisensory somatosensory (S) reaction times with an upper limit of one [min [P(RT<sup>V</sup> ≤ t) + P(RT<sup>S</sup> ≤ t), 1] to the CDF of multisensory VS reaction times [P(RTVS≤ t)] (Miller, 1982; Maris and Maris, 2003; Colonius and Diederich, 2006). For any latency t, the race model inequality (RMI) holds when the CDF of the **actual** multisensory condition [P(RTVS ≤ t)] is less than or equal to the **predicted CDF** [min (P(RT<sup>V</sup> ≤ t) + P(RT<sup>S</sup> ≤ t), 1)]. Note that these CDFs take all RTs into account and have been extensively reviewed and utilized in our previous studies (Mahoney et al., 2015, 2018; Dumas et al., 2016). Acceptance of the above RMI suggests that unisensory signals are processed in parallel, such that the fastest unisensory signal could produce the actual response (i.e., the "winner" of the race). However, when the actual CDF is greater than the predicted CDF, the RMI is rejected and the RT facilitation is the result of multisensory interactions that allow signals from redundant information to integrate or combine non-linearly.

In order to calculate the race model violation, RTs must be sorted by condition in ascending order and the RT range across all stimulus (V, S, or VS) conditions must be calculated on an individual level. RT data are then quantized into 20 bins from the fastest RT (or zero percentile) to the slowest RT (hundredth percentile) in 5% increments (0%, 5%, . . . , 95%, 100%) separately for each condition. For example, let us suppose that for one individual the fastest RT was equal to 100 ms and the longest RT was equal to 1,000 ms (regardless of stimulus condition). Here, the single fastest RT of 100 ms would be represented at the 0th percentile. The next cumulative percentile bin (5%), would take all RTs that fell within the 100 ms + [5% of the range (1,000–100 ms = 900 ms range <sup>∗</sup> 5% = 45 ms)] into account, so RTs between 100 and 145 ms. The 10% bin, would then consider RTs that were between 100 and 190 ms and so on until we reached the last RT (or 100%) percentile bin which would take all RTs from the 100–1,000 ms range into account. This method is implemented once for each of the three stimulus conditions and the probability of any RT occurring within each bin is calculated and transformed into a CDF. The CDF of the multisensory VS RTs represents the **actual** multisensory CDF, while the summation of the CDFs for both the visual and somatosensory CDFs (with an upper limit of 1) represents the **predicted** CDF. The difference in these two CDFs represents the Race Model inequality (RMI), where positive values are indicative of successful multisensory integration (also referred to as a violation of the race model). **Figure 2** depicts the group-averaged difference between **actual** and **predicted** CDFs (dashed trace), where positive values (shaded area between 0 and 10th percentile) are indicative of VS integration (i.e., rejected RMI). The RMI was tested using Gondan's permutation test 2010 over the fastest 10% of responses and a robust violation was observed (tmax = 13.43, tcrit = 2.05, p < 0.001).

As in our previous study (Mahoney et al., 2018), actual CDF difference values for these three violated percentile bins (0, 5, and 10%) were used to (1) calculate the area-under-the-curve (AUC) TABLE 1 | Demographic and clinical characteristics of study sample overall and by classification\*


.

\*Values are presented as mean ± SD for continuous variables and % for dichotomous variable.

#Area under the curve of the CDF difference wave over the 0–10% percentile.

<sup>∧</sup>Result of Between Groups One-Way ANOVAs.

which served as our independent variable of 'magnitude of VS integration' for further statistical modeling and (2) determine VS integration classification. VS integration classification was assigned based on the number of violated percentile bins (0, 1, 2, or 3) during the 0–10th percentile. Classification definition was operationalized as follows: if all percentile values violated the RMI the individual was considered a "superior" integrator; if two values violated the RMI, the individual was considered a "good" integrator; if one value violated the RMI, the individual was considered a "poor" integrator; and if zero values violated the RMI, the person was considered a "deficient" integrator." **Figure 2** also depicts race model difference waves by integration classification (solid grayscale traces).

# Clinical Evaluation

Global cognitive status was assessed using the Repeatable Battery for Assessment of Neuropsychological Status (Duff et al., 2008). As in our previous studies, global health scores (range 0–10) were obtained from dichotomous rating (presence or absence) of physician diagnosed diabetes, chronic heart failure, arthritis, hypertension, depression, stroke, Parkinson's disease, chronic obstructive pulmonary disease, angina, and myocardial infarction (Mahoney et al., 2011, 2014, 2015, 2018; Dumas et al., 2016).

# Gait Evaluation

Quantitative gait assessments were conducted using a 28 foot instrumented walkway with embedded pressure sensors

that provide various spatial and temporal gait parameters (GAITRite, CIR Systems, Havertown, PA). GAITRite, a valid system for measuring gait performance with excellent test-retest reliability (Bilney et al., 2003; Menz et al., 2004; Brach et al., 2008), is widely used in clinical and research settings (Verghese et al., 2007a). Here, steady-state locomotion was captured over a distance of 20 feet; data from the first and last 4 feet of the instrumented walkway (void of sensors) were purposefully excluded to eliminate initial acceleration and terminal deceleration. Participants were asked to walk on the mat at their "normal walking speed" in a quiet and well-lit room (see Verghese et al., 2002).

Similar to our previous work (Verghese et al., 2007b), factor analysis using the principal component method was performed on eight individual spatiotemporal gait parameters: gait velocity, stride length, percentage of double support, stride time, stance time, cadence, stride length variability, and swing time variability. The advantage of a factor approach using orthogonal varimax rotation is to reduce a large number of potentially correlated variables (while retaining most of the information) into a smaller number of uncorrelated independent factors that reduces the redundancy across individual variables. We identified a total of three independent gait factors (namely: Pace, Rhythm, and Variability) which later served as dependent variables in subsequent analyses. The pace factor includes three spatial variables: gait velocity, stride length, and percentage of immobilized gait or double support. The rhythm factor includes three temporal variables: stride time, stance time, and cadence which is number of steps per minute. The variability factor comprised both spatial (stride length variability) and temporal (swing time variability) facets of gait measured in SD units.

# Statistical Analysis

Data were inspected descriptively and graphically and the normality of model assumptions was formally tested. Descriptive statistics (M ± SD) were calculated for continuous variables and between group ANOVAs were conducted. All data analyses were run using IBM's Statistical Package for the Social Sciences (SPSS), Version 24.

Three linear regression analyses (one for each gait factor) were performed with pace, rhythm, or variability serving as the dependent variable and VS integration as the independent variable in unadjusted models. Additional covariates were entered in a stepwise manner. In Step 2, age and gender were added as independent variables. In Step 3, additional independent variables included presence of moderate visual loss, presence of mild neuropathy, and global health score. If adjusted associations were significant, additional sensitivity analyses were conducted to determine whether adjustments for Overall RT or RBANS Total Index score impacted the association of VS integration with the dependent measure. In an effort to further scrutinize the variability factor, given our a priori hypotheses, two additional regression models were conducted to examine the individual association of VS integration with spatial (stride length variability) and temporal (swing time variability) variability components.

# RESULTS

Demographic information is presented in **Table 1** for both the overall cohort and for each multisensory classification group. Results demonstrate significant RMI violation over the fastest 10% of RTs using an established permutation test (Gondan, 2010); suggesting robust multisensory effects for the entire cohort. Difference values between actual and predicted CDFs

TABLE 2 | Factor loading of quantitative variables on three independent gait factors# .


#Rotation method: Varimax with Kaiser Normalization.

Bold values indicate loadings.

were individually calculated for the violated percentile bins (0, 5, and 10%) and used to determine (1) multisensory integration classification group and (2) magnitude of VS integration. Based on our operational definition, our sample consisted of 95 superior integrators; 87 good integrators; 96 poor integrators; and 55 deficient integrators.

Factor analysis with varimax rotation yielded three orthogonal factors that accounted for over 87% of the variance in quantitative gait performance (**Table 2**). The factor with the highest variance, pace, had strong loadings by spatial parameters including velocity, stride length, and percent of gait cycle spent immobilized (double support). The second factor, rhythm, had strong loadings by temporal parameters including swing time, stance time, and cadence (or number of steps per minute). The last factor, variability, loaded highly on stride length (spatial), and swing time (temporal) variability measures. Mean factor score was 0 (SD 1), and factor scores can be conceptualized as summary risk scores with high scores representing worse performance.

Results from the linear regression analyses (**Tables 3A–E**) reveal that VS integration processes, as quantified by the amount area-under-the-curve in the CDF difference wave, are associated with pace (β = 0.16, p ≤ 0.001) and variability (β = −0.12, p ≤ 0.04) factors, but not rhythm (β = 0.00, p = 1.00) in unadjusted models. The multisensory effect remained associated with pace even after controlling for age, gender, visual impairment, neuropathy, and global health score in models 2 through 3 (β = 0.12, p < 0.05), but not variability (β = −0.11, p = 0.051). Given our hypothesis regarding the association of VS integration with spatial and not temporal gait factors, we further examined the association of multisensory integration with stride length variability and swing time variability separately. Our findings revealed that only stride length variability (spatial aspect) was associated with VS integration in fully adjusted models (β = −0.13, p < 0.05). Sensitivity analyses further confirmed the significant association between VS integration and Pace even when adjusting for Overall RT (β = 0.13, p = 0.015) or RBANS Total index score (β = −0.13, p = 0.017). As well, the association between VS integration and Stride Length variability remained significant when adjusting for Overall RT (β = −0.12, p = 0.035) or RBANS Total index score (β = −0.13, p = 0.023).

# DISCUSSION

The main objective of the current study was to determine whether ability to integrate concurrent VS information was associated with specific aspects of gait performance in older adults. Our findings reveal robust, but differential VS integration effects; 29% of the current study sample were superior VS integrators, while 26, 29, and 16% were considered good, poor, and deficient VS integrators, respectively. Our results demonstrate that magnitude of VS integration (i.e., area-under-the-curve in the CDF difference wave) was a strong predictor of spatial aspects of gait (i.e., pace factor). Magnitude of VS integration was not associated with temporal aspects of gait performance (rhythm), including swing time variability. The fact that magnitude of VS integration was however associated with stride length variability, does demonstrate a two-level dissociation between VS integration and spatial aspects of gait that is in keeping with our initial hypothesis.

In an effort to unpack the association of VS integration with spatial aspects of gait, we compared participants with poor or deficient multisensory integration abilities (n = 140) to those participants with superior or good multisensory integration abilities (n = 193). Results revealed that participants with good or superior VS integration maintained significantly faster gait velocity (103.55 vs. 95.93 cm/s; p = 0.001); longer strides (119.84 vs. 113.25 cm; p = 0.002); less percentage of gait cycle spent in double support (31 vs. 33%; p = 0.001) and less stride length variability (3.46 vs. 4.03 SD units: p = 0.01) compared to those with poor or deficient VS integration. While this information is helpful in characterizing the various spatial facets of gait, there are clear advantages to the application of a principal component approach when analyzing quantitative gait data.

Our finding that increased VS integration is linked to better goal-directed locomotion is directly in line with our hypothesis and likely a result of both processes activating similar neural circuitry. Multisensory integration effects have been linked to cortical [frontal/motor/primary sensory areas/ superior temporal sulcus (STS)] and subcortical (superior colliculus/thalamus) regions in cats, primates, and humans (Meredith and Stein, 1986; Stein et al., 2002; Calvert et al., 2004). The lack of an association between VS integration and rhythm in our study could potentially be related to the fact that the temporal aspects of gait, emanating from brainstem and spinal networks, are less active during early, basic VS processing. While reports have indicated that multisensory inputs from brainstem can affect cortical integration processes, it is clear that the brainstem is primarily concerned with the temporal and spatial attributes of the sensory inputs, and thus the brainstem's role is more involved with modulation of information rather than information processing (Calvert et al., 2004).

Successful functioning and mobility in the real world rely on efficient multisensory integration processes that utilize feedback and feedforward neuronal loops between primary sensory, multisensory, and subcortical regions (see Calvert et al., 2004; Schroeder and Foxe, 2004; Meyer and Noppeney, 2011; Wallace, 2012). The thalamus plays an important role in the integration of sensory information, through corticocortical and cortical-subcortical transmissions (Sherman, 2005). TABLE 3 | (A–E) Summary of linear regression models for predicting gait factors and/or variables.


#### Model summary



#### Model summary




#### Model summary



#### Model summary



TABLE 3 | Continued Model summary


Cortico-cortical and cortico-thalamic loops required for intact multisensory integration and mobility outcomes like balance and gait are notoriously compromised with aging. It is therefore logical that a disruption in shared neural circuitry, resulting from normal aging, disease, or any other potential variable, could adversely impact all processes relying on the functional and structural integrity of said circuit.

In an attempt to highlight the clinical significance of these findings, it should be noted that Verghese and colleagues posit that each 10 cm/s decrease in gait velocity is associated with a 7% increased risk for falls in our study populations (Verghese et al., 2009). The difference in gait velocity between the superior (105.46 cm/s) and deficient integrators (94.47 cm/s) was nearly 11 cm/s. Additionally, we recently revealed the clinical relevance of multisensory integration in aging in the context of balance and fall prediction (Mahoney et al., 2018) and our results indicate that older adults with superior VS integration abilities maintain: (1) better balance performance on the unipedal stance test (16.43 s) compared to deficient integrators (12.57 s) and (2) reduced occurrence of falls compared to deficient integrators for both prevalent (17 vs. 30%) and incident (42 vs. 80%) falls. Our initial studies highlight the significant association of VS integration (i.e., RT facilitation effect) with balance, falls, and physical activity level. However, the directionality of this association was seemingly paradoxical, where larger RT facilitation was associated with worse balance and increased falls (Mahoney et al., 2014, 2015). While a significant association between VS integration and balance and falls still remains (Mahoney et al., 2018), we posit that the directionally of this association is likely influenced by methodological modifications which included a new operational definition of VS integration based on magnitude of race model violation (not RT facilitation) and avoidance of data-trimming procedures that reportedly skew the CDF (Gondan and Minakata, 2016).

In terms of study limitations, a healthy young control group was purposefully excluded given known alterations in unisensory

# REFERENCES


processing with increasing age. The ability to image the brain in motion is essential to determine the actual neural networks associated with the independent gait factors of pace, rhythm, and variability; hopefully continued advances in technology will afford the opportunity to launch this investigation sooner rather than later. Lastly, while overall cognitive functioning as measured by the RBANS was not significantly different between groups, it is possible that better VS integration is associated with better cognition, which could in turn influence the relationship of VS integration with spatial aspects of gait. Future studies should aim to determine the impact of cognition or cognitive status on the association of VS integration and various motor outcomes.

In conclusion, we provide support for the association of increased VS integration with increased gait performance, particularly with regard to spatial aspects of gait (pace) for older adults. Our main finding reveals that deficits in VS integration are linked to slower gait speed, shorter strides, and increased percentage of gait cycle spend immobilized with two feet on the ground (double support %). Additionally, worse VS integration was associated with increased stride length variability which has already been linked to increased fallrisk for older adults. Therefore, the current study continues to provide support for the notion that inefficient multisensory integration may be a potential novel mechanism for falls in older adults.

# AUTHOR CONTRIBUTIONS

JM: study concept and design, acquisition of participants and data, interpretation of data, grant support and manuscript preparation. JV: study concept and design, grant support, and preparation of manuscript.

# FUNDING

This work was supported by the National Institute on Aging at the National Institute of Health (K01AG049813 to JM), (R01AG044007 to JV), and (R01AG036921 to Dr. Roee Holtzer). Additional funding was supported by the Resnick Gerontology Center of the Albert Einstein College of Medicine.

# ACKNOWLEDGMENTS

Special thanks to our participants and our research assistants for their outstanding assistance with this project.


**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 Mahoney and Verghese. 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.

# Effects of Aging on Postural Responses to Visual Perturbations During Fast Pointing

Yajie Zhang1,2\*, Eli Brenner <sup>1</sup> , Jacques Duysens <sup>3</sup> , Sabine Verschueren<sup>2</sup> and Jeroen B. J. Smeets <sup>1</sup>

<sup>1</sup>Department of Human Movement Sciences, Vrije Universiteit Amsterdam, Amsterdam Movement Sciences, Amsterdam, Netherlands, <sup>2</sup>Department of Rehabilitation Sciences, FaBer, KU Leuven, Leuven, Belgium, <sup>3</sup>Department of Kinesiology, FaBer, KU Leuven, Leuven, Belgium

People can quickly adjust their goal-directed hand movements to an unexpected visual perturbation (a target jump or background motion). Does this ability decrease with age? We examined how aging affects both the timing and vigor of fast manual and postural adjustments to visual perturbations. Young and older adults stood in front of a horizontal screen. They were instructed to tap on targets presented on the screen as quickly and accurately as possible by moving their hand in the sagittal direction. In some trials, the target or the background moved laterally when the hand started to move. The young and older adults tapped equally accurately, but older adults' movement times were about 160 ms longer. The manual responses were similar for the young and older adults, but the older adults took about 15 ms longer to respond to both kinds of visual perturbations. The manual responses were also less vigorous for the older adults. In contrast to the young adults, the older adults responded more strongly to the motion of the background than to the target jump, probably because the elderly rely more on visual information for their posture. Thus, aging delays responses to visual perturbations, while at the same time making people rely more on the visual surrounding to adjust goal-directed movements.

#### Edited by:

Paolo Cavallari, University of Milan, Italy

#### Reviewed by:

Patrizia Fattori, Università degli Studi di Bologna, Italy Pedro Ribeiro, Universidade Federal do Rio de Janeiro, Brazil

> \*Correspondence: Yajie Zhang y3.zhang@vu.nl

Received: 23 August 2018 Accepted: 21 November 2018 Published: 04 December 2018

#### Citation:

Zhang Y, Brenner E, Duysens J, Verschueren S and Smeets JBJ (2018) Effects of Aging on Postural Responses to Visual Perturbations During Fast Pointing. Front. Aging Neurosci. 10:401. doi: 10.3389/fnagi.2018.00401 Keywords: postural control, goal-directed, reaching, visual information, target, background, elderly, adjustment

# INTRODUCTION

Reaching out for objects while standing happens often in many daily life situations, such as when preparing a meal. In such situations it is essential to account for the forces that accompany reaching out so that they do not disturb one's balance. This is achieved through anticipatory postural adjustments (Bouisset and Zattara, 1987; Massion and Dufosse, 1988; Aruin and Latash, 1995). Maintaining balance is not only essential because one does not want to fall, but also because allowing balance to be disturbed will challenge the accuracy of the endpoint of the reaching movement (Berrigan et al., 2006). As one gets older, maintaining balance when reaching forward while standing becomes more difficult (Hageman et al., 1995). Do such effects of aging influence the control of goal-directed movements?

People rely on continuously updated sensory information to rapidly adjust goal-directed movements (Cluff et al., 2015; Smeets et al., 2016). Such information comes from vision (Franklin and Wolpert, 2008; Oostwoud Wijdenes et al., 2013), the vestibular system (Keyser et al., 2017) and the somatosensory system (Lowrey et al., 2017). The adjustments' latencies depend on the kind of sensory input. The arm takes between 100 ms and 160 ms to respond to a visually perceived target jump (Brenner and Smeets, 1997; Gritsenko et al., 2009; Oostwoud Wijdenes et al., 2013; reviewed by Smeets et al., 2016) or background motion (Brenner and Smeets, 1997; Whitney et al., 2003; Gomi et al., 2006). Even when adjusting reaching movements in response to such visual perturbations, postural responses can precede the hand's response (Zhang et al., 2018). Does this ability to adjust movements decrease with age? The problems in balance control that develop during aging, combined with weaker muscles (Doherty, 2003) and poorer visual sensitivity and processing speed (Fiorentini et al., 1996; Owsley, 2011; Habekost et al., 2013) suggest that responses might become less vigorous and have longer latencies, both for target jumps and background motion.

Little is known about how aging affects the vigor of responses. Aging could reduce vigor because the muscles become weaker (Goodpaster et al., 2006) due to an age-related loss of spinal motor neurons and motor units, which reduces muscle fiber number and cross-sectional area (Booth et al., 1994). However, it has been reported that, older adults move less vigorously, irrespective of task difficulty in Fitts' Task (Temprado et al., 2013). Therefore, the vigor of hand responses might be constrained by processing the information of the ongoing hand movement rather than by muscle strength. For postural responses, it is relevant that aging is associated with a reduced sensitivity of the proprioceptive (Skinner et al., 1984) and vestibular systems (Anson and Jeka, 2016). Therefore, we expect that older adults will rely more on vision of their surrounding when performing goal-directed movements (Coats and Wann, 2011; Chancel et al., 2018), and thus possibly show more vigorous manual responses to background motion, because manual responses to background motion may also be corrections for assumed self-motion (Gomi, 2008). Therefore, it is interesting to investigate the effect of aging on the timing and vigor of various responses to visual perturbations and to determine whether the effects are related to the general slowing of the movement.

Aging has been reported to delay the onset of fast responses to sudden visual perturbations: hand movement adjustments to target jumps and to background motion take about 20 ms longer in older adults (Kadota and Gomi, 2010; Kimura et al., 2015). It has been argued that these reflexive adjustments are essential for guiding the hand accurately to its target (Scott, 2016; Smeets et al., 2016), so a delayed response in older adults would decrease their accuracy. Additionally, larger postural sway in older adults when standing (Baloh et al., 1994; Blaszczyk et al., 1994; Laughton et al., 2003) may affect the accuracy of the endpoint of the reaching movement (Berrigan et al., 2006). A way to compensate for this reduced accuracy is by increasing the movement duration. There is indeed evidence that older adults move more slowly to maintain accuracy (Goggin and Meeuwsen, 1992; Temprado et al., 2013). We therefore test whether the longer adjustment latencies are related to longer movement times with increasing age.

In this study, we apply lateral visual perturbations (either target jump or background motion) while standing participants make forward reaching movements. The aim of the study is to investigate the effects of aging on responses to such sudden visual perturbations during an on-going reaching movement. The perturbations evoke responses in the goal-directed arm movements, so participants need to adjust their posture as well. We therefore also examine adjustments to the head and trunk.

# MATERIALS AND METHODS

# Participants

Sixteen young adults (28 ± 3 years, seven males) and 16 older adults (74 ± 4 years, nine males) participated in this study. They were all right-handed, had normal or correctedto-normal vision, and had no disease that is known to affect motor or sensory function. The study was approved by the Research Ethics Committee of the Faculty of Behavioral and Movement Sciences, Vrije Universiteit Amsterdam (no. VCWE-2016-176R1). Written informed consent was obtained from each participant.

# Experimental Setup and Procedure

The setup is identical to that used in previous research in our lab (Zhang et al., 2018). Participants stood in front of a horizontal screen (60 Hz refresh rate, 91.9 × 51.6 cm, 1,920 × 1,080 pixel resolution) lying flat, face-up on a heightadjustable table (**Figure 1A**). They stood barefoot with their feet separated by about 10% of their height, 15 cm from the near edge of the screen. Table height was adjusted to align the screen with the participant's hip.

An Optotrak 3,020 motion capture system (Northern Digital, Waterloo, ON, Canada) sampling at 200 Hz was used in the experiment, with a camera located to the right of the participant and another located behind the participant. A photodiode was attached to the far-right corner of the screen to help synchronize the target's appearance and when the target changed position or the background started to move with the movement measurements (to within 5 ms). The posture was recorded with customized cluster markers: three markers attached rigidly to each other in a triangular configuration. Cluster markers were attached to the forehead, 3rd thoracic vertebra (referred to as ''upper trunk''), 1st sacral vertebra (referred to as ''lower trunk'') and the wrist (ulnar side). A single marker was attached to the nail of the index finger of the right hand. This marker was used to control the experiment and analyze the movement of the finger.

The timeline of one trial is shown in **Figure 1B**. A target appeared at a random time between 0.6 s and 1.2 s after the participant placed the right index finger at the starting point. The participant was instructed to tap on the target as accurately and fast as possible with the tip of the right index finger. As soon as the participant started moving towards the target, a visual perturbation (either target jump or background motion) occurred in 80% of the trials. Due to delays in measuring the movement of the finger and rendering images on the screen, the perturbation occurred 60 ms after the finger had moved 5 mm from the starting point. If the target was hit (i.e., if the contact position of the finger was within the target), a sound indicated

success. Otherwise, the target drifted away from where the finger touched the screen.

There were nine conditions in 300 fully randomized trials: one condition with no perturbation (60 trials), and eight conditions with a perturbation (30 trials each). The eight conditions resulted from all combinations of two kinds of perturbation (target jump or background motion), two directions (left or right) and two magnitudes (small or big). The checkerboard-like background (square length: 7 cm) was always present (**Figure 1C**). In the target jump conditions, the target was displaced by either 1 or 4 cm, leftwards or rightwards, across a stationary background. In the background motion conditions, the background moved continuously either leftwards or rightwards at 20 or 60 cm/s, ''behind'' the stationary target. Before the 300 trials of the experiment, the participants practiced for about 20 trials (random conditions). During the experiment, they could rest at any time between trials by delaying placing their finger at the starting point.

In order to be able to judge whether the two age groups differed in their physical ability to reach while standing, we determined the functional reach ratio (the functional reach distance (Weiner et al., 1992) divided by the individual's height) before the experiment. Participants stood normally with their feet about shoulder width apart, close to a wall, with the arm that was closest to the wall pointing forward (90◦ of shoulder flexion). They were instructed to lean forward from this position to reach as far as possible without lifting their heels. A yardstick attached to the wall at the level of the shoulder was used to determine the horizontal distance between the initial and farthest position of the participants' right fingertip. The maximal reach distance of three trials was considered the functional reach distance.

# Data Analyses

The data analysis was similar to that in our previous study (Zhang et al., 2018), with in addition comparisons involving the two age-groups using two-way analysis of variance (ANOVA) and ttests, and an analysis of the correlation between response latency and movement time.

The 3D kinematic data of all markers were filtered using a second order low-pass Butterworth filter with a cut-off frequency of 30 Hz. We determined this cut-off frequency by determining the minimum variance in the distances between the three markers on a cluster (Schreven et al., 2015). We excluded trials (5%) for which the trial duration or the delay in presenting the perturbation was not within ±3 SD of the mean, or for which the moment of the perturbation could not be determined properly (on the basis of the signal picked up by the photodiode).

# Dependent Measures

As a measure of accuracy, we defined tapping error as the distance between the endpoint of the movement and the target center. Movement time was determined for each trial as the time from when the finger started moving (finger lifted higher than 5 mm) until it tapped on the screen (i.e., a trial ends). When using movement time as a measure of how fast a participant moved, we averaged the movement time across all nine conditions.

The focus of our study is on the online adjustment to the perturbations that occurred during the movements. As the perturbations were always perpendicular to the main (sagittal) movement direction, we only analyzed the lateral component of the participants' movements. We did so for the finger, wrist, head, upper trunk and lower trunk. The lateral velocity of the finger was calculated from the measured position data using the central difference algorithm. Responses for each participant were determined by taking the difference in average lateral velocity between trials with a rightward and trials with a leftward perturbation and divided this difference by two. The resulting ''lateral response'' is positive if it is in the direction of the perturbation. The magnitude of the peak velocity was determined for each age group (young and older) and perturbation type (target jump and background motion) by averaging the peak values of the individual mean responses across participants. These values will be close to the peaks in the lateral response if the timing of the responses is consistent across participants.

The response latency was determined by an extrapolation method: the time at which a line through the points at which the lateral response reached 25% and 75% of the peak response intersected the baseline (no response) value (**Figure 1B**; Veerman et al., 2008). We use the slope of this line (acceleration) as our measure of the vigor of the response. We defined time zero as the moment at which the perturbations actually happened on the screen. The baseline value was the average response from 50 ms before to 50 ms after this moment.

The extrapolation method requires a clearly identifiable peak. As the lateral response is very modest with respect to the spontaneous trial-to-trial variability for body parts other than the finger, it had multiple peaks for some participants, so it was impossible to reliably identify response peaks for all individual participants. We therefore determined the latencies from the average response of all participants. We bootstrapped (DiCiccio and Efron, 1996) the trials within each participant to obtain a measure of reliability (resampled with replacement). We averaged the resampled responses of all participants and determined the latency for the average response. Doing so 1,000 times provided a distribution of latencies based on

dots in each panel, one for the target jump (red) and the other for background motion (blue). The red curve in (A) indicates the vigor of a minimal jerk movement adjustment in the time between the onset of the adjustment until the tap. Note that the vigor axis has a different scale in the two panels. The negative values for the vigor in the right panel correspond to participants with head responses in the direction opposite to the target jump.

resampled trials, which we used to determine a Bayesian 95% credible interval. We performed the data-analysis on all participants. As we used the same data for the young participants as in our previous article, this yielded exactly the same results, except for the results of the bootstrapping which involves a random factor in the resampling.

### Statistics

Descriptive data are shown as means or means ± SD across participants. As the initial response (and thus the latency) is independent of perturbation amplitude (Zhang et al., 2018), the results are averaged across the two perturbation amplitudes for all analyses except for the plots of the lateral response as a function of time from the perturbation. A 2 × 3 two-way ANOVA was used to test the effects of age (young and older adults; between participants) and perturbation type (no perturbation, target jump and background motion; within participants) on movement time. As we cannot determine a response for the ''no perturbation'' trials, a similar 2 × 2 ANOVA excluding the ''no perturbation'' type was used to test the effects of aging and perturbation type on finger response latency. The relationship between response latency and movement time was evaluated with a Pearson correlation. Bayesian 95% credible intervals were determined for the average response latencies across all participants. The tapping error, the accuracy and the functional reach ratio of the young and older groups were compared using t-tests. P < 0.05 was considered as significant.

# RESULTS

Both age groups performed the task well (success rate above 95%). The average tapping error was similar for both groups across all conditions: 1.46 ± 0.10 cm for the young adults and 1.41 ± 0.07 cm for the older adults. The functional reach ratio

was slightly lower in the older group (young: 22.7% ± 3.9%, older: 19.9% ± 3.7%, p = 0.043).

The average movement times of the older adults was 526 ± 86 ms, much slower than the 383 ± 44 ms for the young adults (F(1,90) = 141.371, p < 0.001). The movement time did not depend on the perturbation type (F(2,90) = 1.343, p = 0.27) and there was no interaction between age and perturbation type (F(2,90) = 0.182, p = 0.83), so we averaged movement time across all nine conditions of each participant and used this average value for the further analysis.

# Manual, Head and Trunk Responses

The first 100 ms of the lateral responses of the finger and wrist were larger for target jumps than for background motion for the young adults, whereas the opposite appears to be the case for the older adults (**Figure 2**). The difference is mainly due to a much weaker response to target jumps for the older adults (red curves) with a similar response as the young adults for background motion. In general, responses to small and large perturbations had very similar latencies but the larger perturbations gave rise to slightly larger response amplitudes. After averaging the responses to the two perturbation sizes, both for target jumps and for background motion (**Figures 2C,F**), it is clear that all manual responses are delayed for the older adults. Aging also reduced the vigor of the response, but much less so for background motion than for target jumps. The wrist may even respond more strongly to background motion for the older adults than for the young adults (filled blue dot in **Figure 2F** is above the open one; also compare blue curves in **Figures 2D,E**).

It is known that the finger responds less vigorously to target jumps when the (remaining) movement time is long (Oostwoud Wijdenes et al., 2011). The vigor of the finger's response was clearly lower when movement time was longer (red dots in **Figure 3A**), with the young adults (open symbols) being responsible for the shorter movement times. For responses to a target jump, we can determine the optimal smooth response given the remaining time, considering the delays in the equipment and the average response latency (Flash and Hogan, 1985). The red curve in **Figure 3A** is the vigor that one would expect for such an optimal response. The overall pattern in the data of both groups (red symbols) is very similar to what one would expect for an optimal smooth response (curve). For the older adults, we see a more vigorous response to background motion than to target jumps (solid blue dots above the red dots). As it is unclear how much one should correct for background perturbations, we cannot make predictions for the vigor of these responses.

FIGURE 5 | Lateral responses of upper and lower trunk as a function of the time after the perturbation. Details as in Figure 2. In the upper right panel, the latency of the response of the young adults' upper trunk to a target jump was 66 ms, which is outside the plotted range.

In line with our previous study (Zhang et al., 2018), the head does not respond clearly to target motion; this was independent of the age (red traces in **Figure 4**). The response to background motion is considerably larger for older than for young adults (compare filled and open blue dots in right panel of **Figure 4**). Unlike the vigor of finger responses (**Figure 3A**), the vigor of head responses to background motion does not decrease with movement time (**Figure 3B**). This is not inconsistent with an explanation in terms of the remaining movement time, as there is no remaining time for the head. The trunk responded to the perturbations in much the same way as the wrist, with older adults having a clearly smaller response to target jumps than young adults, whereas the responses to background motion did not differ (**Figure 5**).

## Response Latency

It is clear that all response latencies were shorter for the young adults than for the older adults (filled symbols higher than open symbols in **Figure 6**). In line with the results of our previous study (Zhang et al., 2018), the response latency was also shorter for responses to target jumps than for responses to background motion (blue symbols higher than red symbols). For the finger, both the effect of age group and that of perturbation type were significant (F(1,60) = 44.6, p < 0.001; F(1,60) = 42.2, p < 0.001) without a significant interaction (F(1,60) = 0.81, p = 0.37). The same was true for the wrist (age: F(1,60) = 44.5, p < 0.001; type: F(1,60) = 6.57, p = 0.013; interaction: F(1,60) = 2.89, p = 0.094). The latency of the older adults' finger responses was 126 ± 9 for the target jump and 137 ± 8 for background motion, 11–14 ms later than those of young adults (112 ± 7 and 126 ± 6, respectively). Their wrist responses were 16–22 ms later (**Figure 6**). A similar trend can be seen for responses of the trunk and head, but it is less clear because of the large variability in the estimated response latencies.

To investigate whether the longer latencies for the older adults could be related to the individual differences in movement time, we plotted the relationship between movement time and finger response latency (**Figure 7**). The response latency was clearly correlated with the movement time, both for background

groups. Error bars show Bayesian 95% credible intervals that were obtained through bootstrapping (1,000 samples). Data for the young adults are reanalyzed from Zhang et al. (2018).

FIGURE 7 | The relationship between finger response latency and movement time. Each participant is represented by two dots, one for the target jump (red) and the other for background motion (blue).

motion (r = 0.783, p < 0.001, slope = 0.071) and for target jumps (r = 0.811, p < 0.001, slope = 0.088), so the longer response latencies for the older adults are in line with their longer movement times.

# DISCUSSION

In this study, we investigated how aging affects the ability to adjust goal-directed movements to sudden visual perturbations (a target jump or background motion). Additionally, we evaluated whether any effects of aging on the adjustments' timing or vigor could be related to effects on other aspects of movement execution, such as movement time. The patterns of responses to target jumps and background motion were similar to those in our previous study (Zhang et al., 2018). The hand and trunk of young adults responded more vigorously to the target jumps than to background motion, whereas those of the older adults had the opposite pattern of responses (**Figures 2**, **5**). Older adults also had longer movement times and longer response latencies. The increase in response latency with age (about 15 ms) is close to previously reported values of 16–17 ms (Kadota and Gomi, 2010) and 20 ms (Kimura et al., 2015) for fast (∼110 ms) responses. A possible explanation for the longer latencies in older adults is sensory slowing. Aging may have negative effects on visual processing speed (Fiorentini et al., 1996; Habekost et al., 2013). An alternative explanation is that the latencies are secondary to a general slowing of movements.

Aging has different effects on the vigor of the various responses. The reduction of vigor with age could be a manifestation of a general slowing process, in which all factors related to force-impulse control could be involved, such as age-related loss of spinal motor neurons and motor units, a decrease in muscle fiber number and cross-sectional area (Booth et al., 1994) and the associated decrease in muscle strength (Goodpaster et al., 2006). We evaluated this by determining the maximal ability in forward reaching without time constraints. As observed in other studies (Duncan et al., 1990; Hageman et al., 1995), the older adults had a slightly lower functional reach ratio. However, as the perturbation was always at the start of the movement, older adults had more time to correct their movement and could therefore use less vigorous responses to achieve an optimally smooth correction (red curve in **Figure 3A**). Longer movement times could thus be the explanation of the less vigorous finger response to target jumps in older adults. If the reduction of the response vigor with age is related to the remaining time to reach the target, rather than with muscle weakness, we should find very little effect of aging on the responses that are not directly related to reaching the goal. This is indeed the case: the vigor of the finger's response to background motion did not decrease as much with movement time (and thus age) as that to target jumps (blue dots in **Figure 3A**), and the vigor of the head responses to target motion even tends to increase with age (red symbols in **Figure 3B**). A similar pattern can be found in the peak velocities of these responses (right panels of **Figures 2**, **4**).

The increased vigor of the head's response to background motion for the older adults (**Figures 3B**, **4**) suggests that the elderly rely more on vision to keep their head stable. Several authors have reported that the elderly rely more on vision to control posture (Jamet et al., 2004; Bugnariu and Fung, 2007; Poulain and Giraudet, 2008; Slaboda et al., 2011; Agathos et al., 2015). This could be because the precision of other senses (e.g., vestibular) deteriorates faster with age, or might be caused by the elderly being less good at ignoring irrelevant information (de Dieuleveult et al., 2017). Haibach et al. (2009) found that although sway was more sensitive to the optic flow in older as compared to young adults, in accordance with a higher reliance on vision, the sensation of self-motion (vection) did not increase in parallel. This suggests that the subconscious use of optic flow may become increasingly important with age independently of the explicit perception of self-motion. How the weight given to sensory information changes with age depends on the task. For instance, Wiesmeier et al. (2015) reported that when the task was to maintain balance on a moving platform, the elderly relied to a greater extent on proprioceptive rather than visual and vestibular cues.

If the manual responses to background motion are unnecessary adjustments for moving the hand to the target as a result of assumed self-motion (Gomi, 2008), then the pattern of responses to background motion that we found (**Figure 3A**)

# REFERENCES


might be a combination of vigor decreasing with increasing movement time in the same way as for target motion, but being larger for the older adults due to an increase in reliance on vision (optic flow) to compensate for sway. If background motion gives rise to compensatory postural adjustments of the hand, head and trunk in order to stabilize the body when confronted with evidence of self-motion (Mergner et al., 2005), the finger's response to background motion may simply be the result of a misplaced postural correction.

Longer adjustment latencies are clearly related to longer movement times, irrespective of perturbation type (**Figure 7**). Since the latency of responses to visual perturbations is independent of the remaining movement time (Oostwoud Wijdenes et al., 2011), it is unlikely that the longer latencies in the elderly are a result of the reduced temporal constraints given the longer movement times. On the other hand, the reduced vigor of the finger's response in the elderly is probably a result of the longer movement time (**Figure 3A**). Assuming that all participants optimized the combination of speed and accuracy as instructed, the movement time is presumably determined on the basis of the quality of the online control. Thus, most of the age-related differences that we found are probably interrelated, probably with the increased response latency as the origin. Longer latencies in feedback loops lead to unstable behavior unless the gains are low (Burdet et al., 2006), so the corrections are less vigorous in the elderly. The longer movement time is a mechanism for compensating for adjustments being less vigorous and having a longer latency (Salthouse, 1979). With a longer movement time the older adults could perform as accurately as the young adults (though not quite as fast).

In conclusion, our study shows that the general slowing effect of aging includes a longer delay in using visual feedback. The study also confirms that older adults rely more on the visual surrounding to control their movements, and therefore are more affected by background motion. The other effects that we found may be secondary to the increased latency of online adjustments.

# AUTHOR CONTRIBUTIONS

YZ, EB and JS designed this study. YZ collected all the data and analyzed them with the help of EB and JS. All authors contributed to the interpretation of the data and writing of the manuscript.

# FUNDING

This research was funded by the European Commission through MOVE-AGE, an Erasmus Mundus Joint Doctorate program (2011–2015).


fast pointing task. Exp. Brain Res. 236, 1573–1581. doi: 10.1007/s00221-018- 5335-y

**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 Zhang, Brenner, Duysens, Verschueren and Smeets. 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.

# Medical, Sensorimotor and Cognitive Factors Associated With Gait Variability: A Longitudinal Population-Based Study

Oshadi Jayakody<sup>1</sup> , Monique Breslin<sup>1</sup> , Velandai Srikanth1,2 and Michele Callisaya1,2 \*

<sup>1</sup> Menzies Institute for Medical Research, University of Tasmania, Hobart, TAS, Australia, <sup>2</sup> Department of Medicine, Peninsula Health, Monash University, Melbourne, VIC, Australia

Background: Greater gait variability increases the risk of falls. However, little is known about changes in gait variability in older age. The aims of this study were to examine: (1) change in gait variability across time and (2) factors that predict overall mean gait variability and its change over time.

Methods: Participants (n = 410; mean age 72 years) were assessed at baseline and during follow up visits at an average of 30 and 54 months. Step time, step length, step width and double support time (DST) were measured using a GAITRite walkway. Variability was calculated as the standard deviation of all steps for each individual. Covariates included demographic, medical, sensorimotor and cognitive factors. Mixed models were used to determine (1) change in gait variability over time (2) factors that predicted or modified any change.

#### Edited by:

Helena Blumen, Albert Einstein College of Medicine, United States

#### Reviewed by:

Karen Zown-Hua Li, Concordia University, Canada Jeannette R. Mahoney, Albert Einstein College of Medicine, United States

> \*Correspondence: Michele Callisaya

michele.callisaya@utas.edu.au

Received: 09 October 2018 Accepted: 04 December 2018 Published: 18 December 2018

#### Citation:

Jayakody O, Breslin M, Srikanth V and Callisaya M (2018) Medical, Sensorimotor and Cognitive Factors Associated With Gait Variability: A Longitudinal Population-Based Study. Front. Aging Neurosci. 10:419. doi: 10.3389/fnagi.2018.00419 Results: Over 4.6 years the presence of cardiovascular disease at baseline increased the rate of change for step length variability (p = 0.04 for interaction), lower education increased the rate of change for DST variability (p = 0.04) and weaker quadriceps strength increased the rate of change for step width variability (p = 0.01). Greater postural sway predicted greater variability on average across the three phases (p < 0.05). Arthritis, a higher body mass index (BMI), slower processing speed and lower quadriceps strength predicted greater mean step time variability (p < 0.05). Arthritis and a higher BMI predicted greater mean step length variability, while slower processing speed and BMI predicted greater mean DST variability (p < 0.05).

Conclusion: Over a nearly 5-year period, variability in different gait measures do not show uniform changes over time. Furthermore, each variability measure appears to be modified and predicted by different factors. These results provide information on potential targets for future trials to maintain mobility and independence in older age.

#### Keywords: gait, gait variability, longitudinal study, cognition, sensorimotor, older age

**Abbreviations:** BMI, body mass index; CVD, cardiovascular disease; DST, double support time; Postural sway (EC), postural sway eyes closed; Postural sway (EO), postural sway eyes open.

# INTRODUCTION

fnagi-10-00419 December 15, 2018 Time: 18:11 # 2

An estimated 30–35% of adults aged 70 and older have abnormal gait (Verghese et al., 2006), increasing the risk of falls, hospitalization and institutionalization (Montero-Odasso et al., 2005; Verghese et al., 2006). Traditionally, changes in gait speed are used as markers of gait dysfunction. However, a growing body of literature has investigated intra-individual gait variability, the fluctuation in the value of a gait parameter from one step to the next (Callisaya et al., 2010). Gait variability is potentially a more sensitive predictor of adverse events such as falls (Brach et al., 2005; Verghese et al., 2009; Callisaya et al., 2011). The sheer increase in the global older population (Department of Economic and Social Affairs, 2015) and the high prevalence of gait impairments make understanding how gait variability changes in older age and what factors might predict this change an important topic for investigation.

Evidence on whether age is associated with gait variability is limited to cross-sectional studies. Most studies have compared gait variability between younger and older people (Gabell and Nayak, 1984; Hausdorff et al., 1997; Stolze et al., 2000; Grabiner et al., 2001; Menz et al., 2003; Owings and Grabiner, 2004a,b; Woledge et al., 2005; Kang and Dingwell, 2008; Beauchet et al., 2009), reporting either no age-related differences (Gabell and Nayak, 1984; Hausdorff et al., 1997), or greater variability in spatial (Grabiner et al., 2001; Owings and Grabiner, 2004a,b; Woledge et al., 2005; Kang and Dingwell, 2008; Beauchet et al., 2009) and temporal measures (Menz et al., 2003; Kang and Dingwell, 2008) in older age groups. In studies of just older people, advancing age was associated with greater spatial (Helbostad and Moe-Nilssen, 2003; Callisaya et al., 2010; Verlinden et al., 2013) and temporal variability (Hausdorff et al., 2001b; Callisaya et al., 2010; Verlinden et al., 2013; Kirkwood et al., 2016). However, the majority of studies are from small samples of volunteers or older adults from geriatric or rehabilitation clinics (Gabell and Nayak, 1984; Hausdorff et al., 1997; Stolze et al., 2000; Owings and Grabiner, 2004a; Woledge et al., 2005; Kang and Dingwell, 2008; Beauchet et al., 2009), limiting generalizability.

Longitudinal studies would assist in better understanding the role of aging on gait variability. Furthermore, despite crosssectional associations between poorer physical (Hausdorff et al., 2001a; Brach et al., 2008a; Kang and Dingwell, 2008; Callisaya et al., 2009) and cognitive function (Holtzer et al., 2012b; Martin et al., 2012; Beauchet et al., 2014) with greater gait variability, no studies to our knowledge have examined the factors that may modify longitudinal changes in gait variability. Such information is clinically important in determining individuals at increased risk of declining gait and hence adverse outcomes such as falls.

Therefore, the aims of this study are in a populationbased sample of older people: (1) to examine the longitudinal associations between age and a range of temporal and spatial gait variability measures; (2) to examine the demographic, medical, sensorimotor and cognitive factors that predict overall mean gait variability and its change over time.

# MATERIALS AND METHODS

# Study Participants

The Tasmanian Study of Cognition and Gait (TASCOG) is a population based longitudinal study of gait, cognition and brain imaging in older people. A community dwelling sample of adults aged between 60 and 85 years (n = 431) were randomly selected from the Southern Tasmanian electoral roll. Participants were excluded if they were institutionalized, or had any contraindication to MRI (a requirement of the parent study). Individuals were also excluded (**Figure 1**) if they had a history of dementia (n = 3), Parkinson's disease (n = 2), missing gait data (n = 9), used a walking aid (n = 6), or were unable to follow simple commands in English (n = 1). Ethical clearance was obtained from the Southern Tasmanian Health and Medical Human Research Ethics Committee and written consent was obtained from all participants. The inception cohort, assembled from January 2005 (baseline), followed-up in March 2008 (phase 2) and March 2010 (phase 3) used identical methods.

# Gait Assessment

Spatial and temporal measures of gait were determined from the footfalls recorded on the GAITRite system, a 4.6 m computerized walkway, (GAITRite system, CIR Systems, Havertown, PA, United States), with excellent test–retest reliability (Menz et al., 2004). Each participant completed six walks at their preferred speed. To allow for acceleration and deceleration participants walked 2 m before and after the walkway. Gait speed (cm/s) was directly obtained from the GAITRite software and intraindividual variability of 2 spatial (step length, step width) and 2 temporal (step time, DST) gait measures were calculated as the standard deviations (SD) of all steps over the six walks as previously described (Callisaya et al., 2010). Gait variability is commonly quantified as either the SD or the coefficient of variation [CoV = (SD/mean) × 100]. Here we use the SD in order to report change in variability in each of the gait measures' original units. Variability in these gait measures were selected as they have previously been found to be associated with advancing age (Kang and Dingwell, 2008; Beauchet et al., 2009; Callisaya et al., 2010) and risk of falls (Hausdorff et al., 2001b; Brach et al., 2005; Callisaya et al., 2011), and represent both spatial and temporal parameters in the sagittal and frontal planes (Callisaya et al., 2010; Martin et al., 2012). Variability in stance and swing time, although having been examined in prior cross sectional analyses (Hausdorff et al., 1997; Beauchet et al., 2009; Kirkwood et al., 2016), were not included due to high correlations with step time variability (stance time variability r = 0.86; swing time variability r = 0.80). Single support time was also not considered since it is the opposite of DST.

# Baseline Covariates

The following demographic, medical, sensorimotor and cognitive factors were assessed at baseline. Demographic variables included age, sex and level of education (summarized into a binary variable

using high school level education and below as the cut off). Height (m) and weight (kg) were recorded to calculate the BMI.

# Medical History

Presence of lower limb arthritis and CVD (hypertension, hypercholesterolemia, ischemic heart disease, stroke and diabetes mellitus) were recorded with a standardized questionnaire. CVDs were grouped into a summary binary variable based on the presence or absence of any CVD. Mood was assessed using the Geriatric Depression Scale (short version). Participants were classified as depressed, based on a score of >5.

# Sensorimotor Factors

Postural sway, quadriceps strength, edge contrast sensitivity and proprioception were measured with the short form of Physiological Profile Assessment (Lord et al., 2003).

(1) Postural sway: measured on a foam mat for 30 s with the eyes open (EO) and eyes closed (EC) [sum of maximum medial-lateral and anterior-posterior sway (mm); no upper limit].

(2) Quadriceps strength: the maximal isometric quadriceps strength of the dominant leg (kg) measured in sitting (up to 100 kg; >30 kg is considered excellent).

(3) Edge contrast sensitivity: an indicator of visual contrast sensitivity [measured using the Melbourne edge test (dB); range 0–24].

(4) Proprioception: perception of joint and body segments or movement in the space (Sherrington, 1906) (measured with a lower limb matching task using a vertical clear acrylic sheet placed between the seated participant's legs; no upper limit; values of <1 degree considered good).

(5) Grip strength was quantified with a bulb dynamometer as the average of two measurements of dominant and of nondominant hand (pounds per square inch).

# Cognitive Function

The following tests were used to measure cognition: (a) Executive function: the Controlled Word Association Test (as many words as possible in 1 min; using the letters F, A, and S), the Victoria Stroop test (to correct for processing speed the difference in time to compete Stroop color test– Stroop word test was used); (b) Processing speed-attention: the Symbol

Jayakody et al. Factors Associated With Gait Variability

Search (range 0–60), Digit Span (range 0–16) and Digit Symbol Coding (range 0–133) of Wechsler Adult Intelligence Scale-III, (c) Visuospatial function: the Rey Complex Figure copy task (range 0–36) and (d) Memory: the Hopkins Verbal Learning Test—Revised [Immediate recall (range 0–36), delayed recall (range 0–12), recognition range (0–12)] and a 20 min delayed reproduction of the Rey Complex Figure copy task (range 0– 36). Raw test scores were grouped and subjected to principal component analyses deriving summary components for domains of executive function, processing speed-attention, memory and visuospatial ability as previously described (Callisaya et al., 2015).

# Data Analysis

STATA (StataCorp LLC, College Station, TX, United States) version 15.0 was used in all the analyses.

#### Changes in Gait Variability Over Time

Longitudinal associations between time and gait variability were examined using linear mixed effects models. The residual distributions of the step time and DST conditional models, when assessed for normality, showed positive skewness. Therefore, step time variability was transformed using 1/Y<sup>1</sup> and DST variability by 1/(Sqrt(Y), which were chosen based on the results of a likelihood maximization procedure (STATA boxcox). Variables were back transformed for presentation of results. The model building procedure was as follows. Models were firstly adjusted for a priori confounders baseline age, sex and education. Interaction terms between time and each baseline covariate were then tested in separate models to determine if the covariate modified changes in each gait variability measure over time. Next, each covariate was tested individually to determine whether they were associated with gait variability over time. Finally, we built models using the significant interactions and predictor covariates from individual models in a stepwise fashion, with variables retained only if they remained significant. Although not a major aim of this study, but to allow for comparisons with other studies, models were also built to assess longitudinal changes in gait speed and the other absolute gait measures.

There was considerable participant attrition between baseline and the follow-up phases. Linear mixed effects models are able to provide an unbiased estimate of the regression coefficients in the case of such attrition, provided the reasons for dropout depend only on the observed data (Little, 1995). There was no reason to believe otherwise in the case of this data. Attrition was found to depend on some outcome measures at baseline, and provided the model is correctly specified in relation to these variables, estimates of coefficients are unbiased.

# RESULTS

# Sample Characteristics

**Table 1** summarizes participant characteristics (n = 410).

TABLE 1 | Baseline characteristics of the study sample (n = 410).


This table summarizes the demographic, medical, sensorimotor, gait and cognitive characters of the sample at baseline.

SD, standard deviation; BMI, Body Mass Index; CVD, cardiovascular disease; EO, eyes open; EC, eyes closed; COWAT, Controlled Oral Word Association Test; mm, millimeter; kg, kilograms; ms, milliseconds; cm, centimeter; dB, decibel; psi, pounds per square inch.

# Longitudinal Associations of Gait Variability Over Time

Significant increases, independent of baseline age, sex and level of education, were seen in step length variability (β 0.028 95%CI 0.004 to 0.052; p = 0.02), resulting in an increase of 0.14 cm over 5 years and in DST variability (β 0.223 95%CI 0.091 to 0.355; p = 0.001), indicating an increase of 1.12 ms over 5 years. The findings for variability of step time (β 0.085 95%CI −0.039

to 0.208; p = 0.18, corresponding to an increase of 0.43 ms over 5 years) and step width (β −0.001 95%CI −0.018 to 0.017; p = 0.94, a decrease of 0.005 cm over 5 years) were nonsignificant.

# Modifiers and Predictors of Gait Variability

For step length variability the interaction between CVD and time was significant (p-value for interaction = 0.03), indicating greater increases over time in the presence of baseline CVD. For DST variability the interaction between education and time was significant (p = 0.01), indicating greater increases over time in people with lower levels of education. Although step width variability did not increase over time on average, the interaction between quadriceps strength and time was significant (p = 0.02), indicating greater increases in those with weaker quadriceps muscles. None of the interactions for step time variability were significant (p > 0.05). **Table 2** shows the effect of each of these interactions on the time coefficient. **Table 3** shows the associations between factors tested one at a time with gait variability, adjusted for time terms and a priori confounders (age, sex, and education).

**Table 4** shows final models for the four gait variability outcomes. The model for step length variability included a CVD × time term (p = 0.04), arthritis (p = 0.01), postural sway (EC) (p < 0.001) and BMI (p = 0.04). The model for DST variability included an education × time term (p = 0.04), postural sway (EC) (p = 0.002), BMI (p < 0.001) and processing speed (p < 0.003). The final model for step width variability included a quadriceps strength × time term (p = 0.01) and postural sway (EC) (p < 0.001). Although step time variability did not change over time, greater BMI (p = 0.03), arthritis (p < 0.001), lower quadriceps strength (p = 0.02), greater postural sway (EC) (p < 0.001) and slower processing speed at baseline (p < 0.001) were associated with greater mean variability over the three phases.

# Longitudinal Changes and Factors Associated With Gait Speed and Other Temporal and Spatial Measures

Gait speed significantly decreased over time independent of baseline age, sex and level of education (β −1.159 95%CI −1.502 to −0.816; p < 0.001). Step length shortened (β −0.685 95%CI −0.811 to −0.560; p < 0.001), while DST (β 6.582 95%CI 5.285 to 7.879; p < 0.001) and base of support (β 0.139 95%CI 0.097 to −0.182; p < 0.001) significantly increased over time. Step time did not change (β −0.196 95%CI −1.025 to 0.632; p = 0.642) over the 4.6 years. The results of the final models for gait speed and the other absolute gait measures are presented in **Supplementary Table S1**.

# DISCUSSION

This is the first study, to our knowledge, to undertake a longitudinal analysis of gait variability in a population-based sample of older people. Greater increases in variability were seen in people with CVD at baseline for step length, low levels of education for DST, and those with weaker quadriceps strength for step width. Furthermore, a number of baseline factors were associated with higher variability on average over the 3 phases. Greater postural sway (EC), BMI and arthritis predicted higher step length variability. Greater postural sway (EC), BMI and slower processing speed predicted higher DST variability. Greater postural sway (EC) predicted greater step width variability. Although step time variability did not increase over time, greater postural sway (EC), greater BMI, arthritis, lower quadriceps strength and slower processing speed predicted greater variability across the three phases. These findings increase knowledge on how gait variability changes in older age and assist in identifying factors that may be developed into strategies to prevent gait impairments among older people.

Few studies have examined the longitudinal changes in intra-individual gait variability. Prior studies have been in small samples of participants with specific diagnoses such as Alzheimer's disease (Wittwer et al., 2010), subcortical vascular encephalopathy (Bäzner et al., 2000) and Huntington's Disease (Rao et al., 2011), and report increases in stride length (Rao et al., 2011; Wittwer et al., 2010) and temporal variability measures (Bäzner et al., 2000; Rao et al., 2011). In our community-based cohort of people without dementia, changes over time were not consistent over the different temporal and spatial measures. Consistent with prior cross-sectional studies, step length (Callisaya et al., 2010; Verlinden et al., 2013) and DST variability (Callisaya et al., 2010; Verlinden et al., 2013) increased over time. This is important as we previously found that step length and DST variability measures, but not step width or step time variability, were linearly associated with increased risk of falls over a 12 months period (Callisaya et al., 2011).

Importantly, factors were identified that modified these associations, and that of step width variability, over time. Those with CVD had greater increases in step length variability. CVD is linked to vascular changes in the brain (i.e., white matter hyperintensities and brain infarcts), that can cause impairments in both cognition (De Groot et al., 2000; Wang et al., 2016) and gait (Rosano et al., 2007; Callisaya et al., 2013; Wang et al., 2016) even in people without dementia. Furthermore CVD, particularly diabetes mellitus, might disrupt peripheral sensorimotor abilities (i.e., lower extremity sensation and vision) (Brach et al., 2008b), resulting in increased step length variability over time. Therefore, the impact of CVD on central (i.e., disruption of pathways important for attention and motor control) and peripheral mechanisms (i.e., sensory loss) may disrupt dynamic balance resulting in the need to alter step length to maintain postural control whilst walking. These findings offer potential avenues of preventing increased gait variability in older age via controlling the advancement of CVD. Lower education levels accelerated increases in DST variability over time. Education is a known proxy for cognitive reserve (Stern, 2009). People with greater cognitive reserve are known to cope better with either age or pathology related changes in the brain (Stern, 2009), opening up the possibility that they are also better able to compensate for brain changes involving gait control (Holtzer et al., 2012a). Cognitive reserve is developed via lifetime exposure to cognitively stimulating experience (i.e., education, occupation, leisure, and

#### TABLE 2 | Longitudinal changes of gait variability over time (n = 410).


This table shows changes in individual gait variability measures per year; any change modified by demographic, medical, cognitive or sensorimotor factors is shown by presenting change over time at high and low values of the baseline covariate.

All models were adjusted for age, sex, and education; due to interactions with time, change depends on CVD status, educational level or quadriceps strength depending on the variability measure. No other interactions effects were significant and are therefore not presented. CVD, cardiovascular disease; kg, kilograms; quad, quadriceps;<sup>∗</sup> meaning the 25th, 50th, and 75th percentiles of quadriceps strength.

TABLE 3 | Associations between medical, sensorimotor and cognitive factors with gait variability in individual models (n = 410); the effect of each baseline covariate on the average of gait measures over 3 phases.


Models are adjusted for age, sex, and education, CVD × time (models for step length variability) and education × time (DST variability) and quadriceps strength × time (models for step width variability); higher scores on executive function indicate worse performance, whereas higher scores indicate better performance for other cognitive tests; CVD, cardiovascular disease; EO, eyes open; EC, eyes closed; BMI, Body Mass Index; kg, kilograms; cm, centimeter; mm, millimeter; dB, decibel. The associations that are significant (p < 0.05) are in bold.

physical activity) (Stern, 2009), thus enhancing these factors throughout life may assist in maintaining better gait control in older age. The significant interaction effect between quadriceps strength and time suggested those with weaker quadriceps had greater increases in step width variability over time. Greater step width variability has cross-sectionally been associated with advancing age (Helbostad and Moe-Nilssen, 2003; Callisaya et al., 2010; Verlinden et al., 2013), and falls (Brach et al., 2005). Our findings suggest that muscle strengthening may be an important target for clinical trials aimed at preventing increases in step width variability over time. We found no change for step time variability over time. A potential explanation may be that both high and low step time variability have been found to be associated with falls (Callisaya et al., 2011), suggesting that perhaps age-related impairments may result in either high or low variability, canceling out any directional change of effect. In summary it appears variability in different gait characteristics do not show uniform age-related changes over time.

Although not modifying change over time, greater postural sway (EC), higher BMI (except for step width variability), arthritis (step length and step time variability), slower processing speed (DST and step time variability) and lower quadriceps strength (step time variability) were associated with greater variability on average over the 3 phases. Postural sway on a foam mat (EC) is a measure of balance and vestibular ability. Walking is a complex balance activity (Woledge et al., 2005) and may

TABLE 4 | Associations between medical, sensorimotor and cognitive factors with gait variability in final models (n = 410).


Final models show the associations baseline covariates with gait variability in the presence of all significant modifiers and predictors.

CVD, cardiovascular disease; EC, eyes closed; BMI, Body Mass Index; quad, quadriceps; kg, kilograms; cm, centimeter; mm, millimeter.

<sup>∗</sup>Meaning the 25th, 50th, and 75th percentiles of quadriceps strength. The associations that are significant (p < 0.05) are in bold.

require a trade-off in the consistency of timing and length of steps in the presence of poorer balance (Callisaya et al., 2009). Arthritis may increase step length variability through increased pain, stiffness (Kang and Dingwell, 2008), reduced strength (Kang and Dingwell, 2008) or balance (Callisaya et al., 2009). Our findings were independent of strength and balance, suggesting that these other impairments may be at play. A potential reason for the associations between BMI and greater variability may be that the body's fat distribution affects balance. However, in our study BMI was a predictor of greater step length, DST and step time variability independent of postural sway, suggesting other mechanisms such as cerebro- (Rosano et al., 2007) or peripheral-vascular disease might be important (Forhan and Gill, 2013). Similar to prior cross-sectional studies, slower processing speed was associated with higher temporal variability (Brach et al., 2008a), but not spatial measures. It is possible that processing speed and temporal variability measures (both related to timing) may have similar underlying neural mechanisms. Atrophy in widespread brain networks (Blumen et al., 2018), as well as white matter hyperintensities (De Groot et al., 2000) and subcortical infarcts (Baune et al., 2009) that may disrupt white matter fibers are associated with processing speed and are also likely to be important for the co-ordination of a consistent gait pattern (Srikanth et al., 2010; Blumen et al., 2018). However, we were unable to determine whether central slowing of processing speed disrupted the timing of gait, or that of peripheral slowing, with both likely to lead to poorer gait control.

Although not a main aim of this study, similar to prior studies we observed that gait speed slowed over time (Atkinson et al., 2007; Callisaya et al., 2013) and this was greater in the presence of arthritis and poorer proprioception. Changes in other absolute gait measures were associated with a multitude of covariates (**Supplementary Table S1**), but these were different from the covariates that modified or predicted the same measures change in variability. For example, greater increases in step length variability over time occurred in the presence of CVD, whereas greater decreases in absolute step length over time occurred in the presence of greater baseline age, arthritis and poorer proprioception. Interestingly, this suggests that absolute and variability measures may represent different constructs in gait.

Our study has a number of strengths. It is one of the only longitudinal studies in the context of gait variability over time, with 4.6 years of follow up. Our sample was randomly selected from the electoral roll, increasing generalizability to the wider community compared to studies of people with specific diseases (i.e., Alzheimer's disease). Gait variability is multifaceted, and we studied a range of temporal and spatial measures in both the sagittal and frontal planes. Further, we carefully controlled for confounders, examined for interactions and built our models by examining the effect of each variable one by one. However, there are a few limitations to be noted. Gait assessment was conducted in an indoor environment, thus gait variability could differ from outdoor walking. We collected data over 27 mean steps (baseline), where some have suggested a minimum of 400 steps are required to determine gait variability (Owings and Grabiner, 2003). However, we carefully considered this and balanced it with unnecessary fatigue. A diagnosis of dementia was by self-report and it is therefore possible that our sample had undiagnosed dementia. Finally, we have a moderate level of participants lost to follow up (39%) which is not uncommon given the longitudinal nature of our study. The use of a mixed effects model means that baseline data for those lost to follow up was able to contribute to the analysis.

# CONCLUSION

fnagi-10-00419 December 15, 2018 Time: 18:11 # 8

Variability in DST, step length and step width increased over time, but only in those with lower educational levels, CVD presence of CVD and weak quadriceps, respectively. In addition, a range of musculoskeletal, cognitive and sensorimotor factors were found to predict greater variability across the three phases. These results provide important information on targets for future clinical trials to maintain mobility and independence in older age.

# AVAILABILITY OF DATA AND MATERIAL

The raw data supporting the conclusions of this manuscript will be made available by the corresponding author, without undue reservation, on reasonable request.

# AUTHOR CONTRIBUTIONS

OJ analyzed and interpreted the data, and wrote the manuscript. MB performed the statistical analysis, involved in drafting the manuscript, and critically revised the manuscript. VS primary

# REFERENCES


investigator of the study, involved in drafting the manuscript and critically revised the manuscript. MC analyzed and interpreted the data, a major contributor in writing the manuscript, and provided important intellectual content. All authors read and approved the final manuscript.

# FUNDING

This research was supported by National Health and Medical Research Council (Grant Number 403000BH), Physiotherapy Research Foundation (Grant Number BH036/05), Perpetual Trustees, Brain Foundation, Royal Hobart Hospital Research Foundation (Grant Number 341M), and ANZ Charitable Trust and Masonic Centenary Medical Research Foundation.

# ACKNOWLEDGMENTS

We thank the study participants for their support.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnagi. 2018.00419/full#supplementary-material



**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 handling Editor declared a past co-authorship with two of the authors MC and VS.

Copyright © 2018 Jayakody, Breslin, Srikanth and Callisaya. 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.

# Differences in Cognitive-Motor Interference in Older Adults While Walking and Performing a Visual-Verbal Stroop Task

#### Bettina Wollesen<sup>1</sup> \* and Claudia Voelcker-Rehage<sup>2</sup>

<sup>1</sup>Department of Human Science, Faculty of Psychology and Movement Science, University of Hamburg, Hamburg, Germany, <sup>2</sup>Sports Psychology, Institute of Human Movement Science and Health, Faculty of Behavioral and Social Sciences, Chemnitz University of Technology, Chemnitz, Germany

Objectives: Studies using the dual-task (DT) paradigm to explain age-related performance decline due to cognitive-motor interference (CMI) which causes DT costs (DTCs) revealed contradictory results for performances under DT conditions. This crosssectional study analyzed whether differences in demographics, physical functioning, concerns of falling (CoF), and other mental factors can explain positive and negative DTCs in older adults while walking in DT situations.

Methodology: N = 222 participants (57–89 years) performed a single task (ST) and a DT walking condition (visual-verbal Stroop task) in randomized order on a treadmill. Gait parameters (step length, step width) were measured at a constant self-selected walking speed. Demographics [age, Mini Mental Status Examination (MMSE)], physical functioning (hand grip strength), CoF [Falls Efficacy Scale International (FES-I)], and mental factors [Short Form-12 (SF-12)] were assessed. An analysis of variance (ANOVA) was used to reveal subgroup differences. A four-step hierarchical regression analysis was conducted to identify which variables determine the DTC.

#### Edited by:

Eric Yiou, Université Paris-Sud, France

#### Reviewed by:

Nils Eckardt, University of Oldenburg, Germany Rahul Goel, Baylor College of Medicine, United States

\*Correspondence: Bettina Wollesen bettina.wollesen@uni-hamburg.de

Received: 16 September 2018 Accepted: 10 December 2018 Published: 09 January 2019

#### Citation:

Wollesen B and Voelcker-Rehage C (2019) Differences in Cognitive-Motor Interference in Older Adults While Walking and Performing a Visual-Verbal Stroop Task. Front. Aging Neurosci. 10:426. doi: 10.3389/fnagi.2018.00426 Results: Three subgroups were identified: (1) participants (n = 53) with positive DTCs (improvements under DT conditions); (2) participants with negative DTCs (n = 60) in all gait parameters; and (3) participants (n = 109) who revealed non-uniform DTCs. Baseline characteristics between the subgroups showed differences in age (F(2,215) = 4.953; p = 0.008; η <sup>2</sup> = 0.044). The regression analysis revealed that physical functioning was associated with positive DTC and CoF with negative DTC.

Conclusion: The results confirmed a huge inter-individual variability in older adults. They lead us to suggest that factors causing performance differences in DTCs needs to be reassessed. Functional age seems to determine DTCs rather than calendric age. Psychological variables particularly seem to negatively influence DT performance.

#### Keywords: aging, dual task performance, walking, cognition, physical functioning, concerns of falling, mental health

**Abbreviations:** ANOVA, analysis of variance; CoF, concerns of falling; CMI, cognitive-motor interference; DT, dual-task; DTC, dual-task costs; MMSE, Mini Mental Status Examination; SPPB, Short Physical Performance Battery; ST, single task.

# INTRODUCTION

Age is associated with sensorimotor change and changes in the musculoskeletal system. In combination or interaction, these age-related changes lead to decrements of locomotor coordination, and they also have an impact on walking performance by decreasing gait stability. Behavioral data, as revealed by the use of biomechanical measurements, showed that effects on the locomotor system are expressed in reduced step length (Scott et al., 2015), gait speed (Verghese et al., 2009; Morrison et al., 2016), as well as increased double support time (Maki, 1997; Verghese et al., 2009; Scott et al., 2015), step length variability (Maki, 1997; Verghese et al., 2009), step width (Maki, 1997), or stumbling (Berg et al., 1997). Moreover, older people have problems adapting their walking abilities at higher gait speeds (e.g., to catch a bus) or while walking over uneven surfaces (Berg et al., 1997). All of these aspects can be described as external perturbations that have a negative impact on postural control and therefore decrease gait stability (for additional definition see the review by van Emmerik et al., 2016) and cause an increased risk of falling (Hausdorff et al., 2001).

In daily situations, the locomotor system needs to integrate sensory information and to coordinate movements according to the situation. Gait performance also depends on sensorimotor and cognitive functions. It is proposed that with increasing age, sensory and motor aspects of walking performance increasingly require cognitive control, attention, and supervision. However, age is associated with reduced cognitive processing efficiency (e.g., decrease in nerve conduction speed, increased lateralization; Hedden and Gabrieli, 2004) and in turn, a decrease in cognitive performance, such as diminished response time, working memory, and processing of multiple tasks. These age-related cognitive changes might affect daily task performance (Stawski et al., 2006). In this context, more and more studies indicate a correlation between age-related declines in the sensory and motor system, as well as in cognitive functioning (Li and Lindenberger, 2002).

The cognitive processing of locomotion in dual- or multitasking situations is measured to identify people's susceptibility to adopt impaired gait patterns, often resulting in an increased risk of falling (e.g., crossing a street while observing traffic flow; Faulkner et al., 2007). Adding a secondary motor or cognitive task typically reduces gait stability due to interference during information processing (Wollesen et al., 2016) measured as reduced movement accuracy and movement coordination (Al-Yahya et al., 2011). An age-related reduction in cognitive performance or cognitive-motor interference (CMI) affects how older people cope with such dual-task (DT) situations in daily life.

# Cognitive-Motor Interference (CMI) During Dual-Task Walking in Older Adults

Walking in our natural environment can be considered a DT scenario that requires increasing cognitive resources with increasing age. The level to which walking performance is affected by CMI is typically expressed as the DT cost (DTC). DTCs are calculated as the percentage of decrements in performance of a DT relative to the performance of a single task (ST). Age-related declines of performance whilst walking in DT situations have been extensively investigated (Li et al., 2001; Hollman et al., 2007; Bock, 2008). For instance, an age-related decline in gait performance has been observed when conducting arithmetic, memory, or visual tasks concurrently to walking (Lindenberger et al., 2000; Beurskens and Bock, 2012). These performance declines in DT walking situations have been considered in light of several theoretical positions (see Wollesen et al., 2017a for an overview). Recent systematic reviews on empirical findings and theoretical models (Lacour et al., 2008; Wollesen et al., 2017a) showed that CMI rises with increasing task complexity of the motor and/or cognitive task and according to individual abilities and resources (Lacour et al., 2008). Moreover, the task domain (stimulus-response mode) was found to be a critical moderator variable (Riby et al., 2004). Hence, task settings including controlled processes (e.g., inhibiting information) or motor components (e.g., carrying a tray) showed more decrements in DT performance in older adults than other task combinations. Moreover, studies indicate that increasing difficulty levels (from DT to multitask-performance or with different task complexities, e.g., from processing speed to executive tasks) also increase the effects of DT on gait decrements (Hall et al., 2011; Venema et al., 2013; Li et al., 2014; Menant et al., 2014; MacLean et al., 2017). However, in contrast to previous research, a study by Plummer-D'Amato et al. (2012) failed to show effects of different cognitive loads on DT walking performance. They only found an effect for different walking conditions (comfortable vs. fast vs. obstacle crossing). Further, older adults often reveal higher DTC than young adults (Lindenberger et al., 2000; Beurskens and Bock, 2012). Some studies reported inconsistent results (Muir-Hunter and Wittwer, 2016) or even less DTC in older than in younger adults for DT walking conditions, where the cognitive tasks did not require visual attention (e.g., walking with a spelling task; Bock, 2008).

It still remains unclear which individual factors or resources might explain DTCs or decrements in daily situations that require the management of different simultaneously performed tasks when walking speed remains constant. Several factors have been discussed that might influence CMI of older adults. Possible influencing factors include complexity of the motor and/or cognitive task and the task domain, age-related motor or cognitive declines (Muhaidat et al., 2013), task prioritization (posture first hypothesis), and previous falls or concerns of falling (CoF; Ambrose et al., 2013).

The Task Prioritization Model (Yogev-Seligmann et al., 2010) accounts for the individuals' strategies used during DT performance. It proposes that older adults prioritize motor performance, if the motor task may induce loss of balance (Brown and Bennett, 2002; Chapman and Hollands, 2007). This prioritization is used to compensate CMI and to reorganize the cognitive-motor resources (Li and Lindenberger, 2002) or to reduce the risk of falling. Yogev-Seligmann et al. (2012) found that older adults with adequate balance abilities and capacity to identify hazards are able to focus on cognitive performance as long as balance is maintained. This result was discussed as older adults prioritizing walking over memorizing to protect themselves from falls, a view known as ''posture first hypothesis'' (Shumway-Cook and Woollacott, 2000; Schaefer and Schumacher, 2011; see Li et al., 2013; for discussion of mixed results).

Being a faller has also been shown to influence gait performance, such as step width and step length (e.g., Barak et al., 2006; Lindemann et al., 2008; Nordin et al., 2010; Kirkwood et al., 2011), as well as DTC. Fallers are often not able to shift attention to the motor task in DT situations (Schaefer and Schumacher, 2011). Furthermore, the combination of high-risk task settings (e.g., elevated surface) and a secondary task also leads to problems of task prioritization in healthy older adults (Schaefer et al., 2015).

Nevertheless, there is some evidence that older adults with a reduced postural reserve (motor abilities to maintain balance) have more decrements of gait performance regardless of their cognitive performance in ST and DT situations (Holtzer et al., 2014). Most of the studies focusing on fall prevention report higher decrements of gait parameters in fallers, including gait speed, step length, step width, and double support time (Maki, 1997; Beauchet et al., 2009; Muhaidat et al., 2013). These changes apply especially in situations that require adapting a faster gait speed (Barak et al., 2006). Declines are associated with an increased risk of falling (Beauchet et al., 2009; Menant et al., 2014). Furthermore, fallers have poorer motor precondition (e.g., reduced physical fitness or muscle strength; Freire Júnior et al., 2017). Additionally, studies have reported that older adults at risk of falling had poorer mobility judgment in a virtual reality DT walking situation (crossing a street while listening to music or writing messages) and therefore experienced more collisions with oncoming cars (Nagamatsu et al., 2011; Neider et al., 2011). Recent studies added findings showing that impaired executive functioning and attention impact the walking performance of older fallers (MacAulay et al., 2015; Cornu et al., 2016).

Another explanation for DTC of older adults are CoF. Older adults with higher levels of CoF have difficulties to inhibit or ignore irrelevant information from the environment in the process of balance control (Young and Mark Williams, 2015). Therefore, during the cognitive process of movement coordination, the CoF seems to compete for the limited resources of attentional focus to maintain balance control (Young and Mark Williams, 2015), resulting in instability and fall risk. For ST walking performance, a meta-analysis by Ayoubi et al. (2015) revealed significant effects of CoF expressed in increased gait variability. Under DT conditions, Donoghue et al. (2016) found reduced gait speed and step length, especially for older persons who reduced their daily physical activity due to their CoF. Therefore, CoF appears to have an impact on mental processes and might reduce the available resources for task managing in DT situations.

The mental status also seems to play a role. For example, older people with depressive disorders showed reduced DT performance (Nebes et al., 2001). Older adults with unipolar depressive disorders have shown problems inhibiting information, and they also have greater response times in comparison to healthy control groups in DT situations (Gohier et al., 2009). Moreover, Hausdorff et al. (2008) found a correlation between mental well-being and DTC in older adults.

In addition, muscle strength or physical functioning, expressed by reduced hand grip strength (Rantanen et al., 1999; Bohannon, 2008), for example, might influence the DT performance of cognitive-motor DT situations. Reduced hand grip strength has been shown to be an indicator of frailty (Rantanen et al., 1999; Bohannon, 2008), muscle strength, mortality, quality of life, and/or heart health (Norman et al., 2011). In this vein, Guedes et al. (2014) revealed an interaction of frailty (assessed as reduced hand grip strength) and reduced DT performance while walking. Therefore, one might assume that the functional condition can free up cognitive capacity for motor coordination which would otherwise be needed to compensate impaired motor functioning.

In summary, recent literature allows us to derive different explanations for DTCs or decrements. They might be a result of: (1) age-related motor or cognitive declines in general; (2) of task difficulty of the cognitive task or the stimulus-response mode of the cognitive task, especially of tasks that need executive control; (3) the complexity of the motor task (walking situation); (4) task prioritization (posture first); (5) previous falls or CoF; or (6) of mental; or (7) physical functioning, or a combination of several factors.

Nevertheless, extensive research about CMI in older adults has not sufficiently discussed older individuals' preconditions, such as physical functioning (e.g., hand grip strength), psychological factors (e.g., CoF), or mental state (i.e., mental well-being) and the resulting positive or negative DTC. Therefore, the aims of this study were: (1) to identify whether DTC of older adults were positive or negative when performing a visual-verbal Stroop task while maintain walking speed; and (2) to analyze the individuals' preconditions (age, physical functioning, CoF) that might have an impact on positive or negative DTC. We hypothesized that older participants can be clearly classified into groups with and without DTC during DT walking (for step length and step width) based on individual characteristics such as age and CoF.

# MATERIALS AND METHODS

This study consists of a secondary analysis of all baseline data from participants who took part in a larger study program to develop DT managing training. The program was approved by a local Ethics Committee of the Chamber of Physicians (PV4376).

# Participants

Overall, a total sample size of N = 240 participants (mean age and SD: 72.35 ± 5.4 years, age range 57–89, n = 177 female, n = 63 male) was recruited for the study program. Recruitment was conducted using advertisements in local newspapers. The inclusion criteria were: independent living, age 65–85 years, and the ability and mobility [Short Physical Performance Battery (SPPB) > 9; ability to walk without walking aids] to join the study program. Exclusion criteria were: acute or chronic disease with documented influence on balance control (e.g., Parkinson's Disease or Diabetes), use of gait assistance (e.g., walking canes, frames, rolling walkers), a Mini Mental Status Examination (MMSE; Folstein et al., 1975) of less than 25 points indicating any cognitive impairment, and color blindness. A total of 18 participants were excluded (n = 14 due to an SPBB score <9, n = 3 due to an MMSE <25, and n = 1 due to age). All participants were informed about the study goals and risks and signed informed consent prior to any testing according to the Declaration of Helsinki. There was no financial compensation for participating in the study.

All included participants completed a standardized questionnaire assessing demographics, anthropometric data, and comorbidities. Health-related quality of life was examined using the Short Form-12 questionnaire (SF-12 Bullinger and Kirchberger, 1998; see **Table 1**). The analysis includes a physical and mental SF-12 score.

# Outcome Measures

#### Treadmill Walking

Subjects performed a 30-s walking test at a self-selected constant speed on an h/p/cosmos motorized treadmill with integrated sensors to measure peak plantar pressure and other gait kinematics (Zebris, Isny, Germany).

Self-selected walking speed was determined via a staircase method, which means walking up to a certain level of comfortable speed and increasing and decreasing speed until a comfortable pace was achieved (range between 0.7 km/h up to 6.0 km/h). Gait data were collected for both feet at 100 Hz. Standardized measurements of gait kinematics (step length, step width) were conducted with the included FDM-T software: each trial had a duration of 30 s.

Before the test sessions started, all subjects practiced treadmill walking. With familiarization periods of about 5 min, participants were allowed to practice until they felt comfortable with the training device (see Wollesen et al., 2017a). Self-selected gait speed was constantly used for the ST and DT conditions. Participants were secured by a safety harness.

# Cognitive Task

Subjects performed 30-s visual-verbal Stroop tests with 16 events of congruent and incongruent color words (e.g., the word ''blue'' presented in yellow letters). The colors red, blue, yellow, and green were used. Participants had to name the color of the font in which the letters were presented but not the actual word spelled by the letters. The time interval between word insertions varied between 0.8 ms and 1.2 ms to avoid rhythm of occurrence. The tests differed in the sequences of word colors.

To avoid a learning effect, we conducted three different versions of the Stroop test, where congruent and incongruent stimuli were presented via a computer screen in randomized order. All Stroop tests were recorded on video presentation within the software (Garage Band; Apple; Cupertino, CA, USA). The video included the verbal response of the participants to the observed color word on the screen. The number of correct answers was monitored, recorded, and analyzed. The analysis was based on all stimuli, irrespective of the congruency of the stimuli (e.g., the word ''red'' was presented in blue color and the participant answered blue or the word was ''red'' and was presented in red and the participant answered red).

### Condition Cognitive Performance (Sitting and Walking)

In the ST (sitting) and DT (walking) condition, subjects performed the visual-verbal Stroop test with 16 events of color words (written in blue, red, green, yellow). In sitting condition stimuli were projected onto a white wall 2 m in front of the participants (for further details see Wollesen et al., 2016).

### DT Condition Walking

In the DT walking condition, subjects performed the visualverbal Stroop as described above while walking on the treadmill.


BMI, Body Mass Index; SPPB, Short Physical Performance Battery; FES-I, Falls Efficacy Scale International; SF-12, Short Form-12 questionnaire; MMSE, Mini Mental Status Examination. <sup>∗</sup>p < 0.05.

The words were displayed in the size of 40 to 58 cm × 20 cm at a distance of 415 cm. The trial lasted 30 s and its length was matched with the length of the walking sequence. Participants were not introduced to strategies for prioritizing their gait patterns or the cognitive task.

#### Concerns of Falling (CoF)

The German version of the Falls Efficacy Scale International (FES-I, Dias et al., 2006) was used to examine concerns about falling during 16 daily activities. The 16 items are rated as ''not at all concerned'' (1) to ''very concerned'' (4). All items were summed up to a FES-I score. Higher scores are indicative of greater CoF (Delbaere et al., 2010).

#### Physical Functioning

The maximum hand grip strength was measured (Bohannon, 2008) using a Jamar<sup>r</sup> Hydraulic hand dynamometer (Model 5030J1, J. A. Preston Corporation, Clifton, NJ, USA) as a predictor of physical functioning. The hand dynamometer was adjusted to the individual's hand size. Participants were asked for their dominant hand (left or right). Each hand was tested twice with a 1-min rest between trials. The test took place in a standing position with arms extended perpendicular to the body. The maximum value of the two trials for the dominant hand served as the result.

# Data Analysis

Addressing the changes in walking performance under DT conditions, the data analysis focused on the DTC for motor performance while walking. Following Doumas et al. (2009), DTCs were calculated using the formula: (ST-DT/ST) <sup>∗</sup> 100. DTCs were calculated for the walking parameters (step length and step width).

Based on their DTC, participants were separated into three groups:


length or increases in both parameters = non-uniform DT performer.

All statistical analyses were performed using SPSS 24 computer software (IBM statistics Armonk, NY, USA). To analyze differences between the three groups of older adults (negative, positive, non-uniform performer), analysis of variance (ANOVA) were calculated for all DTC outcome parameters (DTC of step length and step width). Significance was set at α = 0.05; normal distribution was tested via the Kolmogorov-Smirnov test. Effect size is presented as partial eta square (η 2 p ; small effect η 2 <sup>p</sup> ≥ 0.08, moderate effect η 2 <sup>p</sup> ≥ 0.20, and η 2 <sup>p</sup> ≥ 0.32 large effect). A Bonferroni correction was applied for all post hoc comparisons.

Furthermore, we analyzed potential influencing factors on DTC. Therefore, Pearson product-moment correlations were computed using all cognitive (right answers for Stroop task performance while sitting and walking) and psychological variables (SF-12 mental score, FES-I-scores), physical characteristics (gait speed, physical functioning, SF-12 physical score), and relevant demographics (age) of the participants. Next, a four-step hierarchical regression analysis was conducted to identify which variables determine the positive, negative, or non-uniform DTC while walking.

In the first step, age, and in the second step all relevant physical characteristics (hand grip strength, SF-12 physical score, and preferred gait speed) were included. In the third step, the psychological components were entered (SF-12 mental score, FES-I-scores). In step 4, the model was adjusted to cognitive DT performance (right answers sitting and walking).

# RESULTS

**Table 1** describes the physical characteristics and demographic conditions of the participants (N = 222).

The only significant group difference observed in **Table 1** was the age of the subjects. Participants with positive DTC were older than the two other groups (F(2,215) = 4.953; p = 0.008; η 2 <sup>p</sup> = 0.044).

The range of positive and negative DTC for step width and step length was between 1% up to 95% for step width, and 1% up to 60% for step length.

**Table 2** shows the correlations between DTC and the physical, cognitive, and demographic characteristics of the participants.


SF-12 phys men, Score of SF 12 Questionnaire; DTC, dual task costs; FES-I, Falls Efficacy Scale International; <sup>∗</sup>p < 0.05, ∗∗p < 0.01. Bold values highlight significant differences. Phys, physical score; Men, mental score.



<sup>∗</sup>Significant post hoc test of group comparisons; l, left foot; r, right foot. Bold values highlight significant differences.

There were some significant correlations between the participants' physical and cognitive conditions. The scores of the FES-I were correlated with hand grip strength (see **Table 2**); participants with higher hand grip strength had reduced FES-I scores. In addition, a higher physical and mental well-being was associated with lower FES-I scores. Gait speed was positively correlated with hand grip strength and was reduced with increasing FES-I scores.

The differences in the examined gait variables for the three subgroups of DT performance are documented in **Table 3**.

Regression analysis of relevant physical, cognitive, and psychological characteristics and demographic conditions of the participants and DTC is shown in **Table 4**.

Steps 1 and 2 of the regression analysis of age and the physical parameters did not indicate a significant effect. In step 3, mental well-being and FES-I were integrated into the model. The overall model was significant (F(6,75) = 2.575; p = 0.025; see **Table 4**). In this step, significant effects for hand grip strength (p = 0.007) and FES-I (p = 0.003) were observed. Participants with negative DTC showed higher CoF. Participants with negative DTC had higher hand grip strength (see **Figure 1**; **Table 4**).

The analysis of step 4 included the cognitive performance in the Stroop test. The significant overall effect remained (F(8,73) = 2.234; p = 0.034), as well as the significant effects for hand grip strength (p = 0.005) and FES-I (p = 0.003; see **Figure 1**).

# DISCUSSION

Motor-cognitive DTC during walking in older adults might be a result of age-related motor, cognitive declines, previous falls, or CoF. However, previous research revealed heterogeneous


Step 1: Age Step 2: Hand grip strength. SF-12 phys, gait speed (km/h), Step 3: SF-12 men, FES-I and Step 4: Right answers sitting and right answers walking, <sup>∗</sup>p < 0.05, ∗∗p < 0.01.

results and did not sufficiently discuss whether other individuals' preconditions, like physical functioning, psychological factors (CoF), or mental factors, might affect DTC positively or negatively. Therefore, the aims of this study were: (1) to identify whether DTC of older adults were positive or negative when performing a visual-verbal Stroop task while walking; and (2) to analyze the individuals' different preconditions that might have an impact on positive or negative DTC. Our main hypothesis was that participants could be clearly classified into two groups revealing either positive or negative influence of the secondary task on walking performance (step length and step width). Overall, we were able to classify three groups with different DTC patterns: (1) participants with positive DTC, which means their step length increased and step width decreased (positive DT performer); (2) participants with negative DTC expressed by reduced step length and increased step width (negative DT performer); and (3) participants that either improved or reduced only one of the gait parameters (nonuniform DT performer) (**Figure 2**). With respect to demographic characteristics, the groups only differed in age. Specifically, the positive DTC group was older than the other two groups. Moreover, physical functioning and CoF might be associated with DTC, as well.

# Positive, Negative, and Non-uniform DTC

We were only able to classify 50 percent of the participants into the groups with overall positive or negative adaptions to the DT situation, which was unexpected. The other 50 percent showed either positive or negative effects on step width or step length, meaning step width and step length increased or vice versa, thus revealing non-uniform DTC. These opposed changes in the gait parameters might be strategies to compensate the additional cognitive load to secure gait performance (Beurskens and Bock, 2012; Wrightson et al., 2016). Thus, results indicate that performance does not necessarily decline under DT conditions as long as there is room for compensation. As the majority of earlier studies focused on one gait parameter only (mostly gait speed) and did not control for different DT performance levels, they might have misinterpreted the negative DTC when analyzing the gait decrements. In our study, gait speed was assessed in the first session to determine comfortable walking speed and then remained constant across the whole trial (motor driven treadmill). Thus, our participants did not reveal declines in gait speed. Nevertheless, all participants revealed performance changes under DTC conditions in at least one gait parameter (step length or width), but more than two thirds revealed either decline or compensation. There were also participants who showed only one or two percent variance between ST and DT performance or even positive DTC. This is why we suggest that DTC of older adults performing cognitive-motor tasks such as walking are not negative in general, but depend of the type of measurement or might be a compensation strategy (Li et al., 2001; Bock, 2008). The observed gait adaptions to the CMI of

all participants might be a result of a compensation process due to the increased cognitive load. It has been suggested that these adaptions are the individual's compensation strategies to increased task complexity (Hausdorff et al., 2001; Schaefer and Schumacher, 2011; Wollesen et al., 2016). In addition, it needs to be reflected that there is still a lack of information about the degree to which a certain change in step width and step length might be a positive or a negative adaption to an increased cognitive load while walking. Moreover, the walking parameters that should be observed are not clearly identified or described by existing studies [absolute values of gait kinematics, like the step length and step width, or measurements of variability, e.g., as expressed by Auvinet et al. (2017) or Hausdorff et al. (2008)]. We only analyzed the absolute values of the measurements, as their might be an error propagation if additional calculations were added to the standard measurements. Previous research of our measurement setup showed poor interclass correlation coefficients (ICC's) for gait variability outcome variables (Wollesen et al., 2017a).

In contrast to standard measurements of cognitive performance, like reaction times, the complex coordination of walking performance cannot be described with only one variable. However, a clear explanation as to which walking variability will be effected most by CMI cannot yet be answered by the existing literature.

Furthermore, with respect to age, we found unexpected group differences in the participants with positive and negative DTC. Participants with performance decrements under DT conditions were younger than participants with positive DTC. These results contradict the findings of Plummer-D'Amato et al. (2012), who hypothesized that there is an overall age-related decrement on DT performance. In addition, it remains unclear why age did not correlate with walking speed as reported by e.g., Donoghue et al. (2016). Our results confirm the idea that age is not the only variable to explain DTC. Individual characteristics, termed as inter-individual variability (see for example Baltes et al., 1999), might have a greater impact on CMI than age itself. On the other hand, it needs to be considered that the age difference between the two groups of positive and negative DT performers was only 3 years. The results might have differed, if there had been a difference of 10 years or more.

# Potential Indicators of DTC

As revealed by the regression analyses, DTC were associated with physical functioning (grip strength) and psychological factors (CoF). Contradictory to our expectations, participants with positive DTC were older and revealed lower physical functioning (reduced hand grip strength). Reduced physical functioning along with higher age has been described as a potential factor for negative DT performance in previous literature (Beurskens and Bock, 2012). Our findings confirm the idea of physical decline with aging, but we found improved walking performance under an increased cognitive load in this group. Therefore, other individual preconditions besides age, like physical or cognitive functioning, also seem to matter.

The observed reduced hand grip strength as one parameter of reduced physical fitness or frailty (Bohannon, 2008; Rantanen et al., 1999) was associated with positive DTC when performing an executive function task while walking. Thus, it might indicate that, next to strength, additional motor preconditions are required to perform and maintain motor performance under more challenging requirements, such as DT conditions. This relationship was also reported by Voelcker-Rehage et al. (2010), who found that physical fitness indexed by muscular strength was related to cognitive performance. However, this idea was not supported by our results.

Another unexpected finding was that, for the older and less physically fit participants, the additional cognitive load benefitted movement coordination during DT walking, as shown by reduced DTC. Comparable results have been found for tasks like cuing for patients with Parkinson's disease (Lim et al., 2005) and could be explained by the Supra postural task model (Stoffregen et al., 1999, 2000, 2007; Swan et al., 2004). Following the Supra postural task model, in contrast to the ''posture first hypothesis'', the secondary cognitive task is the main movement goal and balance performance is organized to fulfill the goal (Stoffregen et al., 1999, 2000). Following this idea, the DT situation becomes the new focus of attention and replaces dysfunctional motorcoordination or executive aspects. The participants are highly concentrated on cognitive performance and motor performance improves. However, the data of this study cannot give a clear explanation of this phenomenon. Additional research comparing participants with positive and negative DTC is needed to gain insights into the mechanisms of resource allocation of older adults.

Since there were no group differences in cognitive performance of the Stroop task, the presented results indicate that all participants used the same strategy. Participants maintained a high level of correct answers during the Stoop task under DTC conditions [90% of correct answers in comparison to 80% correct answers revealed by van Iersel et al. (2008)], indicating that they focused on cognitive performance (as shown in previous studies, see Wollesen et al., 2017a,b) and did not use a ''gait first'' strategy. Hence, the participants in our study did not act according to the ''posture first hypothesis'' as expected by the task prioritization model (Hausdorff et al., 2001). These findings are in line with other studies that also failed to confirm the ''posture first hypothesis'' (e.g., Li et al., 2012; Janouch et al., 2018). The study by Janouch et al. (2018) used a street crossing task in a virtual reality setting with increasing task complexities, while the study by Li et al. (2012) focused on treadmill walking with two different task complexities of an arithmetic task. Since the two studies, as well as our study, used a laboratory setting, the deviating results might be explained by the unreal conditions (virtual reality, treadmill): they could have had an impact on task prioritization, because the participants might have felt secure in the laboratory environment. On the other hand, one could argue that the self-selected gait speed of less than 1 m/s was a security mechanism, which already addressed the situation on the treadmill under the ST condition. Moreover, participants with CoF adopted the additional load mainly by an increased step width to increase the base of support. In contrast to participants without CoF, this might be a posture first mechanism. However, it remains unclear whether this can be specified as a conscious decision by the participants to secure gait performance.

In comparison to the other groups, CoF was higher in participants with negative DTC, and CoF were significantly associated with DTC. Earlier studies also found gait decrements for persons with higher CoF (Rochat et al., 2010; Donoghue et al., 2016; del-Río-Valeiras et al., 2016). Our results confirmed the findings that CoF has (besides physical functioning) the highest impact on walking performance in DT situations. According to the review by Young and Mark Williams (2015), CoF lead to difficulties inhibiting irrelevant information and, together with the cognitive task, this information needs resources of the working memory. All of the resources compete for the attentional focus which is needed for movement control. Following this, fear or CoF might have the same effect as a DT itself (Young and Mark Williams, 2015), and participants with high CoF may have fewer resources available for performing the task itself in comparison to participants with less CoF, and therefore show more gait decrements in DT situations.

Nevertheless, the analysis of the presented DT gait data showed that CMI while walking does not generally occur. Moreover, the question is why we identified such a great number of participants who have positive gait changes in DT situations. Our regression model suggests that a good functional and psychological state, here expressed as grip strength and fewer CoF, might be factors influencing motor performance under demanding DT conditions. Besides the different models that explain CMI in older adults, considering (individual) influencing factors and a broader approach to explain DTC in different task complexities is needed.

# LIMITATIONS

One limitation of this study was that we did not control for cognitive DTC, e.g., reaction times. Assessing the cognitive DTC might give more insights about the adaption processes of the different DTC performers. This aspect needs to be addressed in future studies. However, we controlled cognitive performance by counting the correct answers for the Stroop task.

Moreover, the measurement setup addressed changes of the gait parameters while maintaining gait speed under the ST and DT conditions. According to the literature, the participants might have reduced their walking speed from ST to DT, which was not possible under the conditions of this study.

In addition, participants with CoF should be asked if the treadmill condition increases or reduces their concerns, and what kind of safety strategies they use, if they are afraid of falling.

# CONCLUSION

Our results indicate that individual preconditions should be considered when calculating DTC and when deriving conclusions for appropriate training programs. Similarly, neuroimaging studies found that imagined walking involves more cognitive control and less automated processing in low- compared to well-functioning adults (Godde and Voelcker-Rehage, 2010) and that ST gait training reduces this cognitive involvement, particularly in low-functioning persons (Godde and Voelcker-Rehage, 2017). This leads to the conclusion that we need to control these parameters in our future research projects more carefully. We particularly recommend controlling the physical fitness and CoF as standardized instruments to describe the participants' characteristics for DT studies. Future DT studies should consider inter-individual differences in DTC when developing and evaluating training approaches or fall prevention programs.

# REFERENCES


# AUTHOR CONTRIBUTIONS

BW conducted the study idea and the experimental design was developed by BW and CV-R. The data analysis was done by BW and CV-R. The manuscript was written by BW and added by CV-R.

# ACKNOWLEDGMENTS

We thank Antonius Baehr for editing the manuscript.

relationship between short FES-I test scores and objective assessment of balance in the older people with age-induced instability? Arch. Gerontol. Geriatr. 62, 90–96. doi: 10.1016/j.archger.2015.09.005


mobility decisions in seniors at risk for falls. Psychol. Aging 26, 253–259. doi: 10.1037/a0022929


older adults. J. Geriatr. Phys. Ther. 36, 115–122. doi: 10.1519/jpt.0b013e318 27bc36f


**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 Wollesen and Voelcker-Rehage. 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.

# Motoric Cognitive Risk Syndrome: Could It Be Defined Through Increased Five-Times-Sit-to-Stand Test Time, Rather Than Slow Walking Speed?

#### Harmehr Sekhon1,2, Cyrille P. Launay<sup>3</sup> , Julia Chabot1,4, Gilles Allali<sup>5</sup> and Olivier Beauchet1,6,7 \*

<sup>1</sup> Division of Geriatric Medicine, Department of Medicine, Sir Mortimer B. Davis – Jewish General Hospital and Lady Davis Institute for Medical Research, McGill University, Montreal, QC, Canada, <sup>2</sup> Division of Experimental Medicine, Faculty of Medicine, McGill University, Montreal, QC, Canada, <sup>3</sup> Geriatric Medicine and Geriatric Rehabilitation Service, Department of Medicine, Lausanne University Hospital, Lausanne, Switzerland, <sup>4</sup> Division of Geriatric Medicine, Department of Medicine, St. Mary's Hospital Center, McGill University, Montreal, QC, Canada, <sup>5</sup> Department of Neurology, Geneva University Hospital, University of Geneva, Geneva, Switzerland, <sup>6</sup> Dr. Joseph Kaufmann Chair in Geriatric Medicine, Faculty of Medicine, McGill University, Montreal, QC, Canada, <sup>7</sup> Lee Kong Chian School of Medicine, Nanyang Technological University, Singapore, Singapore

#### Edited by:

Helena Blumen, Albert Einstein College of Medicine, United States

#### Reviewed by:

Andreas Zwergal, Ludwig Maximilian University of Munich, Germany Uros Marusic, Institute for Kinesiology Research, Slovenia

> \*Correspondence: Olivier Beauchet olivier.beauchet@mcgill.ca

Received: 04 September 2018 Accepted: 19 December 2018 Published: 23 January 2019

#### Citation:

Sekhon H, Launay CP, Chabot J, Allali G and Beauchet O (2019) Motoric Cognitive Risk Syndrome: Could It Be Defined Through Increased Five-Times-Sit-to-Stand Test Time, Rather Than Slow Walking Speed? Front. Aging Neurosci. 10:434. doi: 10.3389/fnagi.2018.00434 Background: Slow walking speed, time to perform the five-times-sit-to-stand (FTSS) test and motoric cognitive risk syndrome (MCR; defined as slow gait speed combined with subjective cognitive complaint) have been separately used to screen older individuals at risk of cognitive decline. This study seeks to (1) compare the characteristics of older individuals with MCR, as defined through slow walking speed and/or increased FTSS time; and (2) examine the relationship between MCR and its motor components as well as amnestic (a-MCI) and non-amnestic (na-MCI) Mild Cognitive Impairment.

Methods: A total of 633, individuals free of dementia, were selected from the crosssectional "Gait and Alzheimer Interactions Tracking" study. Slow gait speed and increased FTSS time were used as criteria for the definition of MCR. Participants were separated into five groups, according to MCR status: MCR as defined by (1) slow gait speed exclusively (MCRs); (2) increased FTSS time exclusively (MCRf); (3) slow gait speed and increased FTSS time (MCRsaf); (4) MCR; irrespective of the mobility test used (MCRsof); and (5) the absence of MCR. Cognitive status (i.e., a-MCI, na-MCI, cognitively healthy) was also determined.

Results: The prevalence of MCRs was higher, when compared to the prevalence of MCRf (12.0% versus 6.2% with P ≤ 0.001). There existed infrequent overlap (2.4%) between individuals exhibiting MCRs and MCRf, and frequent overlap between individuals exhibiting MCRs and na-MCI (up to 50%). a-MCI and na-MCI were negatively

[odd ratios (OR) ≤ 0.17 with P ≤ 0.019] and positively (OR ≥ 2.41 with P ≤ 0.019) related to MCRs, respectively.

Conclusion: Individuals with MCRf are distinct from those with MCRs. MCRf status does not relate to MCI status in the same way that MCRs does. MCRs is related negatively to a-MCI and positively to na-MCI. These results suggest that FTTS cannot be used to define MCR when the goal is to predict the risk of cognitive decline, such as future dementia.

Keywords: older inpatients, epidemiology, screening, cognition, motricity

# INTRODUCTION

Motoric cognitive risk syndrome (MCR) is defined as the relationship between objective slow gait speed and subjective cognitive complaint (Verghese et al., 2013). MCR is one of the stages of pre-dementia, similar to mild cognitive impairment (MCI) (Verghese et al., 2013, 2014). MCR does not require a time-consuming comprehensive neuropsychological assessment when compared to MCI, which opens new perspectives in terms of detection of individuals who are at risk of dementia in older populations (Verghese et al., 2013, 2014; Belleville et al., 2017). The past decade has been characterized by an increased interest in identifying and validating biomarkers for early diagnosis and identification of individuals who are at risk of dementia (Belleville et al., 2017). However, the use of biomarkers has limitations in many settings. For instance, access to neuroimaging is difficult and the cost of biological biomarkers limits their use (Handels et al., 2017). Additionally, the highest prevalence and incidence of dementia in the coming years will be observed in low and intermediate income countries, where the accessibility of actual biomarkers is limited (de Jager et al., 2017). Hence, there is a need to optimize and increase the accessibility to clinical risk assessment of dementia in community-dwelling older populations. Using a motor test to predict dementia in older populations may be a solution.

Motoric cognitive risk syndrome has the potential to rapidly screen individuals who are at risk of dementia in a primary care setting, where the under-diagnosis of dementia is estimated to be around 50% in individuals over 65 (Iliffe et al., 2009). This under-diagnosis of dementia is largely related to limited resources and the time required for in-depth assessments of cognitive complaint (Iliffe et al., 2009; Villars et al., 2010). The simplicity of assessment of MCR syndrome could help overcome this issue. However, gait speed, a component of MCR, may be difficult to assess during a primary care visit because of space constraints (Abellan van Kan et al., 2009). Gait speed must be recorded at usual steady state pace rhythm over at least 3 meters (Middleton et al., 2015). Few consultation rooms in primary care possess the features required for the assessment of gait speed, which complexifies the process of consultation and increases physician workload and consult time, when gait speed must be measured. It has been reported that consult time in general practice is very short (around 6.9 min) and depends on the physician, the physician's workload and the type of visit (Petek Ster et al., 2008). There is, therefore, a need in primary care for a simpler mobility test, which can be completed rapidly and within limited space, so as to facilitate MCR diagnosis in primary settings. In addition, the chosen motor test must be proven to show a link to cognitive impairment or risk of dementia, as the objective of a redefined MCR is to identify individuals at risk of dementia.

The five-times-sit-to-stand test (FTSS) is a physical test, which measures the time taken by an individual to repeat five consecutive chair rises as quickly as possible (Whitney et al., 2005). This motor test examines the challenged balance condition, which is the transfer from a sitting position to a stand-up position. The FTSS test possesses the necessary features for assessment of mobility performance to diagnose MCR in primary care, as it can easily and rapidly be performed in limited space and its requirements are limited to a chair and a stopwatch. In addition, this test may be performed at the time of consultation, as its duration is of fewer less than 2 min in length, including explanation and performance (Whitney et al., 2005). Thus, the FTSS test does not increase the physician's workload. The one-leg-balance (OLB) test is another simple motor test to examine the challenged balance condition. In it, the individual is asked to stand unassisted on one leg. An impaired OLB test result – defined as being unable to stand on one leg for 5 s – has been identified as a predictor of injurious falls among community-dwelling older adults and cognitive decline in patients with dementia, but not in non-demented individuals (Vellas et al., 1997). In contrast, increased FTSS time has been associated with low cognitive performance in older community dwellers free of dementia (Annweiler et al., 2011). Because non-demented individuals with poor cognitive performance like MCI are at risk of dementia, this association suggests that poor FTSS performance (i.e., increased time) may be used to identify individuals at risk of dementia and thus, that it could be used as an alternate motor test, as opposed to gait speed, to define MCR. Using FTSS performance instead of gait speed to define MCR value for the prediction of dementia requires an investigation, which will determine whether or not individuals classified as MCR through FTSS performance and gait speed are one and the same. This line of questioning is justified, as the FTSS test explores different subdomains of mobility, when compared to gait speed (Whitney et al., 2005; Annweiler et al., 2011; Beauchet et al., 2017; Sekhon et al., 2017). The FTSS test examines the ability to transfer from a sitting position and depends largely on balance control, muscle mass, strength, and the power of lower limbs (Whitney et al.,

2005). In comparison, gait speed is a surrogate measure of gait ability, which depends on different body movements and higher levels of movement control involving executive and memory functions (Abellan van Kan et al., 2009; Middleton et al., 2015; Beauchet et al., 2017). These differences between the FTSS test and gait speed, therefore, call into question the possible overlap between individuals whose MCR status was determined using either the FTSS test or gait speed, and their relative MCI.

Motoric cognitive risk syndrome and cognitive impairment are both intermediate stages between normal cognitive aging and major neurocognitive disorders (Verghese et al., 2013, 2014; Belleville et al., 2017). A knowledge gap exists regarding the relationship between MCR and MCI syndromes. Recently, we underscored that there exists overlap between MCR – defined through slow gait speed – and MCI in the population of older community dwellers (Sekhon et al., 2017). The prevalence of MCI was higher in individuals with MCR, when compared to those without MCR (47.2% versus 39.5%) (Sekhon et al., 2017). Unfortunately, the relationship between MCR subcategories of MCI syndromes such as amnestic (a-MCI) and non-amnestic (na-MCI) has not been examined in this study. As gait is largely controlled by executive functions (Beauchet et al., 2017), we have hypothesized that MCR as defined by slow gait speed (MCRs) may be more frequently associated with na-MCI, when compared to MCR as defined by increased FTSS time (MCRf), which may be associated with a-MCI. Using the data of the cross-sectional study known as the "Gait and Alzheimer interactions tracking" (GAIT) study (Beauchet et al., 2018), we had the opportunity to explore the overlap between MCR as defined by slow gait speed and increased FTSS time, and their relationship with a-MCI and na-MCI. This study aims to (1) compare the characteristics of participants of the GAIT study with and without MCR as defined by slow gait speed and increased time of FTSS, and (2) examine the relationship between MCR, and a-MCI and na-MCI. Comparing gait speed and FTSS as a construct of MCR, as well as their relationship with MCI subtypes, may provide new insight into the interaction between motor and cognitive impairment in the aging population.

# MATERIALS AND METHODS

# Population and Study Design

A subgroup of older individuals recruited in the GAIT study were selected for the present study. The GAIT study is a crosssectional design-based study, which was conducted in France between November 2009 and 2015 (Beauchet et al., 2018). All GAIT participants were relatively healthy community-dwelling individuals, who were recruited during a visit in the memory clinic of Angers University Hospital, in France, for cognitive complaint evaluation. The GAIT study inclusion criteria were: (1) age 65 years and over, (2) living at home in the community, and (3) an adequate understanding of French. Exclusion criteria included acute medical illness (regardless of nature) in the past month; extrapyramidal rigidity of the upper limbs (regardless of etiology); neurological diseases [past history of stroke, (NPH), multiple sclerosis, Parkinson's disease, cerebellar disease, polyneuropathy, and vestibular disease]; psychiatric diseases (past history of psychosis, personality disorders or severe depression as well as active depression as defined by a 4-item geriatric depression scale (GDS) score above 1) (Shah et al., 1997) other than cognitive impairment; severe gait-affecting medical conditions which left potential participants with the inability to walk unassisted for 15 min, such as rheumatologic diseases (spine, pelvic, and joint arthritis with deformation); and ophthalmic diseases with severe vision abnormality. In addition, for the present study, we also excluded participants with NPH or presenting vascular brain abnormalities (i.e., lacunar lesions and strokes) on brain imaging [i.e., computed tomography (CT) or magnetic resonance imagery (MRI) scan] performed during the assessment, suffering from dementia, using a walking aid, and presenting no gait speed or FTSS time data. A total of 663 participants were selected, after applying these selection criteria.

# Study Assessments

The selected GAIT participants had a full-standardized clinical examination, a comprehensive neuropsychological assessment, brain imaging (i.e., MRI or CT) and blood tests including Vitamin B12, TSH, calcemia and other serum electrolytes, creatinine, and urea. Age, sex, educational level [evaluated by number of years of schooling and categorized by high school level (i.e., yes or no)], number of drugs taken daily and body mass index (BMI; kg/m<sup>2</sup> ) were recorded. Maximal isometric voluntary contraction (MVC) strength of hand was measured with the help of a computerized hydraulic dynamometer (Martin Vigorimeter, Medizin Tecnik, Tutlingen, Germany). The test was performed once on each side. The highest MVC value recorded was used in the present data analysis. Binocular distance vision was measured at 5 m with a standard Monoyer letter chart and scored from 0 (i.e., worst performance) to 10 (i.e., best performance) (Lord et al., 1991b). Vision was assessed with corrective lenses if needed. Lower-limb proprioception was evaluated with the help of a graduated tuning fork placed on the tibial tuberosity, so as to measure vibration threshold (Buchman et al., 2009). The mean value obtained for the left and right sides ranged between 0 (i.e., worst performance) and 8 (i.e., best performance) and was used in the present data analysis. Gait speed was measured with the help of GAITRite <sup>R</sup> (Gold walkway, 972 cm long, active electronic surface area 792 cm × 610 cm, total of 29,952 pressure sensors, scanning frequency 60 Hz, CIR System, Havertown, PA, United States). Time to perform FTSS was also measured. A trained evaluator demonstrated the test procedure while giving standardized verbal instructions. Moreover, before testing, participants were allowed to practice the sit-to-stand test twice. Participants began by crossing their arms upon their chest and sitting with their back against the chair (45 cm above the floor). The chair was padded and armless. They were prompted not to bounce off the chair when returning to the standing position, and reminded to fully straighten their legs when elevating. Participants were instructed to stand up and sit down five times as quickly as possible. Performance was measured

with a stopwatch in seconds, from the time at initial seated position to the time at final seated position, after completing five stands. A bedside face-to-face neuropsychological assessment was also performed using the mini-mental state examination (MMSE) (Folstein et al., 1975) and frontal assessment battery (FAB) (Dubois et al., 2000), the French version of the free and cued selective reminding test-total recall (FCSRT-TR) (Van der Linden et al., 2004), parts A and B of the trail making test (TMT) (Brown et al., 1958), the Stroop Test (Stroop, 1935), and the instrumental activities of daily living scale (IADL) (Pérès et al., 2006). The diagnosis was determined, following a standardized procedure and consensual definition, during multidisciplinary meetings involving geriatricians, neurologists, and neuropsychologists of Angers University Memory Clinic. It was based on the aforementioned neuropsychological tests, physical examination findings, blood tests, and MRI or CT scan of the brain. First, the cognitive status was determined using the performances obtained during the neuropsychological assessment. Participants were classified within cognitively healthy individuals (CHI), a-MCI and na-MCI categories, diagnosis of MCI being in accordance with the criteria detailed by Dubois et al. (2010). CHI were individuals who exhibited normal cognitive function with all cognitive scores using the referent age-appropriate mean value. Participants with a-MCI and na-MCI were individuals, who have an objective impairment in the memory (i.e., a-MCI) or non-memory (i.e., na-MCI) domains, respectively, defined as a score >1.5 SDs beneath the ageappropriate mean, and who have not impaired daily living activities (i.e., normal IADL score). Second, the etiology of MCI (i.e., related to neurodegenerative brain lesions versus secondary to metabolic disorders) was determined using the results of blood tests and the brain MRI.

# Definition of Motoric Cognitive Risk Syndrome and Categorization of Participants

Different definitions of MCR were used for each subgroup. First, the diagnosis of MCR was made through slow walking speed (MCRs) in accordance with the criteria described by Verghese et al. (2013): a combination of cognitive complaint and slow gait, with the absence of dementia or any mobility disability. As cognitive complaint was the reason for referral to the memory clinic for participants of the GAIT study, all of them met the criteria for cognitive complaint. Slow gait speed was defined as gait speed of one SD or greater, beneath the age-and sexappropriate mean values established by the present cohort, as done in previous studies (Verghese et al., 2013, 2014). Second, MCR was also defined using increased FTSS time (MCRf) defined as time one SD or greater, above the age-and sexappropriate mean values established by the present cohort. Five subgroups of individuals were identified: (1) those with MCRs using gait speed exclusively; (2) those with MCRf using FTSS time exclusively; (3) those with MCR with abnormal scores in both gait speed and FTSS time (MCRsaf); (4) those with MCR irrespective of mobility test used (MCRsof); and (5) those without MCR.

# Standard Protocol Approvals, Registrations, and Patient Consent

This study was conducted in accordance with the ethical standards set forth in the Helsinki Declaration (1983). Participants in the study were included after obtaining written informed consent for research. The local Angers Ethics Committee approved the study protocol (n◦ 2009-A00533-54).

# Statistics

The participants' characteristics were summarized using means and SDs or frequencies and percentages, as appropriate. Between-group comparisons were performed using a Kruskal-Wallis or Chi square test, Mann-Whitney, independent t-test; unpaired t-test or Chi square test, as appropriate. Uni and multiple logistic regression analyses were performed to examine the relationship between MCR (i.e., dependent variable) and MCI (i.e., independent variable), relative to participants' characteristics. P-values less than 0.05 were considered statistically significant. All statistics were performed using SPSS (version 23.0; SPSS, Inc., Chicago, IL, United States).

# RESULTS

**Table 1** illustrates the participants' characteristics and their comparisons between the different subgroups of participants based on MCR definition. A total of 76 (12.0%) participants were classified as having MCRs, 39 (6.2%) MCRf, 15 (2.4%) MCRsaf, and 130 (20.5%) MCRsof. Individuals with MCR, irrespective of the type of MCR, had the same clinical characteristics, except for sex and level of education. Prevalence of women varied between the different subgroups of MCR (P = 0.029), the highest prevalence being observed with MCRs. Participants with MCRs displayed a lower level of education when compared to those with MCRf (P = 0.008). Participants with MCR displayed lower limb proprioception when compared to non-MCR participants (P = 0.042). The prevalence of MCI syndrome, regardless of type, was significantly different between the three subgroups of MCR (P = 0.039). The prevalence of a-MCI was lower in individuals with MCRs when compared to those with MCRf (P = 0.010) and MCRsaf (P = 0.018), whereas the prevalence of na-MCI was higher in individuals with MCRs when compared to those with MCRf (P = 0.010) and MCRsaf (P = 0.018). Those displaying MCRf registered greater walking speeds when compared to those with MCRs (P ≤ 0.001) and MCRsaf (P ≤ 0.001). Time to perform FTSS was lower in individuals with MCRs (P ≤ 0.001) and MCRsaf (P ≤ 0.001). Comparisons between individuals with MCR, irrespective of definition, and without MCR show that all characteristics differed significantly (P < 0.05), except for age.

Multiple logistic regressions have shown a positive relationship between MCRs and a-MCI and a more marked (**Table 2**) negative relationship between MCRs and na-MCI (P ≤ 0.020). All MCR, irrespective of definition, displayed a positive relationship with MCI (all categories P = 0.010, a-MCI P = 0.040 and na-MCI P = 0.046). These last relationships


 633).

fnagi-10-00434 January 21, 2019 Time: 18:38 # 5

dynamometers

fork placed on the tibial tuberosity; P-value significant (i.e., P < 0.05) indicated in bold.

 expressed in Newton per square meter; ‡‡, binocular vision acuity at distance of 5 m with a standard Monoyer letter chart; ¶¶, mean value of left and right side and based on graduated diapason tuning

TABLE 2 | Logistic regressions showing the association between motoric cognitive risk syndrome and mild cognitive impairment (n = 633).


OR, odd ratio; CI, confidence interval; MCR, motoric cognitive risk syndrome; FTTS, five times sit-to-stand; <sup>∗</sup> , separate model for gait speed; FTSS, gait speed and FTTS, gait speed and/or FTTS; †, Exclusive (i.e., only participants with mean value of the motor test below 1 standard deviation); Model 1, adjusted for age and sex; Model 2, adjusted for age, sex, number of drugs daily taken, body mass index, educational level, handgrip strength, distance vision acuity, and lower-limb proprioception. P-value significant (i.e., P < 0.05) indicated in bold.

remained insignificant when logistic regressions were adjusted for all participant characteristics.

# DISCUSSION

The study findings demonstrate that the use of gait speed and FTSS time to define MCR results in the selection of different subgroups of individuals, with infrequent overlap (2.4%). In contrast, there existed significant overlap between MCR and na-MCI participants (up to 50%). In addition, only MCRs exhibited a significant relationship with the MCI subgroups, the na-MCI subtype, relating positively and the a-MCI relating negatively to this MCR subgroup.

There existed infrequent overlap between individuals with MCR as defined by gait speed and FTSS time. This result suggests that impaired performance in these two motor tests tracks different clinical phenotypes of individuals. But it is not consistent with a previous study, which used alternate gait parameters to define MCR and reported greater overlap (68%) (Allali et al., 2016b). Comparatively, in the previous study the definition used for the different subtypes of MCR involved low performance of gait parameters (mean and variability of stride length and swing time). In our study, even if both gait speed and FTSS examine a condition of dynamic balance in which the body's center of gravity is maintained within a narrow base of support while moving (Lord et al., 1991a; Dubost et al., 2005), they relate to different brain regions, which may explain the infrequent overlap observed (Lord et al., 1991a; Nutt et al., 1993; Dubost et al., 2005; Rosano et al., 2007; Wittenberg et al., 2017). For instance, gray matter volumes in the left frontal lobe were correlated with usual gait speed in healthy older adults, whereas reduced volumes in putamen and superior posterior parietal lobule were associated with balancing difficulty in semi-tandem stance (Rosano et al., 2007). Functional brain imagery study findings point to involvement of the premotor, supplementary motor, and parietal cortex in standing balance control, whereas the hippocampus and premotor cortex are the key region for gait control (Janssen et al., 2002; Rosano et al., 2007; Beauchet et al., 2009, 2012, 2016; Spyropoulos et al., 2013; Wittenberg et al., 2017). Subcortical regions have also been identified as key regions for gait control including the cerebellar locomotor region, the mesencephalic locomotor region, and the subthalamic locomotor region (Bohnen et al., 2011). Gait speed is a surrogate measure of gait, which is the medical term used to globally describe the human locomotor movement of walking (Nutt et al., 1993; Beauchet et al., 2017). Gait is a complex movement in terms of biomechanics and motor control (Nutt et al., 1993; Rosano et al., 2007; Beauchet et al., 2009, 2012; Wittenberg et al., 2017). It has been highlighted that even the simplest walking condition, such as straight-line walking at a comfortable steadystate pace without any disturbances, involves cortical networks and cognitive functions (Nutt et al., 1993; Rosano et al., 2007; Beauchet et al., 2009; Wittenberg et al., 2017). This association may explain the predictive value of slow gait for the occurrence of dementia (Beauchet et al., 2016). In contrast, FTSS time explores the performance of body transfer movement from a seated position (Whitney et al., 2005). This movement is more unstable in terms of biomechanics, when compared to walking at a comfortable steady-state pace without any disturbances (Spyropoulos et al., 2013). It involves an unstable movement from a static and stable position to a quasi static position (Janssen et al., 2002; Whitney et al., 2005; Bohnen et al., 2011; Schofield et al., 2013; Spyropoulos et al., 2013; Lee et al., 2017). Thus, FTSS time is strongly related to several physiological sensory and motor subsystems which contribute to the dynamic postural control, the most important ones identified in older adults being the muscle strength, lower-limb proprioception, vestibular, and vision subsystems (Janssen et al., 2002; Bohnen et al., 2011; Schofield et al., 2013; Spyropoulos et al., 2013). Balance control like gait control deteriorates with the progression of dementia (Lee et al., 2017). This is similar to the decline of gait control

with the progression of dementia (Annweiler et al., 2011). Increased FTSS time has been associated with low cognitive performance in older adults free of dementia (Annweiler et al., 2011). This association has mainly been reported through bedside cognitive tests exploring global cognitive functioning, such as the MMSE the modified mini mental state (3MS) and Pfeiffer's Short Portable Mental State Questionnaire (Hirsch et al., 1997; Raji et al., 2002; Rosano et al., 2005; Annweiler et al., 2011).

The second main finding of our study is the significant overlap between MCI and MCR, irrespective of the criteria for MCR definition. The prevalence of MCI was significantly higher in individuals with MCR when compared to those without MCR, and ranged from 52.6% for individuals with MCRs to 66.7% for individuals with MCRsaf. In addition, the overlap between MCR and MCI was greater for na-MCI when compared to aMCI. This result concords with our previous study (Sekhon et al., 2017) and underscores the strong relationship between MCR and impaired cognitive performance, which explains the ability for both syndromes to predict dementia (Verghese et al., 2013, 2014; Beauchet et al., 2016; Sekhon et al., 2017). Cognition and locomotion are two human abilities, which are controlled by the brain (Nutt et al., 1993; Beauchet et al., 2009, 2012, 2016). Their decline is highly prevalent with physiological and pathological aging, and is greater than the simple sum of their respective prevalence, suggesting complex age-related interplay between cognition and locomotion (Nutt et al., 1993; Rosano et al., 2007; Beauchet et al., 2009, 2012, 2016; Wittenberg et al., 2017). Recently, a systematic review and meta-analysis provided evidence that poor gait performance could predict dementia (Beauchet et al., 2016). We have previously reported that individuals who exhibited both syndromes had poorer cognitive performance in all domains when compared to participants with MCI without MCR, and to participants with isolated MCR (Sekhon et al., 2017).

Furthermore, the present study concludes that MCR related positively to na-MCI and negatively to a-MCI. This result may be related to studies that reported executive dysfunction in individuals with MCR (Kumai et al., 2016; Belleville et al., 2017; Sekhon et al., 2017). This correlation between MCRs and na-MCI (but not with a-MCI) suggests that in our cohort (i.e., memory clinic based), MCRs is associated with an underlying non-AD process, such as vascular dementia or dementia with Lewy bodies, but not with an underlying AD process. This double dissociation is supported by the observation that at disease onset, gait speed is more affected in non-AD dementia than in AD dementia (Allali et al., 2016a). The absence of relationship between FTSS time and MCI status suggests that there is no interaction with cognitive performance in cognitively impaired individuals, such as MCI individuals. This result was not expected because of the

# REFERENCES

Abellan van Kan, G., Rolland, Y., Andrieu, S., Bauer, J., Beauchet, O., Bonnefoy, M., et al. (2009). Gait speed at usual pace as a predictor of adverse outcomes in community-dwelling older people an International Academy on Nutrition and Aging (IANA) task force. J. Nutr. Health Aging 13, 881–889. doi: 10.1007/ s12603-009-0246-z

previous positive relationship reported in CHI (Hirsch et al., 1997; Raji et al., 2002; Rosano et al., 2005; Annweiler et al., 2011), which underlines a non-linear complex relation between FTSS time and decline in cognitive performance. This relationship between MCRs and na-MCI supports that MCR appears to be a good predictor of non-Alzheimer's dementia, and in particular of vascular dementia (Verghese et al., 2013, 2014). Furthermore, the absence of any relationship with MCRf suggests that increased FTSS tracks a profile for older adults which is not relevant in identifying older adults who are at risk of dementia.

Our study has certain limitations. First, the cross-sectional design does not allow us to make any causal association. Secondly, the recruitment of participants was performed in one center. Thirdly, all participants in this study presented a cognitive complaint, preventing the generalization of study findings to all non-demented community-dwelling older adults. Indeed, the non-MCR/non-MCI participants cannot be considered as strictly cognitively intact, but as participants with subjective cognitive impairment (SCI). SCI is a prodromal state of MCI and is considered the earliest clinical stage of dementia (Jessen et al., 2014). Fourthly, we have no brain imaging data on the subset of GAIT participants selected for this study.

# CONCLUSION

The findings revealed that individuals with MCRf are distinct from those with MCRs. MCRf status does not relate to MCI status in the same way that MCRs does. A-MCI related negatively to MCRs, whereas it related positively to na-MCI. All these results suggest that using FTSS time in the definition of MCR is not appropriate in order to identify older adults who are at risk of dementia.

# AUTHOR CONTRIBUTIONS

HS and OB studied the concept and design. OB and CL acquired the data. HS, CL, GA, and OB analyzed and interpretated the data. HS drafted the manuscript. CL, JC, GA, and OB critically revised the manuscript for important intellectual content.

# FUNDING

The "Gait and Alzheimer Interactions Tracking" study was financially supported by the French Ministry of Health (Projet Hospitalier de Recherche Clinique National No. 2009- A00533-54).

Allali, G., Annweiler, C., Blumen, H. M., Callisaya, M. L., De Cock, A. M., Kressig, R. W., et al. (2016a). Gait phenotype from mild cognitive impairment to moderate dementia: results from the GOOD initiative. Eur. J. Neurol. 23, 527–541. doi: 10.1111/ene.12882

Allali, G., Ayers, E. I., and Verghese, J. (2016b). Motoric cognitive risk syndrome subtypes and cognitive profiles. J. Gerontol. A Biol. Sci. Med. Sci. 71, 378–384. doi: 10.1093/gerona/glv092



Wittenberg, E., Thompson, J., Nam, C. S., and Franz, J. R. (2017). Neuroimaging of human balance control: a systematic review. Front. Hum. Neurosci. 11:170. doi: 10.3389/fnhum.2017.00170

**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 handling Editor declared a past co-authorship with several of the authors GA and OB.

Copyright © 2019 Sekhon, Launay, Chabot, Allali and Beauchet. 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.

# Effects of Aging and Task Prioritization on Split-Belt Gait Adaptation

Danique Vervoort 1,2 \*, A. Rob den Otter <sup>1</sup> , Tom J. W. Buurke<sup>1</sup> , Nicolas Vuillerme2,3 , Tibor Hortobágyi <sup>1</sup> and Claudine J. C. Lamoth<sup>1</sup>

<sup>1</sup> Center for Human Movement Sciences, University of Groningen, University Medical Center Groningen, Groningen, Netherlands, <sup>2</sup> AGEIS, University Grenoble-Alpes, Grenoble, France, <sup>3</sup> Institut Universitaire de France, Paris, France

Background: Age-related changes in the sensorimotor system and cognition affect gait adaptation, especially when locomotion is combined with a cognitive task. Performing a dual-task can shift the focus of attention and thus require task prioritization, especially in older adults. To gain a better understanding of the age-related changes in the sensorimotor system, we examined how age and dual-tasking affect adaptive gait and task prioritization while walking on a split-belt treadmill.

Methods: Young (21.5 ± 1.0 years, n = 10) and older adults (67.8 ± 5.8 years, n = 12) walked on a split-belt treadmill with a 2:1 belt speed ratio, with and without a cognitive Auditory Stroop task. Symmetry in step length, limb excursion, and double support time, and strategy variables swing time and swing speed were compared between the tied-belt baseline (BL), early (EA) and late split-belt adaptation (LA), and early tied-belt post-adaptation (EP).

Results: Both age groups adapted to split-belt walking by re-establishing symmetry in step length and double support time. However, young and older adults differed on adaptation strategy. Older vs. young adults increased swing speed of the fast leg more during EA and LA (0.10–0.13 m/s), while young vs. older adults increased swing time of the fast leg more (2%). Dual-tasking affected limb excursion symmetry during EP. Cognitive task performance was 5–6% lower during EA compared to BL and LA in both age groups. Older vs. young adults had a lower cognitive task performance (max. 11% during EA).

Conclusion: Healthy older adults retain the ability to adapt to split-belt perturbations, but interestingly age affects adaptation strategy during split-belt walking. This age-related change in adaptation strategy possibly reflects a need to increase gait stability to prevent falling. The decline in cognitive task performance during early adaptation suggests task prioritization, especially in older adults. Thus, a challenging motor task, like split-belt adaptation, requires prioritization between the motor and cognitive task to prevent adverse outcomes. This suggests that task prioritization and adaptation strategy should be a focus in fall prevention interventions.

Keywords: split-belt walking, adaptive gait, aging, older adults, dual-task, task prioritization

#### Edited by:

Helena Blumen, Albert Einstein College of Medicine, United States

#### Reviewed by:

Pierfilippo De Sanctis, Albert Einstein College of Medicine, United States Gilles Allali, Geneva University Hospitals (HUG), Switzerland

#### \*Correspondence:

Danique Vervoort d.vervoort@umcg.nl

Received: 18 October 2018 Accepted: 11 January 2019 Published: 29 January 2019

#### Citation:

Vervoort D, den Otter AR, Buurke TJW, Vuillerme N, Hortobágyi T and Lamoth CJC (2019) Effects of Aging and Task Prioritization on Split-Belt Gait Adaptation. Front. Aging Neurosci. 11:10. doi: 10.3389/fnagi.2019.00010

# INTRODUCTION

Humans adapt their gait to environmental challenges, which allows walking on uneven surfaces, avoiding obstacles, and maintaining balance when slipping or tripping. Advancing age modifies the locomotor system, which reduces the ability of older adults to adapt to environmental perturbations while walking (Bierbaum et al., 2011; McCrum et al., 2016). The age-related changes include the distal-to-proximal shift in joint torques and powers (DeVita and Hortobagyi, 1985), and an increase of co-activation of agonist and antagonist lower extremity muscles (Schmitz et al., 2009). While the age-related changes increase joint stability, these changes may have negative effects, as the metabolic cost of walking increases ∼20% with aging (Hortobágyi et al., 2011). Not only are there quantitative adaptations, but age also modifies the strategies used to negotiate obstacle perturbations while walking. Older adults increase step length to avoid the obstacle, which increases the sense of stability by decreasing the forward momentum of the center of mass (Weerdesteyn et al., 2005a,b).

Much less is known however about whether and how older adults adapt their gait to asymmetrical gait perturbations that are ubiquitous in daily life. A good ability to react to such perturbations could be essential to maintain walking balance and prevent falls. Perturbation studies have shown that an increase in the ability to perform reactive responses is associated with a lower number of falls (McCrum et al., 2017). A split-belt treadmill allows us to study not only the reactive responses, but also locomotor adaptations to a sustained perturbation during walking in a controlled environment (Reisman et al., 2005). By setting each belt to a different speed, both the immediate changes in step characteristics, i.e., early adaptation, and the time course of adaptation over several minutes until late adaptation can be determined.

Split-belt walking initially creates an asymmetry in step length and double support time. Young adults adapt their gait to reestablish symmetry in both step length and double support time (Reisman et al., 2005). During the entire adaptation phase there is an asymmetry in limb excursion, i.e., stride length on the splitbelt treadmill (Hoogkamer et al., 2014), due to the longer stride on the fast belt during split-belt walking (Reisman et al., 2005). When returning to tied-belt walking, the so-called early postadaptation, participants show aftereffects, such as asymmetry in step length in the opposite direction from early adaptation (Reisman et al., 2005). An inability to re-establish symmetry in step length and double support time during split-belt adaptation and the absence of aftereffects are presumable markers of reduced gait adaptability (Vasudevan et al., 2011; Bruijn et al., 2012). The early detection of reduced gait adaptability can help in identifying older adults at risk for adverse reactions to perturbations and thus prevent falls.

Healthy older adults are capable of adapting to split-belt walking, but age affects adaptation strategy, especially during the early adaptation phase. Older adults are termed as "speed" adaptors, with a ±0.15 m/s greater decrease in swing speed of the slow leg. In contrast, young adults are "timing" adapters, indicated by a ±5% shorter swing time of the slow leg. To the best of our knowledge, the previously mentioned study is the only work that examined the effects of age on gait adaptation strategies (Bruijn et al., 2012).

Besides single-task gait adaptation, dual-task gait adaptation is of interest. Distraction-free gait is rare and dual-tasking is rather the norm than the exception in daily life. There is however no consensus concerning the effects of motor-cognitive dual-tasking on gait adaptation. Dual-tasking slowed the rate of adaptation on step length symmetry in the adaptation phase in young (Malone and Bastian, 2010) and in middle-aged adults (Malone and Bastian, 2016). Motor-cognitive dual-tasking also increased stance time on the fast belt and double support time on the slow belt in young adults (McFadyen et al., 2009), and another study reported that older adults exhibited a larger step time asymmetry (Saito et al., 2013).

The inconsistent results concerning motor-cognitive dualtasking might be due to the types of cognitive tasks performed or the gait outcomes used, but could also be due to task prioritization. An integrated task prioritization model suggests that healthy adults perform and focus on the secondary cognitive task as long as the threat to postural control is low. A challenging environment or a demanding motor task can shift the focus of attention from the secondary cognitive task to the motor task in order to maintain gait (Yogev-Seligmann et al., 2012).

While young adults can flexibly allocate attentional resources between two concurrent tasks (Raffegeau et al., 2018), attentional capacity decreases with age, thus motor-cognitive dual-tasking becomes more challenging (Huxhold et al., 2006). During dualtask split-belt walking, especially older adults may prioritize gait in order to adequately adapt to the perturbation of the split-belt. Indeed, after short perturbations while performing a cognitive task, older adults prioritized dynamic stability, as shown by a sharp increase in the number of errors on the cognitive task (Mersmann et al., 2013). While walking on elevated or narrow surfaces with a dual-task, older adults not only increased the number of errors on the cognitive task, but also committed more missteps, suggesting that prioritization of the motor task in high-risk settings might fail for older adults (Schaefer et al., 2015).

The ability to adapt to the perturbation induced by the split-belt treadmill is essential to continue walking. However, this ability might be altered or affected by age, dual-tasking or task prioritization. Therefore, our aims were to determine: (1) The effect of split-belt adaptation and age on gait strategy and symmetry; (2) The effect of motor-cognitive dual-tasking on gait adaptation, and (3) Age-related differences in task prioritization during split-belt adaptation.

We hypothesized that adaptations to split-belt walking occur independent of age, but young vs. older adults are "timing" and "speed" adaptors, respectively. We also expected that motorcognitive dual-tasking would affect the ability to adapt to

**Abbreviations:** BL, late slow baseline; EA, early adaptation; LA, late adaptation; EP, early post-adaptation; SL, step length; LE, limb excursion; DS, double support time; SLS, step length symmetry; LES, limb excursion symmetry; DSS, double support symmetry; SwT, swing time; SwS, swing speed; DTC, dual-task cost; CTP, cognitive task performance.

split-belt walking in the spatial or temporal gait outcomes. Furthermore, we hypothesized older vs. young adults will prioritize the motor adaptation task over the cognitive task in the most challenging early period of adaptation to split-belt walking.

# MATERIALS AND METHODS

# Participants

Healthy young (21.5 ± 1.0 years, 40% male, n = 10) and older adults (67.8 ± 5.8 years, 58% male, n = 12) who could walk without aids and follow verbal instructions were included in the study. Criteria for exclusion were orthopedic, neurological, and/or psychiatric disorders that might affect gait, and prior experience with split-belt walking. The Ethical Committee of the Center of Human Movement Sciences at the University Medical Center Groningen approved the study. All participants signed a written informed consent before the measurements.

# Instrumentation and Procedure

Participants walked on an M-Gait treadmill (Motekforce Link, Amsterdam, The Netherlands), with two belts that can be controlled separately. With the embedded force plates in the treadmill, ground reaction forces were sampled at 1,000 Hz. Infrared emitting diodes were placed on the feet (5th metatarsal head) and ankle (lateral malleolus) and recorded at 100 Hz (Optotrak, Northern Digital, Ontario, Canada).

Participants walked on the split-belt treadmill, starting with 3 min of tied-belt walking at both 1.0 m/s and 0.5 m/s (baseline), then split-belt walking with one belt moving at 1.0 m/s and the other belt at 0.5 m/s for 6 min (adaptation) and finally with tied-belts at 0.5 m/s for 6 min (post-adaptation; **Figure 1**).

In the first condition (single-task), participants walked on the split-belt treadmill for a total of 18 min without a cognitive task. In the second condition (dual-task), participants walked while performing a cognitive task. This task order was chosen in order to minimize the learning effect of re-exposure to splitbelt walking, which could potentially affect single-task adaptation and task prioritization during the second condition. Even though we cannot exclude that there were effects of motor learning in the second condition, due to the first condition, earlier studies have revealed that participants still show adaptation during re-exposure to the split-belt (Malone et al., 2011; Malone and Bastian, 2016), allowing us to test our hypothesis on task prioritization.

Between the two conditions, participants could rest for 2 min or longer if needed. The fast belt was randomly assigned to the left or right leg. In the dual-task condition, the fast belt was assigned to the other side as in the single-task condition, in order to minimize the learning effect of re-exposure to the split-belt.

The cognitive task was the Auditory Stroop test. The Auditory Stroop test consists of the words "high" and "low" in a high or low pitch. Participants were instructed to call out the pitch of the word they heard, ignoring the actual word presented (McFadyen et al., 2009). Participants had to be able to hear the words of the Auditory Stroop test. The test was performed in a control condition while sitting and during the dual-task condition at the last minute of the slow baseline (BL) until the first minute of adaptation (EA); the last minute of adaptation (LA) until the first minute of post-adaptation (EP); and the last minute of postadaptation (LP; **Figure 1**). No instructions on task prioritization were given. Verbal responses were recorded to determine the number of correct responses.

# Data Analysis

Data were analyzed off-line with custom made Matlab codes (R2015b, The MathWorks Inc.). Vertical ground reaction forces were filtered with a second-order low-pass Butterworth filter (15 Hz cut-off). Data from the Optotrak markers were filtered with a second-order high-pass Butterworth filter (0.5 Hz cut-off).

Heel-strike and toe-off were determined at the moment the vertical forces crossed the threshold of 50 N. The foot contact moments were then used to calculate the outcome variables for the first and last five strides of the BL, EA, LA, and EP phases, allowing the following three comparisons: EA vs. BL (effect of the perturbation), LA vs. EA (adaptive change), EP vs. BL (aftereffects). For both spatial variables, limb excursion and step length, the foot contact moments were resampled to match the Optotrak sample frequency.

Symmetry variables were calculated for spatial variables step length (SL) and limb excursion (LE) and the temporal variable double support time (DS). SL was calculated as the anterior posterior distance between the ankle markers of both legs at heelstrike of the leading leg (Reisman et al., 2005). With x as the position of the lateral malleolus marker (latmal) at the i th sample.

$$SL\_{fast}(\mathbf{i}) = \left. \times\_{latmal\_{fast}} \left[ t\_{helstrike\ fast} \begin{array}{c} \mathbf{(i)} \right] - \left. \chi\_{latmal\_{slow}} \begin{array}{c} \left[ t\_{helstrike\ fast} \begin{array}{c} \mathbf{(i)} \right] \end{array} \right. \end{array} \right]$$

LE was defined as the distance traveled by the ankle marker in the anterior-posterior direction from heel-strike to toe-off of one limb (Hoogkamer et al., 2014).

$$LE\left(i\right) = \chi\_{latmal} \left[t\_{healstrike}(i)\right] - \chi\_{latmal} \left[t\_{toeoff}(i)\right]$$

DS was defined as the time (t) both feet were in contact with the ground (Reisman et al., 2005).

$$DS\_{\text{fast}}(i) = \mathbf{t}\_{\text{toeoff slow}}(i) - \mathbf{t}\_{\text{heelstrike fast}}(i)$$

Symmetry in SL (SLS), LE (LES), and DS (DSS) was calculated as follows (Malone and Bastian, 2010):

$$\text{Symmetry (i)} = \frac{\text{Fast (i)} - \text{Slow}(i)}{\text{Fast (i)} + \text{Slow}(i)}$$

To reduce the SL asymmetry induced by the split-belt perturbation, the fast leg needs to be placed further forward than the slow leg. This can be achieved by using two strategies: 1) increasing the time spent in swing, or 2) increasing swing speed. Therefore, the strategy variables percentage swing time (SwT) and swing speed (SwS) were calculated as (Bruijn et al., 2012):

$$\text{Sw}\_{\text{fast}/\text{slow}}(i) = \frac{t\_{\text{helstrike}(i)} - t\_{\text{toe}} \text{gf}(i)}{t\_{\text{helstrike}(i+1)} - t\_{\text{helstrike}(i)}} \times 100$$
 
$$\text{SwS}\_{\text{fast}/\text{slow}}(i) = \frac{LE(i)}{t\_{\text{helriteke}(i)} - t\_{\text{toe}} \text{gf}(i)}$$

For all gait variables, the dual-task cost (DTC) was determined from the single and dual-task condition values, with DTC values above zero indicating a larger value in the single-task condition and values below zero indicating that there was an effect of the dual-task (Raffegeau et al., 2018).

$$DTC = \frac{(single\ task - dual\ task)}{single\ task} \* 100$$

Cognitive task performance (CTP) was calculated with the following formula, with n as the number of stimuli. A values of one indicates a perfect performance.

$$CTP = \frac{n\_{correct}}{n\_{total}}$$

# Statistical Analysis

To examine the effect of split-belt walking and if young and older adults differ on split-belt adaptation (aim 1), a repeated measures ANOVA was conducted with within factor Phase (BL-EA-LA-EP) and between factor Group (young vs. older adults) during single-task split-belt walking for the dependent gait variables. When a main effect of phase was found, a post-hoc dependent samples t-test with Holm-Bonferroni correction was applied to the following phase comparisons: EA vs. BL (effect of the perturbation), LA vs. EA (adaptive change), EP vs. BL (aftereffects). For significant interaction effects, the difference between age groups was tested with an independent samples t-test for each phase separately.

To address the second aim, the differences between single and dual-task split-belt walking were assessed using planned comparison t-tests to determine if DTC was different from zero for each of the four phases (BL-EA-LA-EP).

For the third aim, the differences in prioritization during dual-task split-belt adaptation between young and older adults, adaptation effects were first tested with a repeated measures ANOVA for the dual-task condition with within factor Phase (BL-EA-LA-EP) and between factor Group (young vs. older adults) for all gait variables. Post-hoc testing was done similarly as for the first aim. The cognitive task performance was tested with a similar repeated measures ANOVA with the CTP as dependent variable. Differences between the age groups during the control task (sitting) were tested with an independent samples t-test to test if there were differences between the age groups during the single task. Statistical analysis was performed using SPSS (24.0, IBM Corp. Armonk, NY, USA). Level of significance was set at p < 0.05.

# RESULTS

# Single-Task Split-Belt Adaptation and Differences Between Young and Older Adults

For the single-task split-belt condition, a significant main effect of phase was found for the symmetry variables, step length symmetry [SLS; F(2.3, 47) = 28.3; p < 0.001], limb excursion symmetry [LES; F(1.8, 35) = 130.8; p < 0.001] and double support symmetry [DSS; F(2.1, 42) = 46.0; p < 0.001], as well as for the strategy variables, swing time of the fast [SwTfast; F(3,60) = 122.6; p < 0.001] and slow leg [SwTslow; F(2.3, 46) = 23.0; p < 0.001] and swing speed of the fast [SwSfast; F(3, 60) = 102.5; p < 0.001] and slow leg [SwSslow; F(2.2, 44) = 64.1; p < 0.001]. Post-hoc testing revealed that for the symmetry variables SLS and DSS, an asymmetry occurred in early adaptation (EA) due to the perturbation of the changing belt speeds, while symmetry was re-established in late adaptation (LA). The early post-adaptation phase (EP) showed aftereffects of SLS and DSS asymmetry in the opposite direction from early adaptation. During the entire adaptation phase, both early and late adaptation, there is an asymmetry in LES due to the longer stride on the fast belt (see **Figure 2**). For the strategy variables, an increase was seen in swing time of the fast leg and swing speed for both

FIGURE 2 | Adaptation plots of the symmetry variables for young and older adults during the single-task split-belt condition. The adaptation plot shows the development of the mean and standard deviation of the symmetry variables over the split-belt condition, starting with the 90 steps of the slow baseline, then 230 steps of the adaptation phase and then the post-adaptation phase with 180 steps. The symmetry values for step length (A,B), limb excursion (G,H), and double support time (M,N) and the separate values for the fast and slow leg [step length (C–F), limb excursion (I–L), and double support time (O–R)] are shown for young (left) and older adults (right).

legs during early adaptation, while there was a decrease in swing time of the slow leg. While swing time of the fast leg and swing speed of the slow leg slightly decreased until late adaptation, swing speed of the fast leg continued to increase (see **Figure 3**). **Table 1** presents the direction of the changes of the gait variables over the phases and the corresponding post-hoc statistics.

There were no main effects of group for the gait variables, but Phase by Group interactions were significant for LES [F(1.8, 35) = 4.4; p = 0.024], SwTfast [F(3, 60) = 4.8; p = 0.006], SwSfast

FIGURE 3 | Adaptation plots of the strategy variables for young and older adults during the single-task split-belt condition. The adaptation plot shows the development of the mean and standard deviation of the strategy variables swing time (A,B) and swing speed (C,D) over the split-belt condition. The adaptation plot starts with the 90 steps of the slow baseline, then 230 steps of the adaptation phase and then the post-adaptation phase with 180 steps. Values of the fast leg are shown in black, values of the slow leg are shown in gray.

TABLE 1 | Post-hoc tests for the main effect of phase in the single-task split-belt condition.


The table presents results of the post-hoc t-tests, with the t-value [t(df) ], the p-value and the differences (Diff) between the two phases that were tested. Diff values are mean ± standard deviations. P-values are highlighted in bold if the phases were significantly different after the Holm-Bonferroni correction. Diff, difference score; BL, baseline; EA, early adaptation; LA, late adaptation, EP, early post-adaptation; SLS, step length symmetry; LES, limb excursion symmetry; DSS, double support symmetry; SwT, swing time; SwS, swing speed.

[F(3, 60) = 5.3; p = 0.003], and SwSslow [F(2.2, 44) = 4.2; p = 0.018]. Post-hoc testing revealed that during baseline (BL) there was a slight asymmetry of LES in opposite directions in young and older adults [t(20) = 2.3; p = 0.033; see **Figure 4**]. In early adaptation there was a difference of 0.06 greater asymmetry in LES in older vs. young adults [t(20) = −2.5; p = 0.023]. SwTfast showed that young adults had a trend of ±2% higher swing time of the fast leg compared to older adults during early [t(20) = 1.8; p = 0.094] and late adaptation [t(20) = 1.7; p = 0.100]. During both early and late adaptation older vs. young adults had, respectively a 0.13 and 0.10 m/s higher swing speed of the fast leg [respectively t(20) = −2.3; p = 0.034 and t(20) = −2.3; p = 0.032; see **Figure 4**]. Older compared to young adults had a 0.09 m/s higher swing speed of the slow leg during baseline [t(20) = −3.0; p = 0.006].

# Difference Between Single and Dual-Task Split-Belt Walking

No significant effects for dual task cost were found between the single and dual-task condition, except for LES during early post-adaptation [t(21) = 2.3; p = 0.033]. During motor-cognitive dual-tasking, the aftereffects of LES were lower than during the single-task.

# Dual-Task Split-Belt Adaptation for Young and Older Adults

There was a significant main effect of Phase for both symmetry, SLS [F(2.3, 47) = 28.3; p < 0.001], LES [F(1.8, 35) = 130.8; p < 0.001] and DSS [F(2.1, 42) = 46.0; p < 0.001], and strategy gait variables, SwTfast [F(3,60) = 122.6; p < 0.001], SwTslow

FIGURE 4 | Bar plots of the symmetry (SLS, LES, DSS) and the strategy variables (Swing time, Swing speed) for young and older adults on the single-task split-belt condition. The four phases: baseline (BL), early adaptation (EA), late adaptation (LA), and early post-adaptation (EP) are shown next to each other. The bar plots show the median, the upper, and lower quartiles and the min and max value of the age groups. The dots show the individual data of the young (black) and older adults (blue). There were interaction effects for LES, Swing time of the fast leg, and Swing speed of both legs. SLS, step length symmetry; LES, limb excursion symmetry; DSS, double support symmetry.

[F(2.3, 46) = 23.0; p < 0.001], SwSfast [F(3, 60) = 102.5; p < 0.001], and SwSslow [F(2.2, 44) = 64.1; p < 0.001], similar to the single-task condition. Post-hoctesting revealed that the symmetry variables SLS and DSS showed the similar pattern of asymmetry during early adaptation and re-established symmetry during late adaptation, with opposite aftereffects during early postadaptation (see **Table 2** for the direction and post-hoc statistics of all gait variables). Thus, there was still adaptation during this second exposure of split-belt walking necessary for testing task prioritization (see **Figures 5**, **6**). No significant age or interaction effects were found in any of the gait variables.

# Differences in Cognitive Task Performance Between Young and Older Adults

In the seated control condition, the two age groups did not differ in cognitive task performance (CTP; Young: 0.95 ± 0.05, Old: 0.96 ± 0.07; see **Figure 7**).


TABLE 2 | Post-hoc tests for the main effect of phase in the dual-task split-belt condition.

The table presents results of the post-hoc t-test, with the t-value [t(df) ], the p-value and the differences (Diff) between the two phases that were tested. Diff values are mean ± standard deviations. P-values are highlighted in bold if the phases were significantly different after the Holm-Bonferroni correction. Abbreviations: Diff, difference score; BL, baseline; EA, early adaptation; LA, late adaptation; EP, early post-adaptation; SLS, step length symmetry; LES, limb excursion symmetry; DSS, double support symmetry; SwT, swing time; SwS, swing speed.

During motor-cognitive dual-tasking, there was a significant main effect of phase for CTP [F(2.3, 46) = 6.5; p = 0.002]. Post-hoc testing showed that all adults made on average 5–6% more errors on the Auditory Stroop task during early adaptation compared to their performance during baseline [t(21) = 3.0; p = 0.006] and late adaptation [t(21) = −3.1; p = 0.006]. There was a significant main effect of group on CTP over all the split-belt phases (BL-EA-LA-EP). Young vs. old adults performed better on the cognitive task [F(1, 20) = 11.0; p = 0.003; see **Figure 7**], with the largest difference of 11% seen in EA.

# DISCUSSION

The overall aim of the present study was to gain insight into the effects of age and dual-tasking on adaptation to a perturbation induced by a split-belt treadmill and task prioritization. More specifically, we examined the effects of age and dual-tasking on gait symmetry and strategy.

Both young and older adults adapted to split-belt walking and re-established gait symmetry, but the two age groups achieved this using a different strategy. Older adults increased swing speed of the fast leg, whereas young adults showed a trend of increased swing time of the fast leg. Dual-tasking compared with single-task split-belt walking did not affect gait adaptation strategies, but only affected limb excursion symmetry during the early post-adaptation phase, as indicated by smaller aftereffects. The lack of dual-task effects on all other gait variables and phases is likely due to task prioritization. Task prioritization is clearly present within the dual-task condition in the early adaptation phase, as revealed by a worse cognitive task performance, even more so for older compared to young adults. We discuss these results with a perspective on how healthy older adults retain the ability to adapt to the split-belt perturbation.

In line with results of previous research (Reisman et al., 2005; Malone and Bastian, 2016), both young and older adults reestablished symmetry in step length and double support time after the initial perturbation. The re-established symmetry by the older adults in this study indicate a retained ability to continuously adapt to split-belt walking. Although symmetry variables are widely reported and nicely reflect short as well as longer term split-belt adaptation, it is nevertheless important to examine the contribution of both legs (fast and slow) to clearly show the effects of gait adaptation (Hoogkamer et al., 2014). The effects on gait symmetry can be harder to distinguish at lower speeds or speed ratios, like the 1:2 speed ratio of 0.5–1 m/s used in this study for feasibility. In the current study, we found most of the well-documented adaptation effects on gait symmetry, even with the limitations of these lower speeds and a small sample size with some variation within the age groups (see **Figure 4**). It would therefore be interesting to further examine adaptive gait and the variation between participants with a larger sample size at slightly higher speeds to show even clearer differences within and between groups.

While both age groups adapt to split-belt walking, our results confirm previous data showing that the adaptation strategies were age-dependent (Bruijn et al., 2012). Swing time of the fast leg showed a trend with a 2% larger increase for young compared to older adults, while swing speed of the fast leg increased 0.10–0.13 m/s more for older than young adults. With the limitation of a small sample size in the current study, only a trend of adaptation in 'timing' for young adults was found, but a larger sample size in future studies could further confirm the adaptation strategies. The current results do agree with the concept that adaptations occur in 'timing' and 'speed' in young and older adults, respectively (Bruijn et al., 2012). However, we found the 'timing' and 'speed' effects for the fast instead of the slow leg, possibly due to the treadmill being stopped between the baseline and adaptation phase in the previous study, while our participants continued walking. This causes a difference in acceleration of the belts between the two studies in that in our study only one belt accelerated with max 0.5 m/s<sup>2</sup> , while in the previous study both belts accelerated with max 0.3 m/s<sup>2</sup> (Bruijn et al., 2012).

development of the mean and standard deviation of the symmetry variables over the split-belt condition, starting with the 90 steps of the slow baseline, then 230 steps of the adaptation phase and then the post-adaptation phase with 180 steps. The symmetry values for step length (A,B), limb excursion (G,H), and double support time (M,N) and the separate values for the fast and slow leg [step length (C–F), limb excursion (I–L), and double support time (O–R)] are shown for young (left) and older adults (right).

The age-related change in adaptation strategy might be beneficial for older adults. Switching from a time to a speed strategy in response to a split-belt perturbation decreases the time spent in swing and thus decreases single leg stance, which increases gait stability during split-belt walking, especially immediately after the perturbation (Bruijn et al., 2012). Maintaining dynamic stability during split-belt walking is also important over the longer period of adaptation, as shown by the adaptation of gait stability during the continuous perturbation of split-belt walking (Park and Finley, 2017; Buurke et al., 2018). Furthermore, asymmetry in gait is associated with poor dynamic stability in stroke survivors (Lewek et al., 2014). In the context

FIGURE 6 | Adaptation plots of the strategy variables for young and older adults during the dual-task split-belt condition. The adaptation plot shows the development of the mean and standard deviation of the strategy variables swing time (A,B) and swing speed (C,D) over the split-belt condition. The adaptation plot starts with the 90 steps of the slow baseline, then 230 steps of the adaptation phase and then the post-adaptation phase with 180 steps. Values of the fast leg are shown in black, values of the slow leg are shown in gray.

of the present study, the age-related differences in strategy to re-establish gait symmetry might be essential for older adults to maintain gait stability while walking on the split-belt to prevent falling.

Beyond gait stability, age-related decreases in neuromuscular function could contribute to the age-related differences in timing and speed strategies. The age-related reductions in leg muscle strength (Nigg et al., 1994; Hayashida et al., 2014) and power (Bean et al., 2002) are associated with changes in the walking pattern including a slowing of gait speed, an increase in stance and double support time, and a decrease in swing time (Winter et al., 1990; Samson et al., 2001; Laufer, 2005). The neuromuscular changes might thus limit swing time during split-belt walking in old adults. Future research should determine the relationship between the age-related changes in neuromuscular function and the gait adaptation strategies observed in the present study. A comprehensive analysis of muscle activation patterns during split-belt walking, in both timing and contributions of muscles, could provide further insights into neuromuscular mechanisms underlying the age-dependent variations in gait adaptations.

Dual-tasking did not affect gait adaptation as there were no significant effects on dual-task cost, except for the aftereffects of limb excursion symmetry. The smaller aftereffects could be due to the fact that the dual-task caused participants to retain less of the split-belt adaptation. Since there was no randomization between the two conditions, we cannot exclude that fatigue might have had effects on the dual-task results, which is a limitation of the study. The lack of further dual-task effects is in contrast to the previously discussed motor-cognitive dual-tasking studies during split-belt walking (McFadyen et al., 2009; Malone and Bastian, 2010, 2016; Saito et al., 2013). We however propose that participants prioritized gait over the cognitive task, minimizing any effects of the secondary cognitive task on the gait variables.

At the onset of the speed differences between the belts, the early adaptation phase, both young and older adults performed worse on the cognitive task, as indicated by fewer responses to the Auditory Stroop stimuli compared to baseline and late adaptation. This result suggests that immediately after being exposed to split-belt walking there is a need to prioritize gait over performing the cognitive task for the sake of safety. Our findings support the theory of task prioritization (Yogev-Seligmann et al., 2012) that people tend to prioritize the motor task over the cognitive task in a more challenging environment, in this case, a challenging motor task, which requires attentional capacity to maintain gait.

Furthermore, during dual-task split-belt walking older vs. young adults had 3–11% fewer correct responses on the cognitive task, with the largest differences also during the early adaptation phase. This implies that performing an additional task while reacting to the split-belt perturbation is harder for older adults. Therefore, it seems that older adults even more than young adults need to prioritize gait. The increased need to prioritize gait may be due to age-related changes. Interference from dual-tasking on walking ability starts at a lower level for individuals with less taskrelevant resources, like older adults (Huxhold et al., 2006). Older adults might also have a higher need to focus their attention on foot placement in response to the split-belt perturbation, since afferent feedback is impaired in old age (Goble et al., 2009). Afferent feedback facilitates adaptive gait, and a reduction in feedback from muscles leads to a poorer detection of errors that are important for accurate gait adaptation (Bastian, 2011). By prioritizing the motor task over the cognitive task during splitbelt walking, older adults show that they have retained the ability to compensate for the age-related neuromuscular decline, which could help in preventing adverse reactions to perturbations.

# CONCLUSIONS

Age did not affect gait symmetry after a split-belt perturbation, but did affect adaptation strategy, with young and older adults adapting through "timing" and "speed," respectively. The role of this change in adaptation strategy is likely to increase gait stability in older adults to prevent falling. Task prioritization of the motor over the cognitive task may underlie the lack of dual-tasking effects on gait adaptation. This is supported by a decline in cognitive task performance during early adaptation, more so in older compared with young adults. We conclude that healthy older adults retain the ability to adapt to split-belt perturbations, but interestingly they adapt through a different adaptation strategy. Moreover, in challenging motor tasks, like split-belt adaptation, this requires task prioritization to

# REFERENCES


prevent adverse outcomes. This suggests that task prioritization and adaptation strategy should be a focus in fall prevention interventions. Furthermore, future research should determine the relationship between gait adaptation strategies, gait stability, and neuromuscular function to understand the underlying mechanisms of age-related differences in split-belt adaptation.

# AVAILABILITY OF DATA AND MATERIAL

The data set used and analyzed in the current study is available from the corresponding author on reasonable request.

# AUTHOR CONTRIBUTIONS

DV, CL, TH, and NV designed the study protocol. DV collected and analyzed the data with supervision of CL. Results were interpreted by DV, AdO, NV, TB, TH, and CL. DV wrote the first draft under supervision of CL. DV, CL, AdO, NV, TB, and TH contributed significantly to revising the manuscript. All authors read and approved the final manuscript.

# FUNDING

The publication of this work was supported by the French National Research Agency in the framework of the Investissements d'avenir program (ANR-10-AIRT-05 and ANR-15-IDEX-02).

# ACKNOWLEDGMENTS

We thankfully acknowledge all participants. We thank Anne de Hoop and Ylva Visser for their help during the measurements, and Wim Kaan, Dirk van der Meer, and Emyl Smid for their technical support.


speed in healthy subjects due to age, height and body weight. Aging 13, 16–21. doi: 10.1007/BF03351489


**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 Vervoort, den Otter, Buurke, Vuillerme, Hortobágyi and Lamoth. 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.

# Instability Resistance Training Decreases Motor Noise During Challenging Walking Tasks in Older Adults: A 10-Week Double-Blinded RCT

#### Nils Eckardt1,2 \* and Noah J. Rosenblatt<sup>3</sup>

<sup>1</sup> Department of Training and Movement Science, Institute for Sport and Sports Science, University of Kassel, Kassel, Germany, <sup>2</sup> Department of Sport and Movement Science, Institute of Sport Science, Carl von Ossietzky University of Oldenburg, Oldenburg, Germany, <sup>3</sup> Dr. William M. Scholl College of Podiatric Medicine's Center for Lower Extremity Ambulatory Research (CLEAR), Rosalind Franklin University of Medicine and Science, North Chicago, IL, United States

#### Edited by:

Eric Yiou, Université Paris-Sud, France

#### Reviewed by:

Sjoerd Bruijn, VU University Amsterdam, Netherlands Didier Pradon, INSERM U1179 Handicap Neuromusculaire: Physiopathologie, Biothérapie et Pharmacologie Appliquées (END-ICAP), France

> \*Correspondence: Nils Eckardt nils.eckardt@uol.de

Received: 28 August 2018 Accepted: 04 February 2019 Published: 27 February 2019

#### Citation:

Eckardt N and Rosenblatt NJ (2019) Instability Resistance Training Decreases Motor Noise During Challenging Walking Tasks in Older Adults: A 10-Week Double-Blinded RCT. Front. Aging Neurosci. 11:32. doi: 10.3389/fnagi.2019.00032 Locomotor stability is challenged by internal perturbations, e.g., motor noise, and external perturbations, e.g., changes in surface compliance. One means to compensate for such perturbations is to employ motor synergies, defined here as co-variation among a set of elements that acts to stabilize, or provide similar trial-to-trial (or step-to-step) output, even in the presence of small variations in initial conditions. Whereas evidence exists that synergies related to the upper extremities can be trained, the extent to which lower limb synergies, such as those which may be needed to successfully locomote in complex environments, remains unknown. The purpose of this study was to evaluate if resistance training (RT) in unstable environments could promote coordination patterns associated with stronger synergies during gait. Sixty-eight participants between the age of 65 and 80 were randomly assigned to one of three different RT modalities: stable whole-limb machine-based RT (S-MRT), instability free-weight RT (I-FRT), and stable machine-based adductor/abductor RT (S-MRTHIP). Before and after RT, participants walked across an even lab floor and a more challenging uneven surface with and without holding a weighted bag. The uncontrolled manifold control analysis (UCM) was used to calculate the synergy index (i.e., strength of the kinematic synergy) related to stabilization of our performance variable, the mediolateral trajectory of the swing foot, under each condition. Regardless of RT group, there was no effect of RT on the synergy index when walking across the even lab floor. However, the synergy index during the two uneven surface conditions was stronger after I-FRT but was not affected by the other RT modalities. The stronger synergy index for the I-FRT group was due to improved coordination as quantified by an overall increase in variability in elemental variable space but a decrease in the variability that negatively affects performance. The unstable environment offered by I-FRT allows for exploration of motor solutions in a manner that

**94**

appears to transfer to challenging locomotor tasks. Introducing tasks that promote, rather than limit, exploration of motor solutions seems to be a valuable exercise modality to strengthen kinematic synergies that cannot be achieved with traditional strengthening paradigms (e.g., S-MRT).

Clinical Trial Registration: www.ClinicalTrials.gov, identifier NCT03017365.

Keywords: irregular surface, unstable resistance training, uncontrolled manifold, motor redundancy, elderly, gait, perturbation

# INTRODUCTION

Falls are a leading cause of injuries and mortality in older adults and the risk of falling increases with age (Rubenstein, 2006). Many falls in community-dwelling older adults occur during locomotion, particularly when postural stability is challenged by perturbations like slips and trips (Berg et al., 1997). Locomotor stability is generally realized through accurate positioning of the swing-foot relative to the center of mass (CoM) (Bruijn and van Dieën, 2018). In the frontal plane this requires active control by the central nervous system (Kuo, 2002; Bruijn and van Dieën, 2018) realized through activation of the swing limb hip musculature in response to states of the stance limb (Bruijn and van Dieën, 2018). However, increased neuromotor noise and associated motor variability (Kang and Dingwell, 2009) may negatively affects control of mediolateral (ML) foot placement and increases variability in ML placement of the foot (Bruijn and van Dieën, 2018), which may increase fall risk (Brach et al., 2005). Nonetheless, if an individual can compensate for increased variability, by channeling it into a subspace that does not affect performance, then the high variability would not be hazardous. The uncontrolled manifold (UCM) analysis provides a means to quantify the extent to which motor variability may or may not "be hazardous" (Latash et al., 2007).

The UCM analysis quantifies the extent to which all available degrees of freedom (DoF) that contribute to a task-relevant performance variable co-vary so as to stabilize (limit trial-totrial variation in) that performance variable (Scholz and Schöner, 1999; Latash et al., 2007). The analysis decomposes variability in a set of elemental variables into two components: "good" variance that has no effect on the performance variable and "bad" variance that results into deviations of the performance variable. A positive synergy index, which quantifies the relative amount of "good" variance compared to "bad" variance, implies that the performance variable is stabilized by a synergy (Scholz and Schöner, 1999; Latash et al., 2007). Such stabilization allows secondary tasks that rely on the same set of elemental variables to be performed without affecting the primary task. It is generally thought that the effects of aging on motor coordination manifest as low amounts of "good variance" (Kapur et al., 2010). However, in response to challenging locomotor conditions such as uneven surface walking, healthy community-dwelling older adults are able to counteract perturbation-related increases in "bad" variability by channeling elemental variability into "good" variability (Eckardt and Rosenblatt, 2018). Similar findings have been reported for upper extremity tasks (Kapur et al., 2010). Thus, the aging human CNS possess the ability to harness motor flexibility, i.e., to increase the synergy index by increasing "good" variance through exploration of motor solution space. Given its' importance, there is a need to understand whether this ability is trainable and if so, what exercises are optimal to train this ability.

With regard to the upper extremity, it has been demonstrated that non-repetitive tasks performed under conditions of manipulated stability can help promote large amounts of "good" variance (Shim et al., 2008; Wu and Latash, 2014). Similarly it has been suggested that exercise interventions which introduce tasks that promote, rather than limit, exploration of motor solutions may be particularly appropriate for promoting motor flexibility and the coordination patterns utilized to ambulate in complex environments (Rosenblatt et al., 2014). In turn, such exercises may help to reduce risk of falling and their introduction into fall prevention interventions would represents a considerable departure from traditional exercises, such as resistance training (RT) that targets muscle strength and power to achieve this goal (Benichou and Lord, 2016). Indeed, machine-based resistance training (S-MRT) does not seem suited to promote exploration of motor solutions due to restricted movements during exercise execution; a combination of balance and resistance training, i.e., "instability free-weight resistance training" (I-FRT), may be better suited to do this.

Instability free-weight RT is an exercise modality which involves tasks that specifically promote exploration of motor solutions by having participants engage in RT training while standing on destabilizing surfaces. The inherent instability during execution of the I-FRT results in greater overall muscle activation of the lower limbs due to the constant need for postural readjustment (Lawrence and Carlson, 2015). Indeed, the greater demands of I-FRT may explain why in our recent 10-week RCT I-FRT elicited similar increases in balance, power, and strength in older adults compared to S-MRT despite using half the training load (Eckardt, 2016). We hypothesized that inter-and intramuscular coordination may be the driving reason for increases in the respective outcomes. Nonetheless, strength, power, and balance are measures of performance that do not directly address changes in coordination such as those quantified by synergies within the UCM analysis. There is evidence, albeit limited, that RT can improve synergies; one study has evaluated the effects of finger RT on finger synergies, independence, force control and adaptations in multi-finger coordination (Shim et al., 2008). If RT does impact synergies, then RT focusing on the hip could be particularly beneficial with regards to improving kinematic synergies related to the mediolateral trajectory of the

swing-foot during gait. Indeed, swing limb hip abductors activity is critical in modulating foot placement and is predicted by the relationship between the CoM and the stance limb (Rankin et al., 2014; Roden-Reynolds et al., 2015). On the other hand, strengthening may not significantly affect swing limb mechanics, which in part contribute to kinematic synergies related to foot placement. For example, increasing strength by 26% (i.e., control condition relative to a weaker nerve block condition) does not affect swing limb kinematics (Pohl et al., 2015). In fact, it is entirely possible that weaker older adults employ stronger synergies to compensate for weakness, as has been argued to occur during sit-to-stand tasks (Greve et al., 2013) such that hip strengthening could lead to a reduction in synergies. Thus, the extent to which RT, and particularly hip-specific RT, can impact motor coordination during locomotion remains unclear.

The purpose of the current study was to quantify how different RT modalities affect kinematic synergies related to the mediolateral trajectory of the swing-foot during normal and perturbed gait (walking across an uneven surface with and without additional asymmetric loading that promote additional imbalance). In addition to evaluating the effect of I-FRT and standard S-MRT on kinematic synergies, we also evaluated a highly specific adductor/abductor resistance training (S-MRTHIP) to better understand the extent to which hip strength affects kinematic synergies related to foot placement. We hypothesized first that kinematic synergies (i.e., synergy index) would stay invariant across groups from pre- to post testing during normal walking, given that normal walking is a habitual task. Second, we hypothesized that only the I-FRT group would increase the kinematic synergy index during perturbed gait (in absence of prior literature, we assumed the null for S-MRTHIP). Third, we hypothesized that in I-FRT the increase of the kinematic synergy index would result from an increase in "good" variance and a decrease in "bad" variance due to improved co-variation of lower-extremities based a previous study (Wu and Latash, 2014).

# METHODS

# Study Design

We conducted a registered three-arm, double-blinded RCT (ClinicalTrials.gov: NCT03017365 on 01/04/2017) examining the effects of three (RT) protocols on kinematic synergies and strength, power, and balance in older adults. The assessors were blinded to the participants' assignments. Participants were naïve to the study hypothesis. The trial was approved by the local ethics committee of the University of Kassel (E052016058) and was complied with the relevant ethical standards of the latest Declaration of Helsinki (WMA, October 2013). All participants provided written informed consent prior to enrollment.

# Participants

In total 82 participants between the age of 65 and 80 were recruited via public advertisement. The only inclusion criteria were the ability to walk independently without any gait aid. Participants were excluded based on pathological ratings of the Clock Drawing Test (CDT) (Nair et al., 2010), the Mini-Mental-State-Examination (MMSE, <24 points) (Lopez et al., 2005), the Falls Efficacy Scale – International (FES-I, >24 points) (Dias et al., 2006), the Geriatric Depression Scale (GDS, >9 points) (Parmelee and Katz, 1990), the Freiburg Questionnaire of Physical Activity (FQoPA, <1 h) (Frey and Berg, 2002) and the Frontal Assessment Battery (FAB-D, <13 points) (Benke et al., 2013). Ultimately, 68 participants successfully completed the trial. **Figure 1** shows the CONSORT flow diagram and the number of participants in the treatment arms at each stage of the trial. Subject' demographics and baseline descriptors of the who completed the 10-week trial are presented in **Table 1**.

# Randomization

Participants were stratified (1:1:1) into one of three groups according to age and sex. An uninvolved researcher then randomly assigned the groups to one of three training modalities: Machine-based stable resistance training (S-MRT), freeweight instability resistance training (I-FRT), or machinebased adductor/abductor resistance training (S-MRTHIP). The randomization sequence was generated using www.randomizer. org and was concealed until groups were stratified.

# Assessment

Data was collected in the biomechanics laboratory of the University of Kassel, Germany.

# Kinematic Data Collection, Processing and Analysis

Twenty-six 12.5 mm reflective markers were attached bilaterally to the legs with double-sided adhesive tape at prominent bony landmarks according the IOR lower-body marker-set (Leardini et al., 2007). A six-camera motion capture system (Oqus 3+, Qualisys AB, Gothenburg, Sweden) operating at 120 Hz was used to record marker trajectories. Participants then walked for 1 min back and forth through a capture volume of 5 m at a self-selected walking speed. The capture volume was preceded/proceeded by ∼2 m which allowed the participant to accelerate/decelerate before entering/exiting the capture volume. Participants completed three conditions: the even surface (ES) (control) condition where they walked across the lab floor; the uneven surface (US) condition where foam panels (terrasensa <sup>R</sup> classic; Huebner, Kassel, Germany; see **Figure 2**) were placed on the floor to create an uneven surface; and the "imbalanced shopping bag" (ISB) condition where they walked across the US carrying a simulated shopping bag – i.e., a tube, lanced with a chain with ends weights attached (5% of the body weight) – in the dominant hand (USISB). The last condition was intended to present an additional challenge to balance above the US condition alone. All data were processed using Visual3D (C-Motion, Germantown, MD, United States). Raw kinematic marker trajectories were interpolated and smoothed with a fourth-order zero-lag Butterworth low-pass filter with a cut-off frequency of 6 Hz. The UCM-analysis was then performed on the processed data using a custom written R-code (R Foundation for Statistical Computing, Vienna, Austria). The custom code is available upon request.

# Primary Measures: Uncontrolled Manifold Analysis

The UCM approach has been described in detail elsewhere (Scholz and Schöner, 1999; Latash et al., 2007) as has its application to stabilizing the swing limb trajectory during gait (Rosenblatt et al., 2015; Eckardt and Rosenblatt, 2018). Briefly, motion capture data was normalized to 0–100% corresponding to left toe off to heel strike. A geometric model was used to express the position of the swing limb at each percent of swing as a function of seven lower limb segment angles. For each step and at each percent of swing, the deviation between each angle and it's across-step average was calculated. A deviation vector was then projected onto a 6-DOF space that did not affect the swing limb position and a 1 DOF space that did, based on the Jacobian of the geometric model. The across-step average length of the projected vectors defined "good" and "bad" variance, respectively, at every percent of swing from which a synergy index was expressed. Consistent with prior studies (Krishnan et al., 2013; Rosenblatt et al., 2015; Eckardt and Rosenblatt, 2018), the variance components and synergy index were averaged across the swing phase for further analysis. The primary outcome from the analysis is the synergy index, which was z-transformed prior to statistical testing (1VZ). The index was calculated as the difference between the "good" variance per DOF (VUCM) and "bad" variance per DOF (VORT) relative to the total kinematic variance per all DOFs (VTOT) such that changes in 1V<sup>Z</sup> can reflect multiple strategies (Wu and Latash, 2014). Therefore, in addition to 1VZ, we also report VUCM, VORT, and VTOT. A more detailed description of the UCM method can be found in **Supplementary Material SII**.

### Secondary Measures

#### **Balance assessment**

We tested proactive balance using the timed-up-and-go test (TUG) (Podsiadlo and Richardson, 1991) and the multidirectional reach test (MDRT) (Newton, 2001). For the TUG, participants were asked to rise from a chair and walk three meters at their habitual walking speed, turn around a cone return to the chair and then sit down. Time was recorded to the

#### TABLE 1 | Subject characteristics and descriptive values.

fnagi-11-00032 February 25, 2019 Time: 18:30 # 5


S-MRT, stable machine based resistance training; I-FRT, instability free-weight resistance training; S-MRTHIP, stable machine based adductor/abductor resistance training; M, mean; SD, standard deviation; f, female; m, male; MMSE, Mini Mental State Examination; CDT, Clock Drawing Test; GDS, Geriatric Depression Scale; FAB\_D, Frontal Assessment Battery, German Version.

nearest 0.01 s using a stopwatch that started on the command "ready-set-go" and stopped as soon as the participants sat down. The MDRT measures the maximal distance participants could reach forward, backward, left, and right while standing without taking a step. Maximal reach distance (cm) was recorded. We added the left/right conditions and calculated the mean mediolateral distance for further analysis. Participants had one practice trial for every test. Two test trials were carried out and the mean was entered into statistical analysis.

#### **Strength and power assessment**

Maximal isometric leg extension strength was examined with the isometric mid-thigh pull test (IMTP) (McMaster et al., 2014). Data was measured with a force plate (Model 9281B, Kistler Instrument AG, Winterthur, Switzerland), operating at 1200 Hz and recorded with QTM (Qualisys AB, Gothenburg, Sweden). Participants stood upright in a squatting position on a solid, elevated metal platform, bridging over the force plate to avoid contact. Cable length was individualized to guarantee a constant knee angle of approximately 135◦ . Participants were then asked to pull upward on a handle connected to the force plate, starting initially with a moderate intensity and slowly increase the intensity to maximum exertion while keeping the upper body extended and upright. To ensure upright posture, an assessor put her hand on the participants' back while pulling. The IMTP shows high within- and between-session reliability (ICC ≥0.87) (Moeskops et al., 2018).

Bilateral isometric strength of hip adduction and abduction, as well as the knee extensor, was measured with a handheld dynamometer (Lafayette Instrument Company, Lafayette, IN, United States) (Arnold et al., 2010). To measure hip adduction and abduction, participants were positioned sideways on a therapy bench. The hand-held dynamometer was placed above the malleolus of the lower (adduction) or upper leg (abduction) respectively, as previously described (Arnold et al., 2010). Participants were asked to adduct and abduct their respective leg. For knee extension strength, participants sat on the therapy bench and were asked to try to extend their leg. The assessor placed the hand-held dynamometer at the lower leg just proximal to the ankle. We recorded two maximum effort isometric contractions for 3–5 s with each muscle group. Interclass correlation coefficients (ICCs) are generally high for hand-held dynamometry (ICCs 0.95–0.99) (Arnold et al., 2010). For all isometric strength testing, we provided one practice trial and then averaged the next two trials. Measures were taken on both limbs and then averaged across limbs prior to statistical analysis. To limit the effects of fatigue we allowed recovery periods (>1 min) between trials.

To assess lower extremity muscle power, we administered the Five Times Sit-to-Stand-Test (STS) (Tiedemann et al., 2008). Participants were instructed to stand up and sit down five times as quickly as possible, without using their arms. They were advised to fold their arms across the upper body. Time was measured by a stopwatch to the nearest 0.01 s. After the countdown "ready-setgo," testing time was started and stopped when participants sat down for the fifth time.

#### **Questionnaires**

Global cognitive function was assessed using the MMSE, a screening tool for mild cognitive impairment (Lopez et al., 2005). The FAB-D consists of six neuropsychological tasks, evaluating cognitive and behavioral frontal lobe functions (Dubois et al., 2000). Physical activity was assessed using the FQPA (Frey and Berg, 2002). Concern about falling was evaluated using FES-I (Dias et al., 2006). The FES-I was the only questionnaire applied pre- and post-testing. All other tests were used for screening purposes and/or to describe the population.

# Exercise Intervention

The exercise intervention took place between January and April. Training was supervised by two trained instructors providing a participant to instructor ratio of 5:1. All intervention groups trained for 10 weeks, twice per week on non-consecutive days for 60 min per day. We began with a 1-week introductory phase and three training blocks lasting 3 weeks each. Training intensity was progressively and individually increased by modulating load and sets for all groups and the level of instability for the I-FRT group (see **Table 2**). After week one, four, and seven the training

TABLE 2 | Detailed intervention program for all groups and phases.


S-MRT, stable machine-based resistance training; I-FRT, instability free-weight resistance training; S-MRTHIP, machine-based adductor/abductor training. bw, body weight; 1-RM, one repetition maximum; BOSU, BOth Sides Utilized; ROM, Range of Motion; ML, mediolateral.

load (weight) was increased following one repetition maximum (1-RM) testing using the prediction equation provided by Epley (Reynolds et al., 2006) for each major exercise. The 1-RM was performed under stable conditions for every group.

#### S-MRT

The main exercises of this group were squats at the Smith machine, placing the barbell at the hip instead of putting it on the participant's shoulders, and the leg-press. Secondary exercise were core exercises and walking with weights across an even surface.

#### I-FRT

This group also performed squats, but instead of using the Smith machine, they exercised using instability devices (i.e., foam pads and BoSU balls) and dumbbells. The second main exercise was the front lunge on instability devices. Secondary exercises were core routines, incorporating instability devices, and walking across an uneven surface (terrasensa <sup>R</sup> classic; Huebner, Kassel, Germany) carrying dumbbells.

### S-MRTHIP

The main exercises for this group were the thigh/hip adductorand abductor resistance machine. As secondary exercises, participants performed additional adduction and abduction exercises using elastic rubber straps. The resistance of the rubber straps was incrementally increased every block (changed by one color). Furthermore, lateral core exercises were introduced. In addition, this group walked across a special motorized treadmill (robowalk <sup>R</sup> , h/p/cosmos, Nußdorf, Germany), which applied a lateral pull via elastic straps at the ankle and/or knee while walking.

Detailed description of the training programs and changes in intensities and degrees of instabilities can be found in **Table 2**.

#### Training Intensity

Training intensity was quantified as the combined load (weight) for the two main exercises during the last training phase as determined from the participants' training sheets.

# Data Analysis

An a priori sample size calculations with G∗Power 3.1.9.2 showed that to detect an expected effect of Cohen's d = 0.3 (Shim et al., 2008; pilot data) at α = 0.05 with 1-β = 0.90 using a repeated measures with a within-between and interaction design, a total sample size of at least N = 15 per group was required. Normality of the data was checked by visual inspection and tested with the Kolmogorov–Smirnov test for each dependent variable per group, prior to the main analysis. Given that ANOVAs are quite robust against violations of distribution (Schmider et al., 2010), we would only employ non-parametrical alternatives in the event that a variable was non-normal for at least two groups. Baseline differences were tested between groups with a one-way ANOVA. Given that gait speed affects gait kinematics, we compared gait speed between pre- and post-testing for all three conditions with dependent two-sided t-tests. The effect of treatment was analyzed separately for each of the primary and secondary outcomes using 2 (time: pre-test, post-test) × 3 (group: S-MRT, I-FRT, S-MRTHIP) ANOVAs with repeated measures on time and between subject factor being group. In the case of a significant interaction (p ≤ 0.05), post hoc tests (dependent twosided t-tests) were used to detect significant pre-post differences within each group. In addition, we investigated differences in the training load between groups using pre-planned independent two-sided t-tests. Ryan–Holm–Bonferroni corrected p-values for all t-tests are reported. Further, we employed Bayesian t-tests and calculated Bayes Factors (BF) to extend explanatory power of the inference t-tests results. We assume a default Cauchy prior width of 0.707. **Table 3** summarizes the common interpretation of BF (Wetzels et al., 2011). To provide additional information, we also calculated the effect size as Cohen's d for ANOVAs. Exploratory

TABLE 3 | Evidence categories for Bayes factor.


Adapted from Jeffreys (1961), cited in Wetzels et al. (2011). H0, Null-Hypothesis; HA, Alternative Hypothesis.

Software for Confidence Intervals was used to calculate Cohen's dunb (an unbiased estimate of the population effect size δ), associated 95% confidence intervals and the t-tests (Cumming, 2012). Following Cohen (Cohen, 1988), d-values ≤ 0.49 indicate small effects, 0.50 ≤ d ≤ 0.79 indicate medium effects, and d ≥ 0.80 indicate large effects. Alpha level was set at 5%. Bayesian t-tests were computed using JASP (Version 0.9.0.1). For all other tests we used IBM SPSS version 23.

# RESULTS

The individual results are deposited as complete dataset in the **Supplementary Material SI**. The machine-based stable resistance training (S-MRT) group had an average attendance of 94%, 95% for the free-weight instability resistance training (I-FRT) group, and 95 % for the machine-based stable adductor/abductor resistance training (S-MRTHIP) group. Gait speed increased from pre- to post-testing for conditions (**Figure 3**): ES [t(66) = 3.74, p < 0.001, dunb = 0.45; 95%-CI (0.21, 0.71); BF<sup>10</sup> = 61.67], US [t(66) = 3.89, p = 0.001, dunb = 0.41; 95%-CI (0.16, 0.67); BF<sup>10</sup> = 21.84], and USISB [t(66) = 2.31, p = 0.024, dunb = 0.29; 95%-CI (0.04, 0.53); BF<sup>10</sup> = 1.61]. Based on our a priori criteria for non-parametric testing, we were able to use parametric test for all variables. All outcomes are summarized in **Figures 4**, **5**.

# Primary Measures: Uncontrolled Manifold

### Kinematic Synergy Index

Regardless of group, 1V<sup>Z</sup> during the ES condition was not affected by RT (S-MRT: −4% change; I-FRT: 2% change; S-MRTHIP: 2% change). However, there was a time x group interaction for both challenging walking conditions (**Table 4**). In particular, 1V<sup>Z</sup> for I-FRT increased by 16% during US and 20% during USISB whereas there was no significant change in the kinematic synergy during either condition for both S-MRT and S-MRTHIP (**Table 5**).

# VUCM

Consistent with the fact that 1V<sup>Z</sup> during ES was not affected by any form of RT, we found no significant effect of time on "good" variance during the ES condition. In contrast, a main effect of time was observed for both unstable walking conditions. During the US condition we observed increases of 21% (S-MRT), 28% (I-FRT), and 50% (S-MRTHIP) for VUCM, with similar effects across groups; i.e., no significant interaction was present. Similar results were seen for the USISB condition; VUCM for this condition increased by 19% following S-MRT, by 28% following I-FRT and by 43% following S-MRTHIP. There was no time x group interaction (**Table 4**).

### VORT

Regardless of group there was no significant effect of time on VORT during the ES condition. However, there was a significant time x group interaction for both of the challenging conditions (**Table 4**). "Bad" variance decreased by 25% in the US condition following I-FRT and decreased by 24% in the USISB condition. In

contrast to the I-FRT group, the other two groups significantly increased VORT by more than 35% during the US condition following training. Similar increases were found for these groups during the USISB condition; S-MRT increased "bad" variance by 35% and S-MRTHIP by 41% (**Table 5**).

Median; dotted line, upper/lower quartile. The width of the plots is scaled to data distribution.

### VTOT

Like VUCM and VORT, the total variance while walking across the even surface did not significantly change as a result of RT. However, there was a significant main effect of time on total variance during the two challenging conditions, with an average increase of 36% regardless of group; there was no significant time x group interaction for either challenging condition (**Table 4**).

# Secondary Measures

#### Balance Assessment

All groups reduced their TUG times by an average of 2%. However, there was no significant effect of time on TUG (**Table 4**).

The effects of RT on proactive balance, measured by the MDRT were consistent across groups; regardless of group, participants improved their forward reaching skills by an average of 8% whereas backward leaning was not significantly improved with RT. There was a significant time x group interaction for mediolateral proactive balance; with the S-MRT and I-FRT groups increasing side reaching by 4 and 14%, respectively, while S-MRTHIP decreased their ability by 12% (**Table 5**). The post hoc tests revealed that the change for S-MRT and S-MRTHIP was not significant whereas the effects were significant for I-FRT (**Table 5**).

### Strength and Power Assessment

Regardless of group, lower extremity muscle power, measured using the Five Times Sit-to-Stand task increased by 10% in all groups. There was no significant time x group interaction (**Table 4**).

On average there was a 19% increase in isometric leg extension strength, but the effects varied by RT group; there was a significant time x group interaction. The post hoc tests revealed a significant improvement for S-MRT and I-FRT, and no significant effect observed for S-MRTHIP (**Table 5**).

There was no effect of time on hip adduction and abduction strength. However, isometric knee extension strength did increase by 14% with time. Nonetheless, changes across groups were similar for all strength variables, thus we found no interaction effect time x group (**Table 4**).

### Training Intensity

We found meaningful differences between groups, demonstrating that I-FRT exercised with considerably lower loads than the other groups. On average, I-FRT exercised on both main exercises with ∼150 kg less than S-MRT and with ∼56 kg less than S-MRTHIP. See **Figure 6**.

#### Questionnaire

Fear of falling, measured with the FES-I, was significantly reduced over time by 3–7% but the effects were similar across groups. M-SRT reduced the FES-I score from 19.3 ± 2.6 to 18.0 ± 2.3, I-FRT from 20.0 ± 3.7 to 18.4 ± 2.8, and S-MRTHIP changed the score from 18.9 ± 3.2 to 18.3 ± 2.4.

# DISCUSSION

The purpose of this study was to quantify the effect of different RT modalities on kinematic synergies (derived using UCM analysis) related to the ML trajectory of the swing-foot during normal and perturbed gait (walking across an uneven surface with and without additional weight/imbalance). Our first hypothesis, that the kinematic synergy for the ES condition would remain invariant for all groups after the intervention was supported. Our second hypothesis that the kinematic synergy during both unstable conditions would increase after training for the F-IRT

#### TABLE 4 | ANOVA outcomes.

fnagi-11-00032 February 25, 2019 Time: 18:30 # 11


dfmain\_effect: 1; dfinteraction\_effect: 2, dferror: 65; 2; S-MRT, stable machine-based resistance training; I-FRT, instability free-weight resistance training; S-MRTHIP, stable machine-based adductor/abductor resistance training; UCM, uncontrolled manifold approach; 1VZ, synergy index; VUCM, "good" variability; VORT, "bad" variability; VTOT, total variability; ES, even surface; US, uneven surface; USISB, uneven surface with imbalanced shopping bag; TUG, timed up and go test; MDRT, multi directional reach test; ML, mediolateral; STS, sit-to-stand test; IMTP, isometric mid-thigh pull test; FES-I, falls efficacy scale. Bold values represent significant effects of the ANOVAs.

group only was also supported. We found decisive evidence for an increased magnitude of the synergy index in the US and in the USISB condition. Consistent with our third hypothesis these stronger synergies were due to reduced "bad" variance for I-FRT while walking across the more challenging conditions; i.e., US and USISB. All groups were able to rely on motor flexibility (proportional increase of "good" variability in relation to "bad" variability) to maintain a kinematic synergy at posttesting and across conditions. To the best of our knowledge, this is the first study investigating the use of the UCM approach to quantify (instability) resistance training induced changes within multi-segmental lower-extremity kinematic synergies to stabilize an important performance variable (i.e., ML swing-foot trajectory).

TABLE 5 | T-test pre-post comparisons.


S-MRT, stable machine-based resistance training; I-FRT, instability freeweight resistance training; S-MRTHIP, stable machine-based adductor/abductor resistance training; M, mean; SD, standard deviation; UCM, uncontrolled manifold approach; 1VZ, synergy index; VORT, "bad" variability; US, uneven surface; USISB, uneven surface with imbalanced shopping bag; TUG, timed up and go test; MDRT, multi directional reach test; ML, mediolateral; IMTP, isometric mid-thigh pull test. Bold values represent signifcant and meaningful effects of the t-tests.

Previous research using the UCM analysis to investigate practice effects on coordination suggested a two-stage process of adaption in multi-segmental coordination (Wu and Latash, 2014). In the first stage "bad" variability drops due to optimized performance control, while "good" variability hardly changes. The second stage is characterized by a reduction in "good" variability, while "bad" variability remains constant. This can be explained by practice-induced optimized control over elemental features, other than the explicit performance variable (Wu and Latash, 2014). However, there are scenarios where a practiceinduced increase of VUCM can be found. An increase of VUCM suggests a more robust and flexible system which can exploit an abundance of motor solutions, especially when being challenged (Wu and Latash, 2014; Latash and Huang, 2015). Our results certainly support the first stage, given the drop of VORT during both US conditions and the accompanying increase in the kinematic synergy. The fact that we observed an increase rather than a drop in VUCM during the US conditions may be explained by increased gait speed. Given that gait speed increased following the intervention and that variability increases with speed, particularly at faster-than-preferred speed (Chien et al., 2015), we expected an increase particularly in performance destabilizing "bad" variability (Rosenblatt et al., 2014, 2015; Chien

et al., 2015). With regard to the specific variance components, increases in movement speed have previously been associated with increased VORT (Chien et al., 2015; Rosenblatt et al., 2015) which is observed in the two machine-based groups. To counter the increased VORT these groups rely on motor flexibility and increase VUCM as well, which is consistent with our previous cross-sectional study in which older adults compensated for challenged stability during walking by increasing VUCM (Eckardt and Rosenblatt, 2018). The fact that I-FRT decreased VORT while walking across both US conditions despite increased gait speed is therefore noteworthy and consistent with the second stage of motor learning (Wu and Latash, 2014). The concurrent increase in VUCM for this group highlights the fact that I-FRT specifically improved coordination, not to compensate for an increase in VORT but as means to develop coordination among elements to allow flexible performance and avoid reliance on a unique solution (Wu and Latash, 2014; Latash and Huang, 2015).

Several prior studies support the idea that resistance training incorporating modalities that promote exploration of movement space may positively affect motor flexibility. For example, Hamed et al. (2018) showed that while both I-FRT (or exercise dynamic stability under unstable conditions as they call it) and S-MRT improved muscle strength and balance recovery in simulated forward falls, only I-FRT increased standing balance abilities. The authors attributed the effects of I-FRT on the fact that the RT is performed under continuously destabilizing conditions that require continuous processing and integration of sensory afferent information to attain an appropriate motor response (Hamed et al., 2018), which is critical to control of dynamic postural stability under the unstable conditions. Exercising under unpredictable/instable conditions requires motor output to remain flexible enough to produce appropriate responses to continuously changing input. Indeed, recent investigations on adaptive mechanisms during uneven surface walking and running shows a widening of neuromuscular synergies (EMG) to provide robust and flexible motor solutions to compensate for perturbations (Santuz et al., 2018). Interventions forcing the CNS to explore an abundant number of motor solutions can elicit robust and more flexible neuromuscular synergies (EMG) (Oliveira et al., 2017).

In addition to increasing kinematic synergies, we found that I-FRT resulted in an increase in leg extension strength, which is in contrast to a previous RCT by our group (Eckardt, 2016). Indeed, the improvements for I-FRT in this study are larger than the prior RCT (Cohen's d: 0.50 vs. 0.65). A different protocol to estimate training load may explain this difference. In our prior study, the 1-RM was calculated on the respective instability device whereas the current study calculated the training load on even surface. However, it may be difficult to elicit a true 1-RM on instability devices. In fact, the 1-RM for I-FRT was slightly higher in this study compared to the prior one (8 ± 4 kg) which may elicit the present effect and the subsequent interaction effect time x group.

It was somewhat unexpected that S-MRTHIP did not improve adductor/abductor strength, which would be predicted based on principles of specificity (Buckthorpe et al., 2015; Contreras et al., 2016). We assume that the hand-held dynamometry test hip strength did not isolate hip adductors/abductors such that the other groups (i.e., S-MRT and I-FRT) could have attained similar testing values by compensating for weaker adductor/abductors with activation of other muscles to coordinate hip movement. Regardless of the reason, the fact that hip adductor/abductor strength did not improve with S-MRTHIP, or any of the other modalities, but that changes in the kinematic synergy were seen in I-FRT suggests that the neuromuscular strategies contributing to mediolateral foot placement and in turn synergistic behavior of lower-extremities may depend more on sensorimotor integration during dynamic (i.e., instable) situations than on force producing capabilities of the hip. Indeed both, S-MRT and S-MRTHIP are quite stationary modalities, at least with regarding to the resistance components which were the primary training components; during resistive exercise execution there is little or no unpredicted movement and minimal counter-rotation of segments relative to the CoM (Hamed et al., 2018). Future work should consider individual effects – e.g., are kinematic synergies different between responders (those who increase adductor/adductor strength) and non-responders or do kinematic synergies scale with strength – to better understand the relationship between hip strength and kinematic synergies during gait.

# LIMITATIONS

Because our participants were healthy older adults (mean TUG time at pre-testing: 8.1 s), generalizability to frail older people needs to be established. In addition, the extent to which I-FRT can strengthen synergies (kinematic or otherwise) related to other performance variables has yet to be determined.

Given that the I-FRT group practiced walking across the US, it is not possible to entirely disentangle the impact of repeated US walking from instability RT on kinematic synergies.

However, given that S-MRT practiced walking with weights and S-MRTHIP practiced resisted walking yet demonstrated an increase rather than a reduction in VORT, it appears that RT training alone, in absence of instability, cannot explain the changes observed in the I-FRT group. In addition, the fact that we observed similar strength changes but differences in motor flexibility between I-FRT and S-MRT suggests that the context in which strength changes occur is critical to promoting motor flexibility.

The high inter-subject variability (see **Figures 4**, **5**) may indicate that there are individual strategies to stabilize the kinematic synergy. Future research should try to identify such subject specific strategies.

# CONCLUSION

The purpose of this RCT was to quantify how different resistance training modalities affect kinematic synergies related to the mediolateral trajectory of the swing-foot during normal and perturbed gait. For all groups the kinematic synergy during normal gait (ES condition) was unaffected by RT. However, I-FRT demonstrated significant increases in the kinematic synergy on the uneven surface which was achieved by reducing motor noise (VORT) and therefore stabilizing the ML trajectory of the swing foot. To our knowledge, this is the first time the UCM approach was used to quantify resistance training induced effects on locomotor stability.

# DATA AVAILABILITY

All datasets analyzed for this study are included in the **Supplementary Material SI**.

# REFERENCES


# AUTHOR CONTRIBUTIONS

NE analyzed the data, wrote the manuscript, and contributed to the conception and design of the study. NR contributed to the study design, assisted with the data analysis and interpretation, and critical review of the manuscripts. Both authors read and approved the final version of the manuscript.

# FUNDING

The trial was supported by a grant of the Regional Government of Hessen (Germany) and task (Center for Transfer and Application in Sports Kassel).

# ACKNOWLEDGMENTS

We thank Meagen L., Kai V., Ingo K., and Elisabeth H. for the data acquisition and Jonas L., Yanneck N., Simon D., and Heiko W. for carrying out the training programs. Further, we would like to thank sensa <sup>R</sup> by Huebner Group for providing terrasensa plates and h/p/cosmos for providing the treadmill. In addition, we thank Daniel W., and Daniel K. for their help with the R-code.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnagi. 2019.00032/full#supplementary-material

MATERIAL SI | Dataset.

MATERIAL SII | Detailed UCM approach.

Individuals. Available at: http://www.panr.com.cy/index.php/article/the-effectof-walking-speed-on-gait-variability-in-healthy-young-middle-aged-andelderly-individuals/



**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 Eckardt and Rosenblatt. 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.

# Multi-Sensorimotor Training Improves Proprioception and Balance in Subacute Stroke Patients: A Randomized Controlled Pilot Trial

Chaegil Lim\*

Department of Physical Therapy, College of Health Science, Gachon University, Incheon, South Korea

Introduction: The objective was to determine whether advanced rehabilitation therapy combined with conventional rehabilitation therapy consisting of sensorimotor exercises would be superior to usual treadmill training for proprioception variation and balance ability in subacute stroke patients.

Methods: Thirty subjects (post-stroke time period: 3.96 ± 1.19 months) were randomly assigned to either a multi-sensorimotor training group (n = 19) or a treadmill training group (n = 18). Both groups first performed conventional physical therapy for 30 min, after which the multi-sensorimotor training group performed multi-sensorimotor training for 30 min, and the treadmill training group performed treadmill gait training for 30 min. Both groups performed the therapeutic interventions 5 days per week for 8 weeks. The primary outcome (proprioception variation) was evaluated using an acryl panel and electrogoniometer. The secondary outcome (balance ability) was measured using the Biodex Balance system before intervention and after 8 weeks.

#### Edited by:

France Mourey, Université de Bourgogne, France

# Reviewed by:

Marialuisa Gandolfi, University of Verona, Italy Alessandro Picelli, University of Verona, Italy

> \*Correspondence: Chaegil Lim jgyim@gachon.ac.kr

#### Specialty section:

This article was submitted to Neurorehabilitation, a section of the journal Frontiers in Neurology

Received: 06 December 2018 Accepted: 07 February 2019 Published: 01 March 2019

#### Citation:

Lim C (2019) Multi-Sensorimotor Training Improves Proprioception and Balance in Subacute Stroke Patients: A Randomized Controlled Pilot Trial. Front. Neurol. 10:157. doi: 10.3389/fneur.2019.00157 Results: The multi-sensorimotor training and treadmill training groups showed significant improvement in proprioception variation and balance (overall, A-P and M-L) (all P < 0.05). In particular, the multi-sensorimotor training group showed more significant differences in proprioception variation (P = 0.002) and anterior-posterior (A-P) balance ability (P = 0.033) than the treadmill training group.

Conclusions: The multi-sensorimotor training program performed on multiple types of sensory input had a beneficial effect on proprioception sense in the paretic lower limb and A-P balance. A large-scale randomized controlled study is needed to prove the effect of this training.

Clinical Trial Registration: https://cris.nih.go.kr/cris/, identifier KCT0003097.

Keywords: sensorimotor training, proprioception, balance, stroke, hemiplegia

# INTRODUCTION

Impaired sensory and functional balance abilities after strokes often make it difficult for patients to return to their activities of daily living (ADL), thus creating a potential burden to family members and society. Approximately 50% of stroke patients experience sensory impairment. Occasionally, these neurological disorders are accompanied by aphasia, hemianopsia, or neglect. For over 50% of patients, sensory defects present on the affected side (1).

Sixty-five percent of stroke patients experience impaired tactile and protective responses, including proprioceptive sensations (1). As a result, stroke patients are less able to transmit information to the brain and spinal cord regarding muscle strength, pressure, joint position, and muscle length, which are required to maintain posture and can be detected in various joints on the paralyzed side (2).

Postural control involves biomechanical constraints, cognitive processing, control of dynamic, orientation in space, movement strategies, and sensory strategies. Sensory information from somatosensory, vestibular, and visual systems is then integrated, and the relative weights placed on each of these inputs are dependent on the goals of the movement task and the environmental context (3). Stroke patients typically have decreased balance reaction times, postural sway strategies, and impaired body weight support of the hemiparetic limb (4). Balance impairment ranks first among stroke disorders. Decreased muscle power, coordination, and sensory make it difficult to maintain balance (5). Furthermore, decreased balance not only potentially increases the risk of falls and femoral neck fracture but also decreases the ability to perform physical activity (6).

In general, stroke patients regain physical ability from sensory stimulation and mass motor exercise or task-oriented practice facilitating neural plasticity (7). Rehabilitation therapy programs can be classified as either conventional or advanced programs according to the theoretical background and clinical trials (8). Conventional rehabilitation therapy programs include the Bobath concept, Brunnstrom approach, proprioceptive neuromuscular facilitation (PNF), and functional strengthening approaches that emphasize motor learning and control, functional activity, or muscle strengthening (9). Regular and repetitive therapies involve sensory input with visual, verbal, tactile, cutaneous, proprioceptive, and auditory assistance for clinical stroke rehabilitation (8). Advanced rehabilitation therapy programs include electrical stimulation (10), robotic therapy (11), and virtual reality (12) for proprioceptive, tactile, visual, and auditory assistance in specific interventions based on neuroscientific evidence (9). Previous studies have recommended trunk control by training on a vibration board (13) and reactive balance training through perturbation in spastic diplegia cerebral palsy (14). According to Aman et al. (15), using a muscle spindle to stimulate active and passive sensorimotor training affects postural control and balance.

Depending on the stroke patient's ability and recovery stage, appropriate advanced rehabilitation therapy combined with conventional rehabilitation therapy consisting of sensorimotor exercises can provide multiple types of sensory input to assist in recovery after stroke (8). As Smania et al. described (16), with a specific training program based on weight transfer and balance exercise performed under different conditions of manipulation of sensory inputs, chronic stroke patients achieved significant improvement in their ability to maintain balance control. Sensorimotor training progressively improves the ability to re-weight and integrate sensory inputs to control balance, even in conditions where somatosensory input has been altered, and to avoid falls (17). However, to our knowledge, previous studies that have targeted multi-sensory input training for stroke patients are very limited.

It was hypothesized that advanced rehabilitation therapy combined with conventional rehabilitation therapy consisting of sensorimotor exercises would be superior to the usual treadmill training for proprioception variation and balance ability in subacute stroke patients.

# METHODOLOGY

# Setting, Study Design, and Participants

This study was a two-arm, parallel, and randomized controlled pilot trial with concealed allocation and participants as well as blinding for the researcher and assistants. All procedures from this study were approved by the Institutional Review Board of Gachon University (IRB No.: 1044396-201803-HR-068-01) and registered at Clinical Research Information Service (CRiS), Republic of Korea (KCT0003097) and all participants signed informed consent prior to beginning the study. In addition, this study conforms to all CONSORT guidelines.

# Procedures

Stroke patients were recruited from a rehabilitation hospital in Incheon, South Korea. Participants (inpatient or outpatient) 50– 71 years of age who experienced their first stroke were enrolled in this study if it had been 6 months or less since the unilateral hemisphere stroke, they could walk for 30 s or more (regardless of using assistance), and they completed the mini-mental state examination (MMSE) with a score of 24 or more. All patients had experienced stroke as defined by computed tomography (CT) or magnetic resonance imaging (MRI). Exclusion criteria consisted of the presence of a cognitive disorder, visual disorder and severe unilateral neglect, cardiorespiratory disorder (with cardiac pacemaker), concurrent neurological disease (e.g., Parkinson disease), orthopedic intervention, having over G2 on the Modified Ashworth Scale and receiving botulinum toxin injections for spasticity within the past 6 months.

# Randomization and Masking

Baseline measurements of patients' abilities were performed prior to randomization. Subsequently, each participant was allocated to one of the two groups via allocation codes included in consecutively numbered, sealed, opaque envelopes. Simple randomization was conducted using Microsoft Excel for Windows software (Microsoft Corporation, Redmond, WA, USA) by a researcher who was not involved in participant recruitment. To ensure masking, protocols and intervention order were not revealed to clinical evaluators. Intervention allocation was recorded in a password-protected document to maintain blinding. All data were measured by the same blinded physical therapist before the intervention and at the end of the 8 weeks intervention period.

# Interventions Procedures

All groups performed conventional rehabilitation therapy for 30 min. Then, the multi-sensorimotor training group performed Stabilize-T and Reha-Bar (Pedalo <sup>R</sup> ; Holz-Hoerz GmbH,

Münsingen, Germany) exercises (18) with transcutaneous electrical nerve stimulation (TENS) for 30 min. The treadmill training group participated in treadmill gait training with placebo TENS (with only adhesive TENS electrodes) for 30 min (19). We applied one of these methods to the patients in accordance with their ability. In all interventions and assessments, if patients complained of discomfort, they were told to stop immediately and take a rest. Both groups received an intervention for 5 days per week for 8 consecutive weeks.

## Conventional Rehabilitation Therapy

Conventional rehabilitation therapy, such as the Bobath approach or the PNF approach, was performed for 30 min. The patients were in the supine position, the trunk and upper and lower parts of the back were aligned, stability of the trunk was ensured, and limb movements of the hip joint, knee joint, and ankle joint were induced. Subsequently, the subjects maintained stability in the trunk and repeated flexion and extension of the lower limbs. A skilled physical therapist provided assistance so that alignment was not altered. The backward tilt of the pelvis involved the use of the pelvis and back muscles. The sitting exercises were performed for smooth pelvic movement and co-contraction of the hamstrings and quadriceps muscles by adjusting the height using a Bobath table to control muscle strength. We performed weight support training on the left and the right in the standing position and one-leg position.

# Multi-Sensorimotor Training Program

The multi-sensorimotor training program included Stabilize-T and Reha-Bar (Pedalo <sup>R</sup> ) (**Figure 1**) exercises (18) with TENS (19). The electrical stimulation therapy was delivered by a dualchannel TENS unit during the bipolar balanced phase, with a pulse duration of 200 µs and frequency of 100 Hz (TENS 7000TM; Koalaty Products Inc, Florida, USA). Adhesive TENS electrodes (5 × 5 cm) were fastened to the paralyzed medial and lateral motor points of the gastrocnemius muscles. Under stimulation conditions, the TENS intensity was adjusted before the start of measurements in increments of 0.01 mA and set at the subsensory threshold of each patient (19). Stabilize-T exercises improved sensory input (proprioceptive and tactile) to help control the posture balance of the locomotor activity (18).

The exercises were performed as follows. First, patients looked ahead while standing with the knee slightly bent on the Stabilize-T (start posture). They closed their eyes for 15 s, opened their eyes for 10 s, and maintained balance for 30 s. After patients stood on the Reha-Bar, they pedaled up and down and rotated the wheels to move forward while holding the safety bar with both hands with the therapist's assistance. Second, patients posed in the same position as the first posture. Then, patients slowly abducted the arms and rotated the neck to the left, center, and right for 30 s each and maintained balance for 30 s. Then, after patients stood on the Reha-Bar again, they pedaled up and down and rotated the wheels to move forward while holding the safety bar with both hands without the therapist's assistance. Third, patients posed in the same position as the first posture. Patients abducted the arms and looked upward for 10 s. Patients repeated the neck extension exercise three times and maintained balance for 30 s. After patients stood on the Reha-bar, they pedaled up and down and rotated the wheels to move forward and backward while holding the safety bar with both hands without the therapist's

assistance. The first exercise was the easiest and the third exercise was the most difficult. These exercises indicated the abilities of the patients in this study (18).

### Treadmill Gait Training Program

The treadmill gait training program involved gait with placebo TENS for 30 min at the paralyzed medial and lateral motor points of the gastrocnemius muscle. Patients participated 5 days per week for 8 weeks (20). Gait training started at low intensity for 10 min (50% heart rate reserve [HRR]). The exercise duration increased 5 min every 2 weeks, and the exercise intensity increased 5% HRR every 2 weeks. The final goal was 30 min at 70% HRR (21).

# Outcome Measurements

Baseline general characteristics were collected through file audit and self-reporting. The primary outcomes were the changes in the proprioception sense. The secondary outcomes were the changes in the balance ability. Primary and secondary outcome measures were collected in the hospital after randomization and after 8 weeks in the same place. Researchers masked to treatment allocation collected and entered all data. The physical therapist recorded their intervention recommendations. A research assistant conducted a blinded content analysis of the recommendations to provide relevant descriptive categories for analysis.

Proprioception variation was assessed using an acryl panel (60 × 60 × 1 cm) and electrogoniometer (JTECH Medical DUALER IQ PRO; Salt Lake City, UT, USA). In the sitting position, patients were asked to close their eyes and aligned their lower limbs on both sides of a clear acrylic panel. Then, the electrogoniometer was attached to the quadriceps and anterior tibia (**Figure 2**) (23). Each trial was performed for 10 s, and resting was required between trials to avoid fatigue and learning effects. The angle of the affected limb was measured after patients memorized the position of the unaffected limb indicated by the therapist. An average of five tests were recorded after two practice sessions (22).

Balance abilities were measured by the Biodex Balance system (BBS; Biodex Medical System, Inc., Shirley, NY, USA). This device focuses on the proprioceptive neuromuscular functions that appear to affect dynamic joint and postural stability. During postural balance testing, the patient's ability to control the platform's tilt angle was evaluated as a deviation from the center. The BBS software (Biodex version 1.08, Biodex, Inc.) presented the degree of deviation in each axis and provided an average sway score. The score levels ranged from 1 (low stability) to 8 (high stability) (24). During three trials, each test was performed for 20 s, followed by a 10 min resting period (25).

# Sample Size Estimation

We estimated that the minimal acceptable sample size would be 21 patients per group to achieve a power of 0.8 with a significance level (α) of 0.05 using a one-sided two sample t-test by G∗Power 3.1.9.1 software for Windows (Uiversität Kiel, Germany). It was decided that 29 patients would be necessary based on an intergroups difference in proprioception improvement in a previous trial (26).

# Statistical Analysis

SPSS 23.0 software for Windows 7 (IBM Corp., Armonk, NY, USA) was used to analyze the data. Data were summarized using means and standard deviation (SD). The normality of the parameter distributions were assessed using the Shapiro-Wilk test. If the data show a normal distribution, data were expressed as the mean ± standard deviation (continuous data) or percentage (categorical data), and parametric tests such as independent samples t-test or the χ 2 test were used to compare the general characteristics of the two groups. For a within-group comparison, a paired t-test was used and comparison between the two independent groups (multi-sensorimotor group and treadmill group) was accomplished with an independent t-test. The level of significance was set at α = 0.05.

# RESULTS

Between August 2017 and April 2018, a total of 49 stroke patients were admitted to the hospital and 37 fulfilled the inclusion criteria. Participants were randomly assigned to the multi-sensorimotor training group (n = 19) or the treadmill gait training group (n = 18). Of 37 participants who began the study, 30 (81%) completed it (**Figure 3**). A total of 7 patients (19%) were lost to follow-up or discontinued intervention. The general baseline characteristics of the participants of the two groups are described in **Table 1**. Recorded characteristics included gender, age, height, weight, lesion side, lesion type, and post-stroke duration. The mean ± SD age of the participants was 60.80 ± 7.03 years, and the mean ± SD post-stroke duration was 3.96 ± 1.19 months. Baseline demographic characteristics such as gender (males/females, 11/8 vs. 10/8), age (62.00 ± 7.30 vs. 59.61 ± 6.77 years), lesion type (ischemic/hemorrhagic, 13/6 vs. 11/7), lesion side (right/left;7/12 vs. 7/11), and post stroke-duration (4.05 ± 1.12 vs. 3.88 ± 1.27 months) were not significantly different between the multi-sensorimotor training group and the treadmill training group (P > 0.05).

The multi-sensorimotor training groups had improved proprioception after rehabilitative training compared to the treadmill training group (P < 0.001; effect size = 0.55; TABLE 1 | General characteristics of the two groups by randomization assignment.


Data are expressed as mean ± SD or n (%).

<sup>a</sup>The P-value was obtained using a χ 2.

<sup>b</sup>The P-value was obtained using an independent t-tests.


TABLE 2 | Changes in balance and proprioception within each group and between the two groups.

Data are presented as mean ± SD.

<sup>a</sup>P < 0.05. The P-value was obtained using a paired t-test.

<sup>b</sup>P < 0.05. The P-value was obtained using an independent t-test.

power 71%). Additionally, the A-P balance ability score of the multi-sensorimotor training group improved more than that of the treadmill gait training group (P = 0.03; effect size = 0.39; power 65%). Both groups had significantly improved balance ability scores (overall, anteriorposterior [A-P], and medial-lateral [M-P]) (P < 0.05) after intervention (**Table 2**).

# DISCUSSION

This study found that proprioception and A-P balance ability significantly improved in those in the multi-sensorimotor training program compared to those in the treadmill gait training group. This method using vibration, tactile, proprioception, and vestibular senses was effective for improving balance, especially in the multi-sensorimotor training group.

In the previous study, Moreside et al. (27) measured the activities of multiple trunk muscles by using electromyography, while the subjects performed horizontal-vibration exercises and showed that the activities of the internal oblique abdominal muscle and external oblique abdominal muscle were the highest. In contrast to this study, the multi-sensorimotor training involved Stabilize-T exercises from the all-direction vibration because the improvements were activated in the internal and external oblique abdominal muscles, the erector spinae muscle, latissimus dorsi muscle, and rectus abdominis muscle. The activation of these trunk muscles suggested that all-direction vibration stimulates improved balance ability (28). Additionally, previous studies reported that using vibration with the eyes closed improved the balance ability of healthy elderly participants because their balance ability had decreased more than that of healthy non-elderly subjects (4).

Neurophysiological observations suggest that changes in the process of sensorimotor integration do not occur at the peripheral level but depend on abnormal central processing of sensory input (29). These study results showed that proprioceptive sensory changes improved by 43% and 23% through multi-sensorimotor training and treadmill gait training, respectively. The multi-sensorimotor program, which comprised the neurological summation of the proprioceptive stimulus, was more effective than the other program. Proprioception involved electrical stimulation of the paralyzed calf muscles for 30 min during the intervention period; tactile, vestibular, and kinesthetic sensations were stimulated through the Stabilizer-T and Reha-Bar exercises. In addition, the muscle response needed to control the postural sway improved. Therefore, it was more effective for improving the A-P balance ability. In the multi-sensorimotor training group, the A-P balance ability was improved more effectively than the M-L balance ability. It can be inferred that the vestibular coordination exercises using the pedal tool effectively enhanced the peri-articular sensations of the surrounding soft tissues and muscles to control A-P balance.

The most direct cause of the balance recovery through TENS is an increase in somatosensory information from the lower limbs because the sensory stimulation through TENS increases the flow of somatic sensation rising from the lower limb that maintains and controls the standing posture (30). Additionally, Golaszewski et al. (31) reported that electrical stimulation increased signaling in the pre- and post-central gyri after cutaneous stimulation. The inferior parietal lobule was also activated in both hemispheres, and it is feasible that additional afferent stimulation might trigger the remaining plastic capacity for sensorimotor reorganization in the brain and might thus facilitate functional recovery in chronic stroke.

We also noted that there are significant improvements in overall and M-L balance after intervention in the multisensorimotor training group, but there were no intergroup differences. This may be because the improvements were observed in treadmill training groups by conventional therapy and treadmill training in the subacute phase.

This study has several limitations. First, 30 participants completed the study, which was insufficient to identify intergroup changes. Second, the long-term effects of the training could not be confirmed. In addition, we could not exclude the learning effect for each evaluation system. Third, we could not evaluate the postural control in ADL and the fear of falling. Finally, the level of the ankle muscle activity could not be directly proven. Therefore, future studies need to include more participants. In addition, a method that can directly quantify ankle and trunk muscle strength, such as electromyography (EMG) activity, should also be attempted, and the studies should be designed to explore whether the training effects are still present months after the experiment.

# CONCLUSION

This study provided evidence that combined rehabilitation methods significantly enhanced the proprioception and balance ability during the subacute phase of recovery after stroke. Therapists have an important role in the achievement of maximum benefits throughout the rehabilitation process after stroke. The optimal intensity and duration of specific interventions have been systematically evaluated, and

# REFERENCES


it has been indicated that combining valuable training exercises for multiple senses is believed to be a good method for facilitating the restoration of proprioception and balance ability.

# AUTHOR CONTRIBUTIONS

CL made substantial contributions to conception and design, acquisition of data (two research assistants helped), data analysis, interpreting the data, drafting the article, and revising it critically for important intellectual content and final approval of the version to be submitted.


before and after whole-hand afferent electrical stimulation. Scand J Rehabil Med. (1999) 31:165–73.

**Conflict of Interest Statement:** 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.

Copyright © 2019 Lim. 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.

# Fear of Falling Contributing to Cautious Gait Pattern in Women Exposed to a Fictional Disturbing Factor: A Non-randomized Clinical Trial

Guilherme Augusto Santos Bueno1,2 \*, Flávia Martins Gervásio<sup>2</sup> , Darlan Martins Ribeiro2,3 , Anabela Correia Martins <sup>4</sup> , Thiago Vilela Lemos <sup>2</sup> and Ruth Losada de Menezes <sup>1</sup>

<sup>1</sup> Postgraduate Program in Health Sciences and Technologies, University of Brasília, Brasília, Brazil, <sup>2</sup> Movement Laboratory Dr. Cláudio A. Borges, College of Sport, State University of Goiás, Goiânia, Brazil, <sup>3</sup> Dr. Henrique Santillo Rehabilitation and Readaptation Center, Goiânia, Brazil, <sup>4</sup> Department of Physiotherapy, ESTeSC - Coimbra Health School, Polytechnic Institute of Coimbra, Coimbra, Portugal

Objective: This study aimed to investigate the gait pattern of elderly women with and without fall-history, with high and low fear of falling, when exposed to a disturbing factor.

#### Edited by:

Helena Blumen, Albert Einstein College of Medicine, United States

#### Reviewed by:

Cristiano De Marchis, Università degli Studi Roma Tre, Italy Alessandro Picelli, University of Verona, Italy

#### \*Correspondence:

Guilherme Augusto Santos Bueno bueno.guilherme@aluno.unb.br

#### Specialty section:

This article was submitted to Neurorehabilitation, a section of the journal Frontiers in Neurology

Received: 30 November 2018 Accepted: 05 March 2019 Published: 26 March 2019

#### Citation:

Bueno GAS, Gervásio FM, Ribeiro DM, Martins AC, Lemos TV and de Menezes RL (2019) Fear of Falling Contributing to Cautious Gait Pattern in Women Exposed to a Fictional Disturbing Factor: A Non-randomized Clinical Trial. Front. Neurol. 10:283. doi: 10.3389/fneur.2019.00283 Materials and Methods: Forty-nine elderly women without cognitive impairment agreed to participate. Participants were divided into four groups, considering the history of falls and fear of falling. Three-dimensional gait analysis was performed to assess gait kinematics before and after exposure to the fictional disturbing factor (psychological and non-motor agent).

Results: After being exposed to the perturbation, all showed shorter step length, stride length and slower walking speed. Those without fall-history and with high fear of falling showed greater changes and lower Gait Profile Score.

Conclusion: The gait changes shown in the presence of a fear-of-falling causing agent led to a cautious gait pattern in an attempt to increase protection. However, those changes increased fall-risk, boosted by fear of falling.

Clinical Trial Registration: www.residentialclinics.gov.br, identifier: RBR-35xhj5.

Keywords: aging, accidental falls, perception, motor skills, biomechanical phenomena

# INTRODUCTION

The study of falls and their predictors amongst the elderly has become increasingly important as the consequences of these events lead to traumatic repercussions both physically and psychologically, contributing to changes in mobility and leading to mortality (1, 2). When it does not reach fatal consequences, the fall may bring reduction in both mobility and social participation due to fear, a condition called "post-fall syndrome" (3). As a result, a vicious and dangerous cycle is generated because fear significantly reduces physical activities to protect itself from the conditions that can cause the fall, but this condition leads to increased comorbidities that promote an increased risk of falls (4).

The fear of falling (FOF) is reported as one of the main predictors of falls (5–8). It is as important as impaired balance (9) or, even more important than the history of falls, since it is present even in Bueno et al. Adaptation of the Gait in the Fear of Falling

the older adults who never fell (10). Applying cognitive theory in the study of fear, it is observed that the subject, when exposed to challenging situations, should not only present necessary skills, but believe that they can deal with them (11). Thus, the study of FOF is based on the concept of self-efficacy, establishing itself by the combination of abilities, motivation, and confidence (12).

As well as fall-risk, the fear of falling is a multidimensional phenomenon, influenced by physical, psychological, social and functional factors (3). Several characteristics are related to fear: being female (13–15), older (15), having poor perception of health (14), higher dependence in the activities of daily living (14, 15), reduced muscle strength (15, 16), impaired balance (14, 15, 17) and previous history of falls (14–16).

In dynamic activities the fear of falling is presented with the adoption of a cautious gait pattern, with significant reductions in different parameters, in particular the walking speed (16, 18, 19). The spatiotemporal and kinematic parameters have been reported as critical clinical tools for assessing the risk of falls in the older adults (20–23). However, the lack of investigations of the extrinsic interferences in gait behavior in older adults, makes the ability of these parameters to predict falls in the elderly population not be clear (24).

The mechanisms underlying the relationship between FOF and falling are not well known, and little attention has been given to the study of their relationship creating a research gap (25). Investigations on gait pattern changes during adverse situations, using obstacles, floor interferences, provoking slippage or footwear modifications have already been done (26–29), however no relationship between gait adaptations and FOF were found. One of the possible methods to investigate the influence of FOF without exposing the participant to unnecessary risks is the application of the "affordances" theory. Proposed in 1979 (30, 31), the "affordances" theory has been applied to neuromotor behavior (32), determining that a visual object can potentiate motor responses even in the absence of actual intention or execution of the task proposed by this object (perception drives action) (33). In some behavioral experiments applying the theory, studies show that they have shown that actions can be enhanced after seeing an image of an object that offer some kind of action, but do not do it (34). Findings provide additional support for the notion that the physical properties of objects automatically activate specific motor codes, but also demonstrate that such influence is rapid and relatively short (32).

Differently from previous studies investigating gait modifications arising from motor perturbations (35), the main aim of this study is to investigate gait kinematic changes in the elderly women exposed to a fictional disturbing factor, using Theory of Affordances. Our secondary aims are: to analyze the gait pattern after disturbance in the elderly women stratified by fall-history and fear of falling; investigating whether demographic factors, cognition and muscle strength can be associated with gait modifications.

# MATERIALS AND METHODS

# Study Design

This controlled, non-randomized, clinical trial was approved by the Research Ethics Committee of the University of Brasília-College of Ceilândia, decision number 2.109.807 and was conducted in accordance with the Declaration of Helsinki (36). The study was registered in the Brazilian Registry of Clinical Trials (ReBEC) with the code RBR-35xhj5, receiving the number U1111-1222-4514 from the International Clinical Trials Registry Platform (ICTRP) and followed the recommendations of CONSORT (Consolidated Standards of Reporting Trials) (37).

# Participants

Participants were invited to participate in the study which was conducted at the Dr. Cláudio de Almeida Borges Movement Laboratory of the State University of Goiás, Goiânia, Brazil, from August to November 2017. The inclusion criteria were: (i) woman; (ii) age 65 or over; (iii) independent walking without aids; (iv) body mass index (BMI) < 30 kg/m<sup>2</sup> (38); (v) preserved cognition (Mini-Mental State Examination >24) (39) and >14 points considering the participants the educational level, with illiterate participants (40); (vi) declare that she has not ingested alcoholic beverages within 24 h prior to data collection; (vii) has no prior contact with any gait analysis lab or equipment. The exclusion criteria were: (i) previous surgeries in the lower limbs, pelvis or spine; (ii) have medical diagnosis of rheumatoid arthritis, neuromuscular or neurodegenerative disease, including diabetes mellitus; (iii) visual impairment; (iv) inclusion in other trials. All eligible participants were informed and signed the consent form.

The sample size was determined using G∗Power software 3.1.9.2 (Franz Faul, Universitat Kiel, Germany) (41), considering one-way variance (ANOVA) of the GPS (Overall) index obtained after perturbation. Thus, the sample required to detect a significant and clinically relevant difference from FOF exposure was N = 40 (n = 10, per group), effect size (ω²) = 0.82, p < 0.05, power 0.99.

# Experimental Setup

The participants answered a fall-history questionnaire reporting fall events over the last 12 months. A fall was defined as an "unexpected event in which the participant finds herself on a lower level" (42). To assess FOF, we used the Falls Efficacy Scale-International in its validated version to the Brazilian population (43). It provides information on level of concern about falls for a range of daily activities through 16 questions, each scoring from 1 (not concerned at all) to 4 (very concerned). The final score ranges from 16 to 64. Scores under 27 reveal low concern and over that point, high concern (44). Participants were then assigned into four groups: Faller with low FOF (Fall-LFOF), faller with high FOF (Fall-HFOF), non-faller with low FOF (NonFall-LFOF) and non-faller with high FOF (NonFall-HFOF).

# Data Collection

To perform 3D gait analysis we used the Vicon System (Vicon Motion Systems Ltd <sup>R</sup> , Oxford Metrics Group, Oxford, UK) and the Conventional Gait Model for biomechanical modeling. All data were sampled at 120 Hz and processed using a fourth-order Butterwoth filter with 10 Hz cut-off frequency (45).

Each volunteer walked barefoot over a 9 meters walkway at a self-selected speed. Two fixed squared metal plates were added at midpoint over the course (**Figure 1** in **Supplement A**). Prior

to data collection they went through the walkway five times for familiarization.

After 5 undisturbed gait trials, the participants were warned that the fixed squared objects on the floor could strongly vibrate or deliver electrical discharges when stepped over, introducing a fictional disturbing factor (FDF) to create FOF. Only 2 more trials were collected after introducing FDF to keep participants from getting used to the fictional stimuli (32).

Maximum voluntary isometric contraction (MVIC) was assessed using a manual dynamometer (Laffayete Instrument <sup>R</sup> Evaluation, Ohio, USA) testing the following muscle groups: hip flexors, extensors, adductors and abductors; knee extensors and flexors; ankle dorsiflexors and plantarflexors. Each muscle group was tested 3 times for 5 s with 1-min rest in between. The highest value was used for analysis. The subject was positioned as standardized by others (46). Right and left side's recordings were averaged and normalized by BMI (47). MVIC was collected after gait trials to avoid muscular fatigue effect on gait pattern (48).

# Data Processing

All kinematic data were normalized by the gait cycle using 51 time-normalized samples for each stride. The averaged gait data pre and post-FDF for right and left sides and for each of the four study groups were analyzed.

The Gait Profile Score (GPS) were used to calculate the quality of gait kinematic parameters (49). The GPS consists of nine gait variable scores (GVS) representing the pelvis, hip, knee and ankle kinematic data, presented in degrees. GVS scores can indicate which joint movement abnormalities tend to contribute to a high (worse) GPS. Both scores were calculated as recommended by

#### TABLE 1 | Descriptive and comparative data between NonFall-LFOF, NonFall-HFOF, Fall-LFOF and Fall-HFOF groups.


A, NonFall-LFOF; B, NonFall-HFOF; C, Fall-LFOF; D, Fall-HFOF. Comparative analysis performed by ANOVA one way, considering the F ratio, effect size (ω) and significance of α ≤ 0.05. Post Tukey post hoc analysis, considering effect size (r) and significance of α ≤ 0.05.

Baker and colleagues (49, 50). In this study, the normal group to calculate GPS consisted of 15 women adults with an average age of 24.8 ± 6.8 years old. The data set contained five trials from each subject, resulting in 75 cycles on each lower limb.

# Confounders

Confounders such as age, gender, body weight, body height, BMI were controlled, as well as others that are known to be associated with both fall and FOF repercussions: cognitive level (14); muscle strength (15, 16); and historical fall (14–16).

# Statistical Analysis

Statistical analysis was performed with SPSS Statistics version 23.0 (IBM, Chicago, USA). To assess the normal distribution the Shapiro-Wilk test was used. Tukey's post-hoc analysis of variance (ANOVA) was used to analyze the differences between the four groups in the two moments of the study, considering the effect size for the variance (ω) and post-hoc comparison. The effect of exposure to FOF agent was analyzed by applying the paired t-test, considering the effect size. In order to evaluate the relationship between discriminative variables, muscle strength and temporal space parameters with GPS, the Pearson product correlation was calculated. Correlation of r ≤ 0.3 was considered "weak," 0.31 to 0.69 "substantial" and ≥ 0.7 "strong" (51). The standard level of significance used was 0.05.

# RESULTS

# Demographic Characteristics

During the study period, 91 senior women were eligible to participate in the study. Of these, 52 signed the consent form



Comparative analysis performed by paired t-test, considering the equation, effect size (r) and significance of α ≤ 0.05.

and participated in the previous evaluation for allocation of the groups. At the end of the study, however, 49 participants remained, being NonFall-LFOF (n = 12); NonFall-HFOF (n = 15); Fall-LFOF (n = 12); FallHFOF (n = 10), according to the conditions presented in the flowchart (**Figure 1**). The results discard the absence of interference of confounders such as age, weight, BMI, as homogeneity was found between groups (p < 0.05; **Table 1**).

# Intergroup Comparison of Gait Parameters and MIVM

The step length, stride length, and walking speed showed significant differences between the groups (p < 0.05). However, the paired comparison highlighted the NonFall-HFOF group (r > 0.40), with reduced walking speed and shorter length in spatial variables pre-FDF. After FDF, only the stride length was different between groups, being lower in the NonFall-HFOF group (**Table 1** in **Supplement A**).

The GPS was not different between the groups, pre-FDF. Three parameters of GVS (Left Ankle Dor/Plan; Left Hip Int/Ext; Right Hip Int/Ext) presented differences between groups (p < 0.05) (**Table 2** in **Supplement A**).

After the FOF perturbation, the GPS (Left) and GPS (Overall) presented differences with significant effect between the groups, and the post hoc comparison showed only difference between NonFall-HFOF / Fall-LFOF groups, where again NonFall-HFOF presented higher degree of variation in both parameters (**Table 2** in **Supplement A**).

The difference in MVIC was observed only in the muscular group of the plantiflexors between study groups [F(3.45 <sup>=</sup> 2.809), p = 0.050, ω = 0.13], but did not present significant values in the comparison between the pairs (**Table 3** in **Supplement A**).

# Intra-group Comparison of pre and Post-exposure Gait Parameters

After the FDF the modifications of the spatiotemporal parameters were similar between NotFall-LFOF and NotFall-HFOF groups. The opposit foot off and the foot off were late, there was increase of the double support, and reductions were observed in the stride length, walking speed, and the step length reduced only in the NotFall-HFOF group (p < 0.05; **Table 2**). The Fall-LFOF and Fall-HFOF groups presented reduction of the same variables, being the stride length, step length and walking speed (p < 0.05; **Table 3**).

The parameters of the GPS (Left, Right and Overall) did not increase after FDF only in the Fall-HFOF group, however this group already had GPS higher than the other pre-FDF groups (**Tables 4**, **5**). The GVS data show that pre-FDF in all groups the major contributing joints in the GPS range were hip and knee. After the FDF, these joints increased their variations in all groups, remaining as the main responsible for the GPS modification (**Tables 4**, **5**).



Comparative analysis performed by paired t-test, considering the equation, effect size (r) and significance of α ≤ 0.05.

# Intra-group Correlations Between Confounding Variables and Gait Parameters Pre and Post-exposure to the FOF Agent

The correlation between muscle strength and GPS, showed that the reduction of muscle strength of hip extensors and flexors, and knee flexors contributes to worsening post-FDF gait quality in the NotFall-LFOF group (r > 0.6; p < 0.05). A similar relationship was found for knee flexors in the Fall-LFOF group (**Supplement B**).

In the spatiotemporal parameters, correlations were found with the variation of the GPS with the late opposit foot off, late foot off, and increase of the double support. In the NotFall-HFOF group these correlations were observed pre-FDF, and post-FDF increased (r > 0.6; p < 0.05). Already in the Fall-LFOF group this correlation appeared only post-FDF. And in the Fall-HFOF group, pre-and post-FDF, the correlation was found only between the increase of the double support and the late foot off (**Supplement B**).

# DISCUSSION

This study aimed to examine the gait pattern adopted by older women exposed to FOF perturbation, and how this factor affects faller and non-faller, with low and high FOF, reflecting in worsening or not the spatiotemporal parameters, GPS and GVS. Significant results pointed to different gait patterns pre and post-FDF. After exposure, all groups presented a reduction in stride length, step length and walking speed, assuming a "cautious" pattern.

Results showed that non-fallers with high FOF change their gait pattern to a cautious gait more than fallers do. The decrease of spatiotemporal variables contrasts with studies that highlight more significant decreases amongst elderly fallers (52, 53). The fact that changes were higher in the presence of FOF than with history of falls agrees with another investigation (48). The introduction of a FOF perturbation during gait resulted in a reduction of the stride length, more significantly in subjects with FOF without fall-history. However, the caution observed by the modifications of other spatiotemporal parameters was similar between groups. This same behavior may be due to declines in the attention process in dynamic or disturbed motor activities, generated by the aging process, where motor slowing are required so that attention on the proposed object remains high (52).

Investigation of FOF effect on the nervous system shows that there is no relation with cognitive decline (54), so the understanding generated by the information offered in the experiment does not differentiate the participants by cognitive interference. The FOF tends to generate an illusory motor image in these older adults, where they feel more agile (Time Up and Go


TABLE 4 | Comparison of GPS and GVS parameters between pre and post fictional disturbing factor for each of NonFall-LFOF and NonFall-HFOF, groups.

Comparative analysis performed by paired t-test, considering the equation, effect size (r) and significance of α ≤ 0.05.

test) than they actually are (25). Thus, assuming a motor pattern that does not match the necessary modifications, not preparing for a motor perturbation that they may suffer.

The sum of the two clinical conditions "to have FOF" and "to have fallen," together potentiate a gait pattern with opposite and unconscious protection effect. This fact may justify how history of fall and FOF are great predictors of falls (44) since they lead to a pattern of locomotion that predisposes to fall and does not avoid it. The same is observed by other studies that point to the increase in the risk of falls due to the slowing of walking speed (55–57), increased double support (24, 55) and stride length shortening (24). Also, falls prevention is linked to clinical interventions that seek to increase walking speed (58).

The use of "caution," potentiated by FOF, causes gait perturbation, with changes in the kinematic parameters (59), and the slowing of locomotion will corroborate the loss of gait quality (60). These same adaptations and consequent worsening of gait quality observed with higher intensity in our sample of


TABLE 5 | Comparison of GPS and GVS parameters between pre and post fictional disturbing factor for each of Fall-LFOF and Fall-HFOF, groups.

Comparative analysis performed by paired -test, considering the equation, effect size (r) and significance of α ≤ 0.05.

elderly women who presented high FOF and no fall history. Compensations in kinematics to avoid the reduction of gait quality are noted by all groups, where they prolong the timing of opposite foot off (61), and foot off (62), occurring due to weight transfer and foot release being the less stable periods of the gait cycle (61, 62).

The adjustments to try to maintain the gait quality seem to be inefficient since it was observed that the larger joints such as hip and knee are the greatest responsible for gait abnormality in this sample. A meta-analysis shows that to maintain gait quality with advancing age the hip increases its contribution, but they do not explain to what extent this increase in contribution is good or not to reduce the risk of falls (63). Our data show that the joints of the hip and knee were in all groups the joints that contributed the most to the variation of normal gait measured by the GPS, after perturbation. Studies have indicated that these joints are the ones with the most variations in segmental coordination in periods of gait instability (62–64). Moreover, the motor variation of these joints is more considerable in the presence of FOF (65, 66) and intensified by the need for an organization to an unexpected perturbation or obstacle during walking (65).

Because of that, the strategy to reduce the spatiotemporal parameters of gait is an attempt to promote greater time adjustment, in the dynamic segmental coordination, promoting caution, when going through the disturbing factor. In situations where older adults need to maintain a gait pattern and ensure attention to a stimulus, they end up prioritizing the maintenance of a "cautious" gait pattern in order to reduce the risk of falling (67). It is known that in older adults with fall-risk, gait adaptability in situations that demand attention and adjustment is weakened, and the lack of adaptability increases the risk of falling (68), seek in "caution," to reduce them with a slower gait when approaching targets or obstacles to locomotion (68). However, in the presence of FOF, the adjustments in gait pattern predispose an increase in the risk of falling and do not have the expected protective effect (24, 67, 69), worsening the quality of gait.

FOF produces anxiety in an attempt to predict the effects of a threatening stimuli that can compromise a task, leading to a memory block of usual motor tasks (70, 71), causing them to adopt a more energetic dynamic posture to try to avoid the loss of balance during threatening situations (18, 19). However, this changes compromise performance in dynamic and demanding functional tasks such as walking, leading to the inadequate acquisition of sensory information necessary to plan and execute postural adjustments in these threatening situations (70). When a target is given or alerted to a stimulus evoking FOF, the older person attempts to focus on the target visually, but when close to it, tends to look away from the target, resulting in worse accuracy to hit the target (72). In the anticipated state that the anxiety generated by the FOF promotes, it increases the risk of falling because it produces a step and an inaccurate displacement (70, 71).

Our findings on the influence of confounders on the interpretation of the effects obtained by the exposition to the disturbing factor highlighted that only the muscular strength of large muscle groups acting on the large joints such as hip and knee presented interferences. This relationship was only observed in those who fell and did not fall with low FOF, corroborating that there is no association between muscle strength and FOF (48). However, exposure to a perturbation of fall showed that the needs of gait adjustments is not conditioned to muscle strength. Thus, we pointed out that the FOF contributes more than fallhistory, cognitive level and muscle strength, on the modifications of walking parameters after exposure to a fear agent. Our findings agree with another investigation (73) showing that fall-risk increases only when there are high FOF and poor gait quality.

In the past, the combination of motor skills, motivation, and trust was the most important concept of self-efficacy (11, 12). The subject needs to overcome the FOF in challenging situations, promoting adjustment skills, but also believing that he or she can cope with them (74, 75). It is reasonable to hypothesize that interventions to fall-prevention need to incorporate conditions beyond what is observed in the musculoskeletal system and its functions. The complexity of this is what should move future research addressing the relationship between structure/function of the body and psychological factors.

The findings of this study should also be regarded with some limitations. First, this study was limited by its small sample size, although we followed the values indicated in the sample calculation and considered the homogeneity of demographic variables in the study of aging. A second limitation is that this study was restricted to a group of elderly women, and the findings may differ from elderly men. What is emphasized here is that in the future more external relations may be incorporated in studies of the motor modifications of the elderly population, and thus contributing to prevention and reduction of the risk of falling, with a greater understanding of its complexity and better interpretation for the clinical practice.

# ETHICS STATEMENT

This controlled, non-randomized, clinical trial was approved by the Research Ethics Committee of the University of Brasília-College of Ceilândia, decision number 2.109.807. The study was registered in the Brazilian Registry of Clinical Trials (ReBEC) with the code RBR-35xhj5, receiving the number U1111-1222- 4514 from the International Clinical Trials Registry Platform (ICTRP) and followed the recommendations of CONSORT (Consolidated Standards of Reporting Trials) (37).

# AUTHOR CONTRIBUTIONS

GB: analysis and interpretation of the data, study concept, wrote the manuscript. DR, AM, and TL: analysis of data, critical revision of the manuscript for important intellectual content. FG and RdM: study concept and design, study supervision, critical revisions of the manuscript for important intellectual content.

# FUNDING

This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior–Brasil (CAPES)– Finance Code 001.

# ACKNOWLEDGMENTS

We acknowledge and thank the support of the researchers from the Dr. Cláudio de Almeida Borges Movement Laboratory of the State University of Goiás.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fneur. 2019.00283/full#supplementary-material

# REFERENCES


the american physical therapy association. Phys Ther. (2015) 95:815–34. doi: 10.2522/ptj.20140415


**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 Bueno, Gervásio, Ribeiro, Martins, Lemos and de Menezes. 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.

# Avoiding Virtual Obstacles During Treadmill Gait in Parkinson's Disease

Chiahao Lu\*, Emily Twedell† , Reem Elbasher, Michael McCabe† , Colum D. MacKinnon and Scott E. Cooper

Department of Neurology, University of Minnesota, Minneapolis, MN, United States

Falls often occur due to spontaneous loss of balance, but tripping over an obstacle during gait is also a frequent cause of falls (Sheldon, 1960; Stolze et al., 2004). Obstacle avoidance requires that appropriate modifications of the ongoing cyclical movement be initiated and completed in time. We evaluated the available response time to avoid a virtual obstacle in 26 Parkinson's disease (PD) patients (in the off-medication state) and 26 controls (18 elderly and 8 young), using a virtual obstacle avoidance task during visually cued treadmill walking. To maintain a stable baseline of stride length and visual attention, participants stepped on virtual "stepping stones" projected onto a treadmill belt. Treadmill speed and stepping stone spacing were matched to overground walking (speed and stride length) for each individual. Unpredictably, a stepping stone changed color, indicating that it was an obstacle. Participants were instructed to try to step short to avoid the obstacle. By using an obstacle that appeared at a precise instant, this task probed the time interval required for processing new information and implementing gait cycle modifications. Probability of successful avoidance of an obstacle was strongly associated with the time of obstacle appearance, with earlier-appearing obstacles being more easily avoided. Age was positively correlated (p < 0.001) with the time required to successfully avoid obstacles. Nonetheless, the PD group required significantly more time than controls (p = 0.001) to achieve equivalent obstacle-avoidance success rates after accounting for the effect of age. Slowing of gait adaptability could contribute to high fall risk in elderly and PD. Possible mechanisms may include disturbances in motor planning, movement execution, or disordered response inhibition.

Keywords: obstacle avoidance, Parkinson's disease, adaptive gait, treadmill, postural control

# INTRODUCTION

Parkinson's disease (PD) is a neurodegenerative disease that leads to a progressive decline in motor function, including symptoms of rigidity, tremor, bradykinesia, and postural instability and gait disorder. The gait disorder is one of the most disabling motor symptoms of PD, and at the same time, one of the most refractory to treatment, such as pharmacological (e.g., levodopa) and neuromodulatory (e.g., deep brain stimulation) treatments. One debilitating consequence of postural instability and gait disorders is increased fall risk. Over 60 percent of people with PD reported at least one fall and 70 percent of these people experienced 2 falls within a year (Wood et al., 2002). The prevalence of falls also significantly increases with disease progression (Hiorth et al., 2014).

#### Edited by:

Eric Yiou, Université Paris-Sud, France

#### Reviewed by: Jacques Duysens,

KU Leuven, Belgium Teddy Caderby, Université de la Réunion, France

> \*Correspondence: Chiahao Lu luxxx214@umn.edu

#### †Present address:

Emily Twedell, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD, United States Michael McCabe, Trinity School of Medicine, Kingstown, Saint Vincent and the Grenadines

> Received: 03 January 2019 Accepted: 19 March 2019 Published: 09 April 2019

#### Citation:

Lu C, Twedell E, Elbasher R, McCabe M, MacKinnon CD and Cooper SE (2019) Avoiding Virtual Obstacles During Treadmill Gait in Parkinson's Disease. Front. Aging Neurosci. 11:76. doi: 10.3389/fnagi.2019.00076

**127**

Falling can often occur when individuals with PD fail to negotiate an obstacle during walking (Imms et al., 1977; Campbell et al., 1990; Chen et al., 1996; Defer et al., 1999; Galna et al., 2009; Brown et al., 2010). To prevent this from happening, one needs to inhibit the preplanned step and modify the ongoing cyclical movement to avoid the obstacle, i.e., a form of gait adaptation. Previous research on obstacle avoidance has indicated that movement performance was worse in people with PD compared to healthy individuals. This includes slower crossing speed (Brown et al., 2010), shorter stride length, greater stride and stance phase duration (Stegemöller et al., 2012; Vitório et al., 2010), and a larger step width (Galna et al., 2013b) and an increased number of steps when the obstacle appeared (Caetano et al., 2018). In addition, these gait impairments appear to be associated with the severity of the overall motor symptoms (Michel et al., 2009; Stegemöller et al., 2012; Galna et al., 2013b; Caetano et al., 2018).

While the spatial characteristics of movements to avoidance obstacles have been widely studied, it is also important to understand how the temporal presentation of an obstacle affects the performance. Previous literature in healthy adults has indicated that the timing of obstacle presentation can have a substantial impact on success rate for obstacle avoidance in both young and older healthy adults (Chen et al., 1994, 1996). Their results show that the success rate increases with the available response time relative to obstacle appearance. To our knowledge, no study has investigated how timing of obstacle appearance in people with PD affects their performance with respect to the available response time.

We examined obstacle avoidance in relation to available response time in people with PD compared to healthy controls during treadmill walking. In order to evaluate the performance of avoiding obstacles independent of difficulty keeping up with the speed of the treadmill, we measured each individual's preferred overground walking speed, and examined obstacle avoidance at that speed. We hypothesized that: (1) people with PD would require more response time for obstacle negotiation compared with healthy controls, and (2) the required response time would be associated with overall disease severity in people with PD.

# MATERIALS AND METHODS

# Participants

Fifty-two participants (demographics in **Table 1**) were included in the study: 26 people with PD and 26 healthy controls. All patients included in the study had idiopathic PD (Hoehn and Yahr scale of II–IV) (Hoehn and Yahr, 1967). Participants with PD were tested in the morning after 12-h withdrawal from anti-parkinson's medications (practically defined off-medication state) (Defer et al., 1999). Eight of the PD participants had implanted deep brain stimulators. In these individuals, stimulation was turned off at least 1-h before the testing. Exclusion criteria included medical conditions impairing gait except for PD and diagnosis of dementia or a score of <25 on the Mini-Mental State Exam (Folstein et al., 1975). All participants TABLE 1 | Summary of participants demographics.


<sup>∗</sup>Values are presented as mean ± SD. #MDS-UPDRS stands for Movement Disorder Society-Sponsored Unified Parkinson's Disease Rating Scale Revision. \$Two participants were excluded due to missing scored items. <sup>∧</sup>One participant was excluded due to missing scored axial items. @Not an error: mean ± SD walking speed was the same for Control as for Control (age > 45). N/A, not applicable; DBS, deep brain stimulation; STN, subthalamic nucleus; GPi, globus pallidus internus.

gave informed consent and the protocol was approved by the University of Minnesota Institutional Review Board.

# Protocol

Participants initially walked overground at a self-paced speed and at least 30 valid steps were acquired on an instrumented mat (Gaitrite, CIR systems, Franklin, NJ, United States) [see Galna et al. (2013a) for details on the overground walking protocol]. Average walking speed and step length were calculated and exported immediately after the completion of overground walking. Next, a 14-min practice walking session was conducted to allow participants to familiarize themselves with the treadmill (C-Mill, Motekforce Link, Culemburg, Netherlands), including 5 2-min trials at the participant's previously measured overground walking speed, followed by 2 2-min trials at 15% faster and at 15% slower than the overground speed in random order. All practice trials were tested without projection of virtual objects. Participants were asked to wear their usual comfortable walking shoes during the overground and all treadmill walking sessions.

Finally, the participants performed the obstacle avoidance task on the treadmill (0.7 m × 2.5 m) at their overground gait speed. During this task, stepping stones (blue squares, size: 0.3 m × 0.3 m), were projected onto the treadmill belt, corresponding to left and right footsteps (0.15 m leftright distance between the center of the squares), and spaced according to the individual's average step length measured during overground gait. Movement of the stepping stones matched the speed of the treadmill belt, as if painted onto it, and participants were instructed to step on each stepping stone with the corresponding foot. Randomly and unpredictably, a stepping stone would change color (from blue to a redwhite striped square) indicating that it was now an obstacle

to be avoided. Participants were instructed to "step short" to avoid the obstacle (**Figure 1A**). Despite these instructions, they occasionally avoided an obstacle by stepping long ("overstep") [overstep may be a preferred strategy for elderly (Weerdesteyn et al., 2005a,b)]. When overstep happened, they were reminded of the instructions.

The timing of obstacle appearance was generated by an algorithm of the treadmill software provided by the manufacturer. This algorithm monitored real-time heel-strike and toe-off events detected from the center of pressure (measured by the force plate built into the treadmill). From this data, it predicted the time of occurrence of the next footfall and presented an obstacle prior to the next footfall (one step before) or the one after the next foot fall (two steps before) on a random subset of gait cycles. Fortuitously, the software generated random variance in the timing of obstacles, and we exploited this feature in the present protocol. Since the variance was not under our direct control we simply measured the actual obstacle latencies relative to toe-offs. The timing of obstacle appearance relative to the previous ipsilateral toe-off was a bell-shaped distribution with 95% range from −1.0 to 0.5 s (**Figure 1B**). In addition, the distribution of both obstacle latency and available time to respond were similar between groups (**Figure 1E**, see **Supplementary Material** for more details).

Each participant performed six total trials. In order to reduce the predictability of obstacles, the proportion of stepping stones which changed to obstacles was set to either 1/9 on 3 of the trials or 1/6 on the other 3 trials, with the ordering varied randomly. The trial duration was 2 min for the 1/6 condition and 3 min for the 1/9 condition, which equalized the number of obstacles across trials. A safety harness was worn by all participants during treadmill walking. The harness was attached to an overhead sliding track which afforded protection against falls, without restricting the participant's motion. Seated rest periods (2–3 min) were offered after each trial or upon participant's request. For three participants who could not perform the obstacle avoidance task with 100% overground walking speed, the treadmill speed was reduced. For two participants it was reduced to 85%. In one of these participants, the stepping-stone spacing was modified accordingly based on step lengths measured during the 85% speed

presented to all participants separated by group.

obstacle latency (top x-axis) in one representative participant; (E) density plot of obstacle latency (left)/available time to respond (right) for total number of obstacles

practice treadmill walking, whereas the other participant was tested at 85% speed but with the stepping-stone spacing from 100% overground walking. For the third participant, who could not perform obstacle avoidance at 100% speed, the treadmill speed was reduced to 72% and stepping-stone spacing set to the step length measured during the 85% speed practice treadmill walking trial (see **Supplementary Material** for more details).

Task performance was captured using a digital video camera (HDR-CX700, Sony, Tokyo, Japan) positioned in front of the treadmill to capture the treadmill belt and the lower extremities of the participant. Video was recorded at 60 fields per second with a spatial resolution of 720p.

# Data Analysis

The performance of obstacle avoidance was scored from the video recordings. In order to avoid bias in video processing, a second person, blinded to participant status (PD vs. Control) reprocessed the video data (see **Supplementary Material** for agreement analysis between two raters). Obstacle avoidance was classified as a success when the participant avoided stepping on the obstacle, otherwise it was classified as a failure. Each success was further categorized by two avoidance strategies, i.e., stepping short of the obstacles (as instructed: "short-step") and stepping beyond ("overstep"). The obstacle latency was measured from the video (60 Hz, i.e., 1/60th second resolution) and defined as the time of obstacle appearance relative to ipsilateral toe-off (specifically, to the start of the swing phase which was terminated either successfully or unsuccessfully by either avoiding or stepping on the obstacle). The difficulty of avoiding the obstacle with a given obstacle latency will increase with increasing treadmill speed and decrease with increasing distance between stepping stones. For example, if an obstacle with the same obstacle latency and spacing was presented to two participants walking at different speeds, it would be more difficult for the participant with the faster walking speed to avoid the obstacle. On the other hand, if the obstacle was presented with the same obstacle latency and walking speed to two participants, but the obstacle spacing was different, it would be more difficult for the participant with the smaller obstacle spacing to avoid the obstacle. To account for this confound, we computed available time to respond (**Figure 1C**) with the following equation:

$$ART = OL + \text{SS} / TS$$

where ART, available time to respond (s); OL, obstacle latency to toe-off (s); SS, steppingstone spacing (m); TS, treadmill speed (m/s).

We repeated also our analysis using the raw obstacle latency data without this adjustment (see **Supplementary Figure S2**). This did not affect the results.

This is the time interval from obstacle appearance until the ipsilateral foot would hit the obstacle based on an average, unmodified gait cycle (Chen et al., 1994; Weerdesteyn et al., 2005a; Brown et al., 2006; Maidan et al., 2018). The primary dependent variable was defined as the time to respond at which the probability of success was 50% (ttr50). The ttr<sup>50</sup> was computed for each participant by logistic regression of success/failure on time to respond (**Figure 1D**). To ensure reliability, the p-value of the logistic regression for each participant was verified to be less than 0.00001. The percentage of each type of success (short-step, i.e., as instructed, vs. overstep, i.e., contrary to instructions) was also calculated.

# Statistical Analysis

The primary analysis was a comparison of ttr<sup>50</sup> between PD and Control groups. Because ttr<sup>50</sup> varied with age (lower for younger participants) and age differed between PD and Control groups (due to the inclusion of a group of 8 Control participants less than 45 years of age), the comparison was an analysis of covariance (ANCOVA) with factor of group (PD vs. control) and age as covariate. For additional confirmation, an independent t-test on ttr<sup>50</sup> was conducted, excluding the group of young Control participants.

The same ANCOVA and independent t-test or Mann– Whitney test (for skewed distribution observed in overstep success rates) were also applied to success rates as the secondary analysis. Finally, in the PD group, simple linear regression analysis was used to examine the relationship between ttr<sup>50</sup> and predictors [disease duration and motor scores of Movement Disorder Society-Sponsored Unified Parkinson's Disease Rating Scale Revision (MDS-UPDRS)], respectively. Two-tailed p-value threshold was set at 0.05.

# RESULTS

A summary of the demographics of the participants is listed in **Table 1** by group. The average age was significantly greater in the PD group than the controls [t(50) = −2.7, p = 0.01] due to the inclusion of the eight young Controls and one young PD. The average age was not significantly different between groups when we excluded participants younger than 45 years old [t(41) = −0.3, p = 0.75]. There was no difference in the proportions of male vs. female participants between the PD and control groups [all age, X 2 (1,52) = 0.08, p = 0.78; age > 45, X 2 (1,42) = 0.31, p = 0.58].

A small fraction of obstacle events was excluded from the analysis (0.5% in Control and 2.7% in PD) due to: footfalls not aligned with stepping stones on the gait cycle preceding the obstacle, when the participant avoided the obstacle by stepping to one side of it, or by keeping the ankle dorsiflexed so as to bear weight only on the heel. The performance of obstacle avoidance

TABLE 2 | Summary of performance in obstacle avoidance (mean ± SD).


ttr50, the time to respond at which the probability of success was 50%.

by each group is listed in **Table 2**. About 4 out of 5 successes were classified as stepping short (as instructed) in both control (mean ± SD, 77.6 ± 16.7%) and PD (80.6 ± 13.2%) groups.

The average ttr<sup>50</sup> value was approximately 1.2 times greater in the PD group compared to healthy controls (**Figure 2A**). A simple between-group comparison was significant [t(50) = −4.4, p < 0.001]. The ANCOVA controlling for age confirmed the results showing a significant main effect of group in the ttr<sup>50</sup> [F(1,49) = 12.2, p = 0.001]. The age covariate was also significant [F(1,49) = 13.4, p < 0.001, **Figure 3A**]. For further confirmation, a simple between-group comparison, excluding participants younger than 45 years, was also significant [t(41) = −3.1, p = 0.003]. Results were the same when we repeated this analysis using raw obstacle latency rather than available time to respond (see **Supplementary Figure S2**).

In the secondary analysis, success rates were higher in controls than in PD participants (**Table 2**), but this difference did not reach significance in ANCOVA [F(1,49) = 0.16, p = 0.69; t-test, t(41) = 0.06, p = 0.95, **Figure 2B**] though the age covariate was significant [F(1,49) = 9.7, p = 0.003]. We also found no significant group effect for short-step success rates [t(41) = −0.78, p = 0.44] in participants older than 45 years of age. For overstep success rate, the between-group difference in participants older than 45 years of age was also not significant (U = 274, p = 0.24).

Linear regression analysis in the PD group showed MDS-UPDRS motor score [F(1,22) = 2.08, p = 0.08, R <sup>2</sup> = 0.08, two participants excluded due to incomplete scores] was not significantly related to the ttr<sup>50</sup> although there was a nonsignificant trend to increasing ttr<sup>50</sup> with increasing motor scores in MDS-UPDRS (**Figure 3B**). However, the MDS-UPDRS motor axial subscore [Factor 1 items in Goetz et al. (2008)] was significantly related to ttr<sup>50</sup> [F(1,23) = 5.14 (missing data in one participant), p = 0.03, R <sup>2</sup> = 0.15, **Figure 3C**]. There was no significant relationship between ttr<sup>50</sup> and disease duration [F(1,24) = 0.58, p = 0.45, R <sup>2</sup> = 0.02] although there was a non-significant trend to increasing ttr<sup>50</sup> with increasing disease duration (**Figure 3D**). There was no relation of success rate to MDS-UPDRS motor scores [F(1,22) = 0.39, p = 0.54, R <sup>2</sup> = 0.03], MDS-UPDRS motor axial subscores [F(1,23) = 2.07, p = 0.16, R <sup>2</sup> = 0.04] and disease duration [F(1,24) = 0.07, p = 0.79, R <sup>2</sup> = 0.04]. In addition, there was no difference in ttr<sup>50</sup> between PD patients with vs. without deep brain stimulator [DBS, t(24) = −1.91, p = 0.07] (All DBS participants tested OFF-stim, with at least 1-h washout) and in ttr<sup>50</sup> between more vs. less affected legs in PD patients [paired t-test, t(24) = 0.02, p = 0.98].

# DISCUSSION

To our knowledge, this is the first study to assess how PD affects the capacity for obstacle avoidance as a function of the available time to respond. Our results showed that the probability of successfully avoiding an unexpected obstacle depended strongly on the available time to respond in both controls and people with PD. In addition, we found that PD participants required more time than controls to achieve an equivalent probability of avoiding the obstacle.

Our control participants replicate the finding by Potocanac et al. (2014), who showed a similar dependence on time-torespond in a smaller sample of neurologically healthy participants (Potocanac et al., 2014). In addition to being larger, our sample of control participants also spanned a wider age range and included elderly participants. We also varied time-to-respond with higher resolution, i.e., continuously, rather than in discrete increments. The logistic regression curve relating the probability of successful avoidance to available time was shifted to the right in older as compared to younger Controls (i.e., older participants required more time than younger ones to achieve an equivalent probability of avoiding the obstacle).

The logistic regression was also shifted to the right in the PD group relative to Controls (i.e., PD participants required more time than Controls to achieve an equivalent probability of avoiding the obstacle) even after accounting for the effect of age. Thus, consistent with our hypothesis, our results demonstrate that people with PD required more time to respond (greater ttr50) in order to avoid obstacles during treadmill walking compared to healthy adults. Although ttr<sup>50</sup> increased with increasing disease duration and with increasing MDS-UPDRS total motor score, neither relation reached significance. However, ttr<sup>50</sup> was significantly related to MDS-UPDRS axial subscores. This suggests that prolongation of ttr<sup>50</sup> is predominantly associated with postural and gait dysfunction, and thus, may be a distinct domain of impairment rather than more generalized measures of disease severity.

Previous obstacle avoidance studies in people with PD used a treadmill and three-dimensional obstacles (van Hedel et al., 2006; Michel et al., 2009; Snijders et al., 2010; Nanhoe-Mahabier et al., 2012; Stegemöller et al., 2012). With three-dimensional obstacles the larger limb movement required to step over the obstacle may be additionally difficult for Parkinsonian patients due to hypometria, a spatial deficit, thereby confounding spatial and temporal impairment. In contrast, we used two-dimensional virtual obstacles with instructions to shorten the step to avoid the obstacle. This reduction in the spatial demands of the task allowed a more direct comparison of the temporal aspect of obstacle avoidance between Control and PD. In addition, although we expected the short-step instructions would provoke episodes of freezing of gait (FOG) (Chee et al., 2009), such episodes were actually rare, even in patients for whom FOG was a prominent symptom. This might simply reflect the irregular and unpredictable nature of FOG which makes it hard to provoke reproducibly in a laboratory setting, or the facilitatory effect of the visual cues ("stepping stones") in our experiments.

Shorter steps (in PD) and faster gait (in controls) reduce the available time to respond to an obstacle, for a given obstacle latency, which would be expected to reduce success probability. However, this cannot explain the difference we saw between PD and control subjects, because we analyzed success probability as a function of time-to-respond, rather than of the obstacle latency. Therefore, the prolonged response time in our PD participants cannot be completely explained by walking speed or step length.

The obstacle avoidance task used in the current study is similar to a classical response inhibition task (Go-NoGo or Stop Signal Task). People with PD do exhibit impaired response inhibition

(Gauggel et al., 2004; Dirnberger and Jahanshahi, 2013). In our task, participants were asked to alter their stride to avoid an obstacle. This differs from a simple response inhibition task as it may require not only inhibiting the default "prepotent" response, but also replacing it with a different one, much like a Stroop test (Stroop, 1935). Furthermore, we observed three different behaviors in response to the obstacle appearance. Not only did we see the prepotent (stepping on the obstacle) and the instructed response (stepping short of the obstacle), but subjects also demonstrated a third behavior, that is, overstepping so that the foot landed on the far side of the obstacle. Overstepping was deployed appropriately to achieve obstacle avoidance, but was contrary to instructions. This appears to represent a preferred obstacle-avoidance response. The fact that overstepping occurred preferentially for obstacles with shorter time to respond (see **Supplementary Material**) suggests that this strategy is another prepotent response which must be inhibited in order to comply with task instructions. Another potential explanation for overstepping could be that lengthening the step would increase the anterior-posterior base of support, thus providing increased stability in the sagittal plane. Response inhibition is often studied in highly artificial tasks. The present study may shed light on how response inhibition modulates a real-world, and functionally very important daily activity, namely gait.

Functional neuroimaging studies have provided evidence that the prefrontal cortex plays an important role in response inhibition or response switching. Maidan et al. (2016) showed a significant increase in prefrontal activation during obstacle negotiation compared to normal walking in people with PD, but this increase was not seen in aged matched healthy adults (Maidan et al., 2016). This is in keeping with the growing body of evidence relating prefrontal cortical attentional and executive function to gait (Herman et al., 2010; Lord et al., 2010; Snijders et al., 2010; Smulders et al., 2013).

Gait speed may be a proxy for PD disease stage (Paker et al., 2015), and in our PD participants, walking speed was inversely related to MDS-UPDRS motor scores. A factor that may have contributed to reduced walking speed and shortened step length

during treadmill walking in the PD group is visual information load. Participants with PD would perceive more stepping stones given the same projection area on the treadmill belt compared to control participants who walked with greater speed and step length. Thus, this additional load of visual information may exacerbate the performance deficit of these participants, which may, in part, contribute to the prolonged time to respond.

Success rates were higher in controls than in PD participants, but this difference did not reach significance. Relatively high success rates in PD patients may be a characteristic of externally cued tasks (van Hedel et al., 2006) or occur when subjects engage a subcortical "fast adjustment network" (Snijders et al., 2010; Potocanac and Duysens, 2017). We also did not find a significant relationship between success rate and the MDS-UPDRS motor scores (PD only). This is expected because success rate depends on obstacle latency. To compare success rates, time-to-respond should be matched, or alternatively, as in the present paper, one can directly compare the function relating success probability to time-to-respond.

In the current study, obstacle avoidance occurred in the context of a baseline walking task in which participants were asked to step on virtual stepping stones. The reason for this instruction was to maintain a stable baseline of stride length and visual attention, so that the step preceding an obstacle was similar for every obstacle. However, our use of stepping stones could have affected our results in several ways. First, although we attempted to set the spacing of the stones to match the individual participant's overground step length, the average of left and right step lengths was used. Since most PD participants had asymmetrical step lengths, this may have normalized the asymmetry of their treadmill gait. Future studies may investigate how gait asymmetry affects obstacle avoidance in PD since it has been shown that elderly with high fall risk may demonstrate more asymmetry in lower limbs when stepping over obstacles compared to young control and elderly with low fall risk (Di Fabio et al., 2004). Second, it is well known that providing a visual cue improves gait in PD (Lim et al., 2005). Third, the requirement to step on the stepping stone may have constituted an additional cognitive load on participants, on which the obstacle-avoidance requirement was superimposed. Therefore, a potential explanation of the increase of the available time to respond in the PD group could be the cognitive cost of dual task, which involved processing of both stepping stones and obstacles. Yet, previous research has shown that simultaneous external cueing did not affect obstacle crossing performance in PD groups with and without FOG during treadmill walking (Nanhoe-Mahabier et al., 2012) (although in that study, the cues were auditory rather than visual).

It is worth mentioning that the average available response time of our neurologically healthy older controls in obstacle crossing was higher (0.6 s) than in a previous study (0.2–0.35 s) (Weerdesteyn et al., 2005a) for similar success rates. The difference could be that the average treadmill speed was substantially slower in the Weerdesteyn et al. (2005a) study compared to the current study (0.8 vs. 1.2 m/s). In addition, their obstacle always dropped at the same location and the sound of the obstacle landing may have acted as a supplemental acoustic cue. Chen et al. (1994) reported 50% success at 280 ms available response time in healthy elderly participants when using a virtual obstacle during overground gait. The constraint on gait speed regulation imposed by the use of a treadmill in our study may have increased the difficulty of our task through a dual-task effect and thus increased the response times. Other implementation details (e.g., visual salience of the obstacles) may also have played a role.

Our study has limitations, and the findings should be interpreted with caution. First, gait could differ between overground and treadmill walking (Bello et al., 2014; Malatesta et al., 2017), although the literature on this topic is equivocal (e.g., Frenkel-Toledo et al., 2005; Riley et al., 2007; Hollman et al., 2016) and may depend on the specific population (e.g., Watt et al., 2010) and on laboratory-specific details. Second, it is possible that the gait speed and step length which were natural for overground walking were less natural on the treadmill. In that case, our results may pertain more to more challenging gait. The gait speed we chose was intended to approximate "natural," i.e., overground gait, subject to the constraint that the experiment required a treadmill. Overground gait seemed more relevant to community ambulation, activities of daily living and fall risk, etc., Although we instructed participants to step short of the obstacle, yet overstepping sometimes occurred. Thus, participants had two strategies that could be implemented, but were instructed to preferentially select one of these (step short). The process of selection and suppression of strategies may have increased response time since this could impose an additional cognitive load analogous to dual tasking, which is known to affect gait in PD (Bond and Morris, 2000; Yogev et al., 2005; Rochester et al., 2008; Amboni et al., 2013) (see **Supplementary Material** for a more detailed analysis of shortstepping/overstepping rates). Future studies are warranted to investigate how the adopted avoidance strategy is influenced by the available time to respond without the constraint of specific instructions. In addition, we were unable to extrapolate our findings to potential fall risk due to lack of comprehensive falls history of our sample population. In order to avoid statistical confounding from systematic differences in step length and gait speed between PD and control groups, we analyzed our data in terms of available time to respond, based on obstacle latency adjusted for gait speed and step length. A limitation of this approach is that that one cannot disentangle effects of speed and target spacing. However, our conclusions were unaffected when data was reanalyzed in terms of raw, unadjusted obstacle latency (see **Supplementary Material**). Another limitation was that we were unable to assess the effect of learning or fatigue on the obstacle avoidance by comparing earlier vs. later trials, or earlier vs. later obstacles within a trial. This is because we pooled all trials in order to fit the logistic regression with the best accuracy. Therefore, we cannot exclude the possibility that the difference between PD participants and controls was that the PD participants required more practice to learn the task. Finally, the obstacle latency was estimated at 60 Hz due to the time resolution of the video.

This study shows that the available time required to respond in obstacle negotiation was prolonged in people with PD

during preferred speed treadmill gait compared to neurologically healthy adults. Furthermore, the prolonged response time was associated with the severity of Parkinsonian axial motor dysfunction.

# AUTHOR CONTRIBUTIONS

CL and SC conceived the study, analyzed the data, developed the methodology, administered the project, and wrote, reviewed, and edited the manuscript. CM conceived the study, investigated the results, developed the methodology, and wrote, reviewed, and edited the manuscript. ET, RE, and MM developed the methodology, analyzed the data, and reviewed and edited the manuscript.

# REFERENCES


# FUNDING

This work was supported by the University of Minnesota Neuromodulation Innovations (MnDrive), the Udall Center grant of the National Institutes of Health Award Number P50NS098573, and the National Center for Advancing Translational Sciences of the National Institutes of Health Award Number UL1TR000114.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnagi. 2019.00076/full#supplementary-material


timing matter? Gait Posture 59, 242–247. doi: 10.1016/J.GAITPOST.2017. 10.023


**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 Lu, Twedell, Elbasher, McCabe, MacKinnon and Cooper. 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.

# Anticipatory Postural Adjustments During Gait Initiation in Stroke Patients

Arnaud Delafontaine1,2 \*, Thomas Vialleron1,2, Tarek Hussein<sup>3</sup> , Eric Yiou1,2 , Jean-Louis Honeine<sup>4</sup> and Silvia Colnaghi 5,6

<sup>1</sup> CIAMS, Université Paris-Sud, Université Paris-Saclay, Orsay, France, <sup>2</sup> CIAMS, Université d'Orléans, Orléans, France, <sup>3</sup> ENKRE, Saint-Maurice, France, <sup>4</sup> VEDECOM Institut, Versailles, France, <sup>5</sup> Department of Public Health, Experimental and Forensic Medicine, University of Pavia, Pavia, Italy, <sup>6</sup> Laboratory of Neuro-otology and Neuro-ophthalmology, IRCCS Mondino Foundation, Pavia, Italy

Prior to gait initiation (GI), anticipatory postural adjustments (GI-APA) are activated in order to reorganize posture, favorably for gait. In healthy subjects, the center of pressure (CoP) is displaced backward during GI-APA, bilaterally by reducing soleus activities and activating the tibialis anterior (TA) muscles, and laterally in the direction of the leading leg, by activating hip abductors. In post-stroke hemiparetic patients, TA, soleus and hip abductor activities are impaired on the paretic side. Reduction in non-affected triceps surae activity can also be observed. These may result in a decreased ability to execute GI-APA and to generate propulsion forces during step execution. A systematic review was conducted to provide an overview of the reorganization which occurs in GI-APA following stroke as well as of the most effective strategies for tailoring gait-rehabilitation to these patients. Sixteen articles were included, providing gait data from a total of 220 patients. Stroke patients show a decrease in the TA activity associated with difficulties in silencing soleus muscle activity of the paretic leg, a decreased CoP shift, lower propulsive anterior forces and a longer preparatory phase. Regarding possible gait-rehabilitation strategies, the selected studies show that initiating gait with the paretic leg provides poor balance. The use of the non-paretic as the leading leg can be a useful exercise to stimulate the paretic postural muscles.

#### Edited by:

Ina M. Tarkka, University of Jyväskylä, Finland

#### Reviewed by:

Ramona Ritzmann, University of Freiburg, Germany Keith M. McGregor, Emory University, United States

> \*Correspondence: Arnaud Delafontaine arnaud\_94150@hotmail.fr

#### Specialty section:

This article was submitted to Neurorehabilitation, a section of the journal Frontiers in Neurology

Received: 28 December 2018 Accepted: 22 March 2019 Published: 17 April 2019

#### Citation:

Delafontaine A, Vialleron T, Hussein T, Yiou E, Honeine J-L and Colnaghi S (2019) Anticipatory Postural Adjustments During Gait Initiation in Stroke Patients. Front. Neurol. 10:352. doi: 10.3389/fneur.2019.00352 Keywords: anticipatory postural adjustments, gait initiation, stroke, rehabilitation, balance

# INTRODUCTION

Gait initiation (GI) is a functional task classically used in the literature to investigate balance control mechanisms in healthy and pathological individuals (1, 2). GI may be a challenging task for balance control in stroke patients because of their sensory and motor impairments (3–5). Better knowledge of the altered balance control mechanisms in these patients is important to allow clinicians to target rehabilitation programs directed at minimizing the risk of falling. However, to date there is no review involving the changes in balance control mechanisms during GI in stroke patients.

GI is composed of a postural phase (i.e., "anticipatory postural adjustments," APA) where the dynamic phenomena necessary for a stable whole-body progression are generated, followed by the foot lift phase that ends at the time of swing toe clearance (i.e., body mass is transferred to the stance leg during this phase), and an execution phase that ends at the time of foot contact (2) (see **Figures 1A,B**).

FIGURE 1 | (A) Example of biomechanical traces obtained for one representative normal subject initiating gait at a spontaneous velocity (one trial). Anteroposterior direction x"G: anteroposterior center of mass velocity; z"G: vertical center of mass velocity; peak vGRF: the maximal moment of vertical Ground Reaction Force; xP, Center of pressure (COP) displacement; xPmax: maximal backward shift of CoP. Mediolateral direction yP, mediolateral COP displacement; yPmax, maximal mediolateral shift of COP. Vertical lines, t0 onset variation of biomechanical traces; FO, Swing foot off; FC, Swing foot contact. FL, Foot lift. Horizontal dashed line: dGI-APA, duration of Gait Initiation-Anticipatory Postural Adjustments; d-EXE, duration of execution phase. SOL-SW, Soleus electromyographical activity of swing leg; SOL-ST, Soleus electromyographical activity of stance leg; TA-ST, Tibialis Anterior electromyographical activity of stance leg; TA-SW, Tibialis Anterior electromyographical activity of swing leg. (B) Stick representation of the different phases and temporal events of gait initiation in a normal subject. QS, quiet standing; GI-APA, gait initiation-anticipatory postural adjustments; EXE, execution phase.

Current literature describes gait initiation APA (GI-APA) components in the sagittal and frontal planes. In the sagittal plane, GI-APA are characterized by a backward center of pressure (CoP) shift with respect to the center of mass (CoM) position, providing the initial propulsive forces to progress forward (6, 7). This CoP shift is caused mainly by a bilateral reduction in soleus tonic activity and an increase in tibialis anterior (TA) tonic activity (8). In the frontal plane, the CoP is displaced toward the leading leg during GI-APA, which serves to move the CoM in the contralateral direction. This CoM shift helps disengage the leading leg for step execution and is crucial for maintaining stability (9). This mediolateral CoP shift is thought to be caused by leading leg abduction, accompanied by trailing leg ankle dorsiflexion and knee flexion (10). In post-stroke hemiparetic patients, damage to central structures results in a reorganization of GI-APA. The goal of this systematic review was to provide an overview of the characteristics of GI-APA in this population.

# METHODS

# Methodology

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology was employed for this systematic review. A bibliographical search was conducted in PubMed and Google Scholar using the following keywords: ("gait initiation" OR "step initiation") AND ("stroke" OR "hemiparesis" OR "hemiplegia"). Only articles available in English or French and published before 31 August 2018 were selected. Additionally, manual searches were conducted from the reference list of each article found in the database until no further citations could be identified. To be included, the studies had to meet all of the following inclusion criteria: (1) anticipatory postural adjustments were explored during the paradigm of gait initiation with a force platform or electromyographical (EMG) activity (6, 8); for review see (2), (2) at least one stroke patent group was studied, (3) all types of studies (i.e., randomized and non-randomized control trials), (4) only articles in English or French, and (5) articles published in peer-reviewed journals.

GI-APA were retained as a main inclusion criteria because it plays a major role in postural stability and GI performance [for review see (2)], notably for stroke patients which are able to develop different motor strategies regarding the leg which initiates the gait [i.e., paretic or non-paretic leg; (3–5, 11, 12)].

On the contrary, studies were excluded if they observed gait without the study of anticipatory postural adjustments or if no stroke patients were included in the protocol.

# Study Selection

Study selection was conducted in three steps (13, 14). First, two reviewers independently applied the inclusion and exclusion criteria to all the citations. Then, abstracts were screened to identify the titles representing a "best fit" with the aim of the present study. Finally, the reviewers read the full articles to reach a final decision on whether they should be included in the systematic review. When the relevance of a study was unclear from the abstract, the full article was read, and disagreements were resolved through group discussions with a third expert until a mutual consensus was reached. The reviewers met at the beginning, midpoint and final stages of the review process to discuss challenges and uncertainties related to study

selection, and to go back and refine the search strategy if needed (**Figure 2**).

# Quality Assessment

The Downs and Black scale was used to assess the risk of bias and the methodological quality in the selected studies. The scale was used for its ability to provide an overview of the study quality of not only randomized controlled trials but also non-randomized studies (16). Additionally, this tool has been demonstrated to have a significant correlation with the valid PEDro scale (17, 18). The Downs and Black scale are divided into 27 questions and the answer to each can be "yes" (one point), "no" or "unable to determine" (zero point). Only the last item was modified from the original scale, as described previously by Trac et al. (16). The maximum score is 28: one point for each question, except for question five (two points).

After all of the articles were assessed and total quality scores were calculated, the level of evidence appraisal was conducted using Grades of Recommendation, Assessment, Development, and Evaluation-Confidence (GRADE) as recommended by The Cochrane Qualitative and Implementation Methods Group (19). The GRADE system classifies the quality of evidence in one of four levels: high, moderate, low, and very low. Evidence based on RCT begins as "high-quality," whereas observational studies start with a "low quality" rating. Grading upwards may be warranted if the magnitude of the treatment effect is very large, if there is evidence of a dose-response relation or if all plausible biases would decrease the magnitude of an apparent treatment effect (20).

# Data Collection

Key items were charted from the selected articles by one of the authors. Item extraction was then checked by another author and disagreements were resolved by a third. The following data were noted for each selected article (**Table 1**):


#### TABLE 1 | Clinical characteristics of the population in the selected studies.


CMSA, Chedoke–McMaster Stroke Assessment; FMA, Fugl-Meyer motor Assessment; FIM, Functional Independence measure; NS, Non-Specified; PMC, PreMotor Cortex; vs., versus.


# RESULTS

### Selected Articles

Thirty-three articles emerged from the PubMed search. Five additional articles were identified through manual searching, for a total of 38 articles, 16 of which met the inclusion criteria (3–5, 11, 12, 21–23, 25–31). A total of 220 patients were included (**Table 2**).

# Patient Characteristics

The patients' age ranged between 21 and 86 years. In 11 of the 15 selected articles, GI was compared in healthy adults vs. stroke patients (5, 11, 21–24, 27–31). In the five remaining articles, only stroke patients were included (3, 4, 22, 25, 26). In 11 articles, participants had experienced stroke at least six months prior to GI recording (4, 5, 21, 23– 26, 28–31). In one study, patients had experienced stroke 3.7 months prior to GI recording (11). In the remaining four studies, the post-stroke interval was not indicated (3, 12, 22, 27). The symptoms always consisted in hemiparesis, although its degree was rarely specified (4, 5, 25, 27–29). The presence and degree of spasticity was acknowledged only in Bensoussan et al. (4).




Frontiers in Neurology | www.frontiersin.org


# Task and Recordings

In four articles, participants were asked to perform a single step in a reaction-time paradigm, i.e., after the deliverance of a GO signal (25, 27–29). In the other articles, participants initiated gait spontaneously and continued walking at their own speed. In two articles (21, 23), the contralesional leg was systematically used as the leading leg. In the other articles, both legs were used as the leading leg.

Dynamic investigation using force platforms was conducted in all articles except one (26). EMG investigation was also conducted in five articles (3, 23, 26, 27, 31).

# Quality Assessment

The results of the quality assessments for each study are given in **Table 3**. Only non-randomized studies corresponded to the inclusion criteria and were evaluated. According to the modified Downs and Black scale, the mean quality index score for the studies was 14 ± 1.6, with scores ranging from 11 (3) to 17 (26, 29). Only non-randomized observational studies corresponded to the inclusion criteria and were evaluated, so the level of evidence was classified as "low quality." No treatment was administered in the studies and no meta-analysis was conducted, so no grading upwards could be done.

# Biomechanical and EMG Features of APA: Stroke vs. Healthy Patients

#### Biomechanics **APA pattern**

Rajachandrakumar et al. (12) showed that when participants with sub-acute stroke initiated gait at their own speed, 78% exhibited at least one trial with a GI pattern classified as "atypical." Atypical GI trial corresponded to a trial performed without APA or with "multiple" APA (i.e., with a multi-peaked CoP trace). In contrast, Martinez et al. (25) observed GI-APA in all trials that involved voluntary stepping in post-stroke patients (mean chronicity: 2.9 years), while APA were absent in 25% of trials involving perturbation-induced stepping.

# **APA duration**

Rajachandrakumar et al. (12) showed that trials with multiple APA had a longer APA duration than normal GI trials (i.e., GI with a single peak CoP trace). A longer APA duration in stroke subjects compared to healthy subjects was reported by many authors (11, 23, 25, 27, 28). In addition, Gama et al. (21) recently established that stroke patients presented lower velocity of the CoP in both the ML and AP directions during the anticipatory phase of gait initiation.

# **Center-of-pressure shift**

Sousa et al. (24) showed that post-stroke patients presented only half of the COP displacements (both in the frontal and sagittal plane) compared to healthy subjects. Gama et al. (21) demonstrated shorter ML and AP CoP displacements as well. A smaller COP backward shift was also demonstrated by Hesse et al. (11).

# **Ground reaction forces**

Melzer et al. (28) and Sharma et al. (22) showed that anteroposterior ground reaction forces generated by both the paretic and non-paretic leg in post-stroke patients during APA were lower compared to healthy controls.

Altogether, these studies indicate that stroke patients present impaired APA compared to healthy subjects, characterized by atypical patterns, longer duration and lower amplitude. The results reported in the paragraph below show that GI parameters can also differ depending on which leg (i.e., paretic or nonparetic) is used as the leading leg.

# Electromyography

Impairments of muscle activation during APA were demonstrated by many authors in stroke patients compared to healthy subjects (3, 23, 26, 31). Sousa et al. (24) showed that when the paretic leg was used as the leading leg, the TA of the paretic leg presented only half of the EMG activity observed in healthy subjects. On the other hand, Ko et al. (26) showed that when gait was initiated with the non-paretic leg, the TA activation in the paretic leg was increased by 27 to 36% compared with the paretic leg as leading leg. Moreover, stroke patients exhibited a bilateral decrease in TA activity during GI-APA and had difficulties in reducing tonic activity in the soleus muscle of the paretic leg (3). Finally, Kirker et al. (31) showed that when stroke patients initiated gait with the paretic leg, GI-APA were characterized by an impaired activation of the gluteus medius muscle of the paretic leg, which was compensated by more prominent activity in the contralateral adductor.

# APA in Stroke Patients: Paretic vs. Non-paretic leg as the Leading or Trailing leg APA Pattern

Rajachandrakumar et al. (12) showed that the frequency of trials with no APA did not differ when gait was initiated with the paretic or non-paretic leg. In contrast, multiple APA were more prevalent when the non-paretic leg was used as the leading leg. Additionally, when stepping with the non-paretic leg, the frequency of trials with no APA was negatively correlated with motor recovery.

# APA Duration

Bensoussan et al. (5) showed that the postural phase was longer when the paretic leg was the trailing leg. In contrast, Hesse et al. (11) and Melzer et al. (28) found no significant difference of postural phase duration when initiating gait with the paretic or non-paretic leg.

# APA Amplitude in the Frontal Plane

Sharma et al. (22) showed that patients with left hemisphere lesions generated greater lateral ground reaction forces with the non-paretic left leg than with the paretic right leg. In contrast, medial ground reaction forces were equivalent for both the paretic and non-paretic leg. Additionally, Hesse et al. (11) reported that the amplitude of the mediolateral CoP shift during APA was lower when gait was initiated with the paretic leg. In


TABLE

3


assessment

analysis

according

to

the

modified

Downs

and

Black

scale.

contrast, the CoM did not move toward the trailing leg when patients started with their non-paretic leg, while the trajectory of the CoM was similar to that of healthy subjects when patients initiated gait with the paretic leg.

### APA Amplitude in the Sagittal Plane

Regarding APA in the sagittal plane, the same authors found that anticipatory anterior-posterior CoM acceleration during APA was lower when starting with the paretic leg (22). Tokuno and Eng (30) found that the anteroposterior impulse generated by the trailing leg was less than half that of healthy subjects. Additionally, Bensoussan et al. (4, 5) found that the ground reaction forces exerted by the paretic leg were directed backwards during APA (retropulsive action) but were directed forward for the non-paretic leg (propulsive action). Altogether, these findings indicate that the use of the paretic leg as the leading leg is associated with difficulties activating the TA and gluteus medius muscles during APA, which result in reduced CoP shift and CoM acceleration. In contrast, the use of the paretic leg as the trailing leg seems to challenge balance to a greater extent during GI.

# DISCUSSION

The aim of this systematic review was to provide an overview of the impact of stroke on the organization of APA associated with gait initiation (GI-APA). Altogether, the reviewed studies suggest that the anticipatory phase of GI is impaired in stroke patients. The findings principally highlight the changes in the pattern and spatiotemporal parameters of GI-APA in comparison with healthy subjects and the differences in initiating gait with the paretic or non-paretic leg. Moreover, this review may provide an evidenced-based theoretical framework for clinicians involved in rehabilitation programs for stroke patients.

The majority of the included studies were scored ≥ 14/28 according to the modified Downs and Black scale. The most common deficiencies in these included studies, in reference to the modified Downs and Black scale, were the lists of the principal confounders (question 5, i.e., not applicable for this review, only for interventional studies) and the adverse effects (question 8, i.e., not applicable for this review, only for interventional studies), the interventions not representative of usual care (question 13, i.e., not applicable for this review, only for interventional studies), the blinding of the patients or the assessors (questions 14 and 15, i.e., not applicable for this review, only for interventional studies), the specification of the time period of recruitment (question 22), the randomization of the subjects in groups (questions 23 and 24, i.e., not applicable for this review, only for randomized control trial studies) and the investigation of the main confounders (question 25, i.e., not applicable for this review, only for interventional studies). These absences were not surprising because the studies did not include intervention and control groups to measure the effects of treatment, but were observational studies comparing GI between subjects during a single session. Because of the absence of randomized controlled trials, a low quality level of evidence for the results may be provided (20). This means that further research with randomized protocols and more standardized protocols is very likely to have an important impact on our confidence in the results. The level of the quality assessment, according to the modified Downs and Black scale, must be considered with caution because there is no treatment group (i.e., interventional randomized controlled trials) in the included studies.

Martinez et al. (25) found that when stroke patients initiated gait, GI-APA were present in 100% of trials, whereas Rajachandrakumar et al. (12) reported the possibility of trials with no GI-APA. The subjects in these two studies were at different post-stroke times and the significant changes in motor recovery occurring in the weeks after stroke may partially explain the observed differences (32, 33). Rajachandrakumar et al. (12) found that trials with no GI-APA were more prevalent among patients who walked slowly, and that motor impairment was correlated with the frequency of trials with no GI-APA. Multiple GI-APA, the other atypical pattern described by the authors, was observed at a higher frequency when gait was initiated with the healthy leg.

The "psychological apprehension" loading of the paretic leg may partially explain longer GI-APA duration and may reflect the priority of the central nervous system to secure mediolateral stability before walking (4, 5, 12). A longer GI-APA duration can also be attributed to delays in soleus deactivation (23) and TA activation (27). These muscle impairments probably lead to difficulties in correctly timing the CoP and CoM shift during GI-APA compared to healthy subjects.

Lower GI-APA amplitude observed in both the frontal and sagittal planes may result in lower stability and lower progression velocity, respectively, during GI (11, 21–23, 28, 30). However, the hypothesis that the lower gait progression velocity observed in stroke patients results from GI-APA impairments requires further investigation (23). Indeed, it has been shown that the amplitude of COP displacement is similar between healthy and post-stroke participants when healthy participants are instructed to walk at a speed similar to that chosen by the post-stroke subjects (30). Thus, the lower GI-APA amplitude reported in stroke patients under spontaneous velocity conditions appears to result from their lower self-adopted speed compared to healthy participants. Furthermore, balance and performance during the execution phase is affected by the choice of the leading leg during the postural phase. GI-APA are dynamic phenomena that could impact balance (34–36) and an inability to support body weight during GI due to weakness in the paretic leg. Thus, lower GI-APA amplitude in stroke patients could result from postural muscle weakness or reflect adaptations to the lower balance state.

Many studies included in the present review confirm the findings of Pélissier et al. (37) who described a reduction in the duration of the swing phase, step length, walking velocity, and magnitude of the propulsion force during the execution phase when initiating gait with the healthy leg. These effects could be the result of decreased plantar flexor, hip flexor and hip extensor strength on the paretic side (38, 39), which may underlie the inability to support body weight. The incapacity to correctly propel the CoM above the paretic leg during the stance phase may explain the higher degree of instability observed when starting with the non-paretic leg (3–5, 11, 22, 30, 31). Additionally, the paretic leg muscles contribute less to the load of the paretic side, generate less power, and may even produce retropulsive forces, which may reduce walking velocity, because all of the propulsive forces are generated by the healthy leg alone (4, 5, 40).

Knowing the main characteristics of postural deficiencies during gait initiation or steady walking in stroke patients may help clinicians assess the efficiency of neurorehabilitation methods and provide an evidence-based theoretical framework for motor recovery. For example, the review of (37) established that the symmetry of stride was not correlated with walking performance (especially with the progression velocity). In addition, partial body support (i.e., on the non-paretic stance leg) was shown to have no effect on the anticipatory phase (21). These findings suggest that it is not necessary to obtain symmetrical static balance before starting gait rehabilitation. On the contrary, gait training should be started as early as possible after stroke. As documented in the literature reviewed above, initiating gait with the paretic leg improves stability and motor performance, so we recommend that rehabilitation programs promote this strategy in the early post-stroke period to improve patient mobility and autonomy.

Gamma et al. (21) recommend the use of an over-ground body weight support to improve mediolateral APA (i.e., stability) without changing the anteroposterior APA velocity (i.e., performance). However, in practice the combination of different rehabilitation strategies, notably the use of functional electrical stimulation and brain-computer interfaces, seems to be more effective than over-ground gait training alone (41).

Moreover, since using the paretic leg as the trailing leg has been found to increase the amplitude of TA muscle activity by between 27 and 36% (26), this strategy can be used as TA strength training to improve postural muscle performance in the paretic leg. Regarding possible gait-rehabilitation strategies, increasing TA muscle activity of the paretic leg during gait initiation seems to be an important target for motor rehabilitation, since bilateral TA activation during APA serves to provide initial propulsive

# REFERENCES


forces that shift the whole-body forward (3). It is also involved in body balance maintenance during step execution (10). It is clear that strength training should not be restricted to the TA of the paretic limb but should involve all paretic leg muscles. Specifically, reinforcing hip abductor muscles may facilitate the anticipatory CoP and CoM shifts in the frontal plane (31), while reinforcing hip extensors, knee extensors and plantar flexors should facilitate body weight support against gravity and provide efficient propulsive forces in the sagittal plane (5, 22, 28, 30).

In this way, the use of virtual reality training during gait initiation could help train and obtain recovery of force and power separately for each lower limb (i.e., paretic and non-paretic limb) in individuals with stroke (42).

# CONCLUSION

This systematic review provides an update on GI-APA reorganization following stroke. Stroke patients present atypical GI-APA patterns, longer GI-APA duration and lower GI-APA amplitude compared to healthy people, regardless of which leg is used as the leading or trailing leg. GI is facilitated when the non-paretic leg is used as the trailing leg because the weakness of the paretic leg leads to difficulties in supporting body weight during the upcoming stance phase. Further experiments should include distinct groups of patients in order to describe GI-APA features in acute, subacute and chronic stroke, and the influence of spasticity and of the lesion site. Understanding the changes in each population could be relevant for personalizing rehabilitation strategies.

# AUTHOR CONTRIBUTIONS

AD wrote the manuscript. TV and TH performed the bibliographical search and selected the articles to be included in the review. EY, J-LH, and SC reviewed 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 Delafontaine, Vialleron, Hussein, Yiou, Honeine and Colnaghi. 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.

# Walking Along Curved Trajectories. Changes With Age and Parkinson's Disease. Hints to Rehabilitation

Marco Godi <sup>1</sup> \*, Marica Giardini <sup>1</sup> and Marco Schieppati 2†

*<sup>1</sup> Division of Physical Medicine and Rehabilitation, ICS Maugeri SPA SB, Pavia, Italy, <sup>2</sup> Department of Exercise and Sport Science, International University of Health, Exercise and Sports, LUNEX University, Differdange, Luxembourg*

#### Edited by:

*Helena Blumen, Albert Einstein College of Medicine, United States*

#### Reviewed by:

*Gammon Earhart, Washington University in St. Louis, United States Elisabetta Farina, Fondazione Don Carlo Gnocchi Onlus (IRCCS), Italy*

> \*Correspondence: *Marco Godi marco.godi@icsmaugeri.it*

#### †Present Address:

*Marco Schieppati, ICS Maugeri SPA SB, Pavia, Italy*

#### Specialty section:

*This article was submitted to Neurorehabilitation, a section of the journal Frontiers in Neurology*

Received: *08 February 2019* Accepted: *03 May 2019* Published: *24 May 2019*

#### Citation:

*Godi M, Giardini M and Schieppati M (2019) Walking Along Curved Trajectories. Changes With Age and Parkinson's Disease. Hints to Rehabilitation. Front. Neurol. 10:532. doi: 10.3389/fneur.2019.00532* In this review, we briefly recall the fundamental processes allowing us to change locomotion trajectory and keep walking along a curved path and provide a review of contemporary literature on turning in older adults and people with Parkinson's Disease (PD). The first part briefly summarizes the way the body exploits the physical laws to produce a curved walking trajectory. Then, the changes in muscle and brain activation underpinning this task, and the promoting role of proprioception, are briefly considered. Another section is devoted to the gait changes occurring in curved walking and steering with aging. Further, freezing during turning and rehabilitation of curved walking in patients with PD is mentioned in the last part. Obviously, as the research on body steering while walking or turning has boomed in the last 10 years, the relevant critical issues have been tackled and ways to improve this locomotor task proposed. Rationale and evidences for successful training procedures are available, to potentially reduce the risk of falling in both older adults and patients with PD. A better understanding of the pathophysiology of steering, of the subtle but vital interaction between posture, balance, and progression along non-linear trajectories, and of the residual motor learning capacities in these cohorts may provide solid bases for new rehabilitative approaches.

Keywords: curved walking, aging, Parkinson's disease, freezing of gait, curved walking rehabilitation

# INTRODUCTION

Every day we frequently follow circular courses and turns when we move (1). Our inherently unstable bipedal gait (2) requires adaptive control during curved walking (3). Stride length and duration of the stance phase differ between the inner and outer leg (4–6), and the center of mass moves toward the interior of the trajectory (7). This is made possible by appropriate placement of the feet to create a gravity-dependent torque that counteracts the centrifugal force. Central commands would take over the control of locomotion at the expense of the spinal automatisms, very much as it happens for crayfish (8) and giant stick insects, where rotation of the body is accomplished by modulation of ground reaction forces among legs (9). Obviously, the brain generates ad-hoc activity that follows the basic laws of physics when moving along circular trajectories, regardless of the number of legs (10).

A thoughtful study of walking and turning in humans had been done more than 140 years ago by Eadweard Muybridge [see (11)]. Foot yaw orientation, body tilt in the frontal plane and pelvis rotation over the stance leg could already be seen in his remarkable photographic freeze frames.

Then, non-linear walking trajectories had been almost ignored by scholars and scientist, until Takei et al. published a note (12) about the walking trajectory of young and older subjects on a circular path in darkness.

The interaction between the spinal circuits with the descending command and the adaptation of the body mediolateral inclination and step length asymmetry to the progression velocity would represent a challenge for the control system. This task requires anticipatory adjustments (13–19) in advance of taking the curve (5, 20–22). It is not implausible that basal ganglia are subjected to an extra load during curved walking, since they operate through two parallel distinct pathways having opposing effects, allowing the execution of movement while gating the antigravity activity of the postural muscles (14, 23). This short review aims to summarize the characteristics of curved walking in young and older adults, and to address gait differences between people with Parkinson's Disease (PD) and healthy subjects. Proposals for an integrated rehabilitation approach in PD are put forward.

# STEERING THE BODY ALONG CURVED PATHS

Successful locomotion along curved trajectories requires fine coordination of body segments' movements. Any minor change in the adjustment of the asymmetric step length of the two legs produces dramatic effects in the kinematics, very much as it occurs with minimal changes in the localization of ground reaction forces underneath the feet during the stance phases (6). Therefore, accurate brain control is required for correct rotation of the lower limbs and inversion or eversion of the ankle for successful placement of the foot on the ground (4, 24–31).

During linear walking, the feet create mediolateral impulses at heel strike that symmetrically move the body toward the contralateral limb. In turning, the body is moved toward the interior of the trajectory by both the internal and the external foot (32). In order to exploit gravity, the center of foot pressure during heel strike and toe off is being slightly displaced with respect to its position under linear walking. This creates a mediolateral torque that produces and controls trunk roll tilt and progression along the circular trajectory (6, 33–35), and generates the proper centripetal force to avoid going off on the tangent (4, 5, 27). Appropriate braking of the body fall toward the interior of the trajectory is exerted by the feet at foot-off (6, 36), such as to counterbalance the reaction forces produced at heel strike (37).

In addition to studying walking along circular trajectories, investigators have focused also on the strategies used to navigate sharp turns (26, 30, 38–40). Others have exploited figures of eight trajectories that include both clockwise and counter-clockwise segments (3, 41–43). Consistently, turns imply reduction of progression velocity, placement of the foot in the direction of the new trajectory (25) and lateral translation of the body in addition to body reorientation to align with the new travel direction (15). Foot placement and trunk inclination move the center of mass toward the new direction in the transition stride. The top-down temporal sequence in body segments reorientation slightly changes as turn proceeds (26, 44). This pattern is robust to turning velocity, therefore inherent in the command to turn (45). Gaze redirection accompanies steering, so that visual or oculomotor deficits should be considered when assessing turning behavior (24, 46).

# WHAT DO WE KNOW OF THE NEURAL COMMAND FOR STEERING?

Motor tasks employ muscle synergies, i.e., one or more sets of muscles synchronously activated and specific to the task (47). Synergy studies concluded that rectilinear and curvilinear walking share the same motor command; however, fine-tuning in muscle synergies is necessary for circular trajectories, where the kinematic strategy conforms to the physical laws that underpin curved walking while keeping balance (14, 27, 48–50). Courtine and Schieppati (28), using the principal component analysis, found that both straight-ahead and curved walking were low dimensional, and two components accounted for more than 70% of the movement variability. Fine modulation of the muscle synergies underlying the straight-ahead locomotion is sufficient for generating the adequate propulsive forces to steer walking and maintain balance (48).

Bejarano et al. (49) found four muscle synergies for both walking conditions. Muscle activation profiles lasted longer and were larger during curvilinear than straight walking, and more so for the muscles of the limb inside than outside the trajectory. However, several deep muscles responsible for intraand extra-rotation of pelvis on thigh had not been recorded. The asymmetric activation of these muscles and the amplitude and time-course of the modulation of their activity might configure an additional synergy peculiar to turning. The contribution of the gluteus medius to the trunk orientation during turning should be also considered (51).

The origin of the adaptation of the motor command to the curved path is a matter of speculation. Jahn et al. (52) have described brain activation for imagined straight walking and for imagined walking along a curved path (53). They observed asymmetric basal ganglia activation at turn initiation, enhanced activity in cortical areas associated with navigation, and decreased activity in areas supposed to process vestibular input. These findings point to the complexity of the organization of the command for producing the curved walking trajectory, while the deactivation of certain brain regions may explain why the vestibular input seems to be down regulated during continuous turning while stepping in place (54), similarly to what occurs at the transition between stance and gait (55).

# SENSORY FEEDBACK DURING WALKING AND TURNING

Asymmetric proprioceptive input elicited by vibration of axial (neck and trunk) muscles produces steering and turning (29, 56–59), whereas proprioceptive input from the leg contributes to fine adjustment of the spinal pattern generators for walking (60, 61). Input from axial muscles

FIGURE 1 | (A) The effect of the trajectory (curved with respect to linear) on spatiotemporal gait variables in older adults. Data are the sample-size-weighted mean of Cohen's d effect-size (ES) of the illustrated variables [calculated from <sup>a</sup> (82); <sup>b</sup>(41); <sup>c</sup> (83); <sup>d</sup>(42); <sup>e</sup> (87); <sup>f</sup> (86); <sup>g</sup> (88); <sup>h</sup> (85); <sup>i</sup> (84)]. Negative values in the x-axis represent a decrease in the variables in curved compared to linear walking. Error bars represent 95% confidence intervals. There is an overall decrease in step length, cadence, and step width during protracted curved path. On the contrary, during sharp turns, step width is increased in the older adults. (B) The effect of Parkinson's disease on walking along curved trajectories. Data are the sample-size-weighted mean of Cohen's d effect-size of the illustrated spatiotemporal gait variables [calculated from a (89); <sup>b</sup>(82, 90) <sup>d</sup>(91); <sup>e</sup> (92); <sup>f</sup> (85)]. Negative values in the x axis represent a decrease in the variable values in patients with PD compared to age-matched controls. Error bars represent 95% confidence intervals. Step length and width decrease, while cadence is unaffected. Velocity is diminished mainly because of reduction in step length.

would play the role of a servo-mechanism, whereby minor asymmetries initiated by asymmetric foot placement (1, 62) would affect the spinal generators to produce the necessary fine changes in leg and foot kinematics accompanying heading changes.

Whether or not continuous walking along a circular trajectory is also favored by a shift in our straight-ahead goes beyond the scope of this short review, but we would note that a shift in subjective straight-ahead occurs after a period of stepping in place on a rotating treadmill (63). In turn, it is not unlikely that a shift in the straight-ahead is produced by the feedback from the muscles producing the rotation of the pelvis and trunk over the standing leg when walking along a curved trajectory or when stepping in place and turning (54, 59). Vision is obviously not necessary for implementing a curved trajectory (5), but the continuous visual field motion would nonetheless favor the fine tuning of the gait synergies underpinning the production of the circular trajectory (24, 64).

These findings suggest that asymmetric proprioceptive input, either produced by the asymmetric kinematics initially produced by the central command to turn or by the artificial activation through muscle vibration, would favor and sustain the steering synergy.

# AGING AFFECTS STEERING

Locomotor impairments are an inevitable consequence of aging, and worsen with the associated cognitive decline (65, 66). However, we did not expand here on the issue of the effects of cognitive decline on locomotion. This was a deliberate choice, because this would require an ad-hoc article, given the growing number of papers on this complex topic [see the reviews by (67–69)]. Furthermore, the effect of cognitive decline on turning has not received the attention it would deserve, yet.

Walking speed is definitely lower than in the young (70–72), and older adults adopt a more cautious attitude when steering (73–75). Normally, during the swing period, young subjects reverse the fall of the center of mass before foot-contact by active braking via activation of the triceps of the stance leg (76, 77). The control of this braking phase is impaired in older adults, and the braking phase is compensated for by reducing the step length (78). Neurological (PD, cerebellar syndromes, neuropathies, hemiparesis, dementia) and non-neurological conditions (cardiovascular and respiratory) contribute to gait problems (79, 80).

In young adults, curved walking significantly decreases walking speed (by about 15%) and stride-length (more so in the leg inner to curvature), whilst cadence is barely diminished (5, 27). In older adults whose linear gait speed is within the limits of normality (81), curved gait speed diminishes by about 20%, with minor differences across the studied cohorts (42, 82–86) (**Figure 1A**). In older adults with poor mobility and linear walking speed below the 0.90 m/s, the reduction in gait speed between linear and curved walking is of about 15% (41) and 5% (87). Likely, frailty and balance-related anxiety (93) reduces speed during linear walking to such low values that the time necessary for the added coordination of posture and progression during curved walking becomes proportionally negligible.

Cadence diminishes by <10% (84–86) for older adults during curved when compared to linear walking. Step length is also reduced by <20% (82, 84). While velocity, cadence, and step length changes are common to different types of turning, step width behaves differently between curved walking and sharp turning. Older adults increase their step width during the transition phase from linear to curved path (88). On the contrary, for protracted curved path, step width is narrow (89, 94) and it decreases until 30% (84). Slow anticipatory adjustments may play a role, as shown by abnormal turning pattern in older adults with balance deficits when the command to turn is unexpected (95).

The variability of step length, cadence, and step width increases during curved walking (84, 89, 91, 96). However, in older people who have not fallen, a moderate amount of step variability is required to adapt to situations that challenge postural control (97, 98), while too little variability is associated with fall history in older adults (99). This could explain why increased spatial variability during curved path identified subjects with better motor skills of walking (86).

Additionally, older adults who frequently fall display a reduced turning angle variability compared with non-fallers (100), denoting a lack of dynamic balance skills necessary to seamlessly modulate turning angles while maintaining balance (101). It seems that the ability to vary step length and step width enables both smooth continuity of the center of pressure path and energy-efficient navigation of curves (86).

The multisegmental control has been studied during planned turning on the spot (102–104). Older adults tend to reorient their head, shoulder and pelvis simultaneously, followed by foot displacement, regardless of visual condition. The command to implement curved walking implies modification of all the fundamental spatio-temporal variables of gait (105), more so in older adults who frequently fall compared with non-fallers (100). Interestingly, multiple fallers show a simplified turning pattern to assist balance control (106). Moreover, changes in timing and sequencing of segment reorientation produce earlier anticipation of turns (102, 107). Activity in prefrontal cortex is increased in older compared to young adults (108) suggesting a higher cognitive cost for gait control even during linear walking. In complex walking, known to depend on appropriate executive function and balance control (109), obstacle negotiation further increases prefrontal activity (110).

These studies point to definite changes in the spatio-temporal gait variables between linear and curved walking. The increased variability in the curved gait kinematics can be the result of a coordination disorder but represents kind of a safety factor for reducing the risk of falling. Adaptive strategies for turning might be driven by the effort to diminish the cognitive cost associated with turning.

# PARKINSON'S DISEASE PROBLEMS DURING STEERING

In patients with neurological disorders, impaired locomotion is a frequent and serious symptom (111). Gait and balance impairments are typical of patients with PD (112, 113). Impaired locomotion is often associated with reduced ability to brake the fall of the body in late stance compared to matched controls (114). This may be a sign of poor coordination between trunk and lower limb, and forces patients to take short steps. A study of muscle synergies has identified a reduced number of synergies during straight walking (115), providing evidence that control of gait is less complex or flexible in patients with PD. Whether this can affect the coordination between trunk and lower limb muscle, and whether synergies are further affected by turning, remains to be established (115).

In neurological patients, additional problems emerge during curved compared to linear trajectories (116–118). Patients with PD are no exception, and all of them are indeed critically challenged by curved walking (119–123), in spite of their diverse phenotype and motor symptomatology. There is a strong indication, based on functional MRI, that patients with PD have reduced activity in the globus pallidus and enhanced activity in the supplementary motor area during imagined walking and turning (124). Functional near infrared spectroscopy shows higher prefrontal cortex activity in the patients compared to healthy subjects (125). Interestingly, prefrontal activation during obstacle negotiation was increased more than during dual task walking (126).

The changes observed in curved compared to linear walking are usually non-negligible. Kinematic analysis demonstrated enbloc rotation of axial segments in patients with PD (127–131). Coordinated axial muscle activation seems to be particularly affected (132). This has been repeatedly shown and assessed quantitatively (133–138). The en-bloc rotation of axial segments in patients with PD contrasts with the lower extremity muscle activation pattern that appears to be overall normal (131). Huxham et al. (139) have shown that stride length reduction appears to contribute more than downscaled rotation amplitude to inefficient turning in patients with PD, possibly because of reduced axial mobility. Compared to straight walking, during curved walking speed diminishes by about 20% (90) and by about 35% (82, 91) in less and more affected patients, respectively. Cadence also diminishes by more than 5% and step length diminishes by about 20% (82, 89) (**Figure 1B**). In contrast to age-matched controls, patients with PD turn with narrower steps (89, 92, 122).

Greater variability of the temporal gait parameters is detectable during curved but not linear walking, over and above the increased variability exhibited by healthy subjects (6). Problems in turning and curved walking are detectable even at an early stage of the disease, and they persist while on-phase (92, 140). Most interestingly, spatio-temporal gait pattern and variability during curved walking are abnormal even in welltreated, well-functioning patients with PD, exhibiting no change in speed in straight-line walking compared to age-matched healthy subjects (85). Anxiety should be considered when evaluating walking performances, since well-treated patients with fear of falling spend more time turning, both in the laboratory and at home (141). This would be connected to the effect of temporal pressure on the control of medio-lateral stability, a critical issue when changing direction (142, 143).

All in all, it seems safe to posit that patients with PD suffer from specific disorders of curved walking and turning, which may not be obvious during straight walking. Again, it may not be immediately clear whether some abnormalities are direct consequences of the neuronal damage or, at least in part, the outcome of a long-duration adaptive process.

# FREEZING IN PEOPLE WITH PARKINSON'S DISEASE

Freezing can occur during all types of gait and is related to the increased stride-to-stride variability (144). However, it is most common during turning (22, 145–153) and is a major cause of falling in these patients (154). Turning during daily activities is more compromised in patients with than without freezing (155).

The increase in freezing events during turning may be due in part to the asymmetric nature of the task and the necessary anticipatory adjustment for ensuring postural stability along the medio-lateral direction. The temporal and spatial asymmetry of steps during turning represents a more complex control problem than forward walking (30, 121, 149, 156, 157), as suggested by increased activation in prefrontal areas accompanying freezing before anticipated turns (125). Perhaps, these changes accompany and compensate for the structural and functional alterations in the brain stem centers for locomotion (158).

Neck and axial rigidity during turning may reduce forward progression (159, 160). Patients take shorter turns with smaller turn angles and more steps and exhibit larger variability with respect to controls (161). Freezers, irrespective of freezing episodes, adopt a narrower step width compared to controls and non-freezers during turning (89, 91).

Problems in fast axial turning appear when stepping is performed on a narrow base (160, 162, 163). Freezing episodes are more frequent at sharp turns (91) and turning in place (164), indicating that problems exist both in adjusting the anticipation to the intended trajectory and in controlling body segment coordination and balance during rotation on the spot, regardless of the speed of turning and the severity of the disease (22, 94). Again, this speaks in favor of a delayed preparation for the change in walking direction.

Anticipatory adjustments are indeed abnormal in patients with PD (165–167) including eye and head anticipatory movements for exploration (53) and movements to correct a lateral disequilibrium (168, 169). Plate et al. (170) have confirmed that anticipatory adjustments are slower and of smaller amplitude, in keeping with the overall bradykinesia of the patients. Interestingly, anticipatory postural adjustments are not followed by coordinated steps (171). Anticipatory adjustments associated with gait initiation may not be strictly abnormal per se in patients with respect to age-matched subjects (169, 172), but may be impaired in people with freezing of gait compared to those without freezing (173). All in all, these findings are in keeping with the hypothesis that freezing would depend on the sheer control of the trunk rotation over the standing leg (54, 85).

Freezing is elusive. However, any voluntary human movement requires a complex array of in-series and in-parallel processes, from anticipatory postural adjustment to ongoing feedbackrelated correction. No wonder some peculiar interaction of aberrant events in patients with PD, from cognitive to reflex nature, can finally produce what we call freezing, in our case of gait, but not necessarily limited to gait [see e.g., (174)].

# REHABILITATION OF CURVED WALKING

The ability to turn, above all in restricted spaces, is very important in autonomy maintenance in everyday life. Furthermore, falls during turning result in more hip fractures than falls during linear gait (175). Therefore, rehabilitation of curved walking and turning may be meaningful since turning impairments are not improved by dopaminergic medication (92). Exercise has potential to improve many clinical issues in patients with PD such as strength, balance, walking, and quality of life (176–179). A recent review, not centered on curved walking (180), confirmed the benefits of physiotherapy in most outcomes over the short term. However, most of the observed differences between treatments were considered of minimal clinical importance.

Rehabilitation should address the critical steps in producing successful steering, the anticipatory adjustments in preparation to turn and direct steps along the curved trajectory, the rhythmic and regular production of steps, axial mobility and headtrunk coordination, and the coordinated activity of the pelvic muscles producing intra- and extra-rotation of the lower limbs. Anticipatory postural adjustments can be possibly enhanced by focused appropriate rehabilitation in older adults (181). This is significant in consideration of the relevance of the delayed release of anticipatory adjustments in these patients (92, 167).

The use of linear treadmill for the rehabilitation of gait in patients with PD has produced modest but significant effects on gait speed and stride length (182–185), indicating that these patients can be trained on treadmill and show some improvement. An earlier study with the use of the rotating treadmill in two patients showed that a short period of walking on the rotating treadmill reduces freezing episodes (186, 187). However, training curved walking by a rotating treadmill produced no improvements in gait or turning after 5 days of training (90). A later study, administering a progressive training with augmenting rotation velocity and trial duration through 10 daily sessions, showed instead a significant benefit on walking velocity along a circular path (188). In the latter study, before and at the end of the treatment, all patients walked over ground along linear and circular trajectories, and the velocity of walking bouts increased post-treatment, more so for the circular than the linear trajectory. Therefore, in spite of their known problems in attentional resources and cognitive strategies (189), patients can learn to produce turning while stepping on the rotating treadmill and this capacity translates into improved over ground curved walking. These findings strengthen the conclusions of independent reports by Cheng et al. (190, 191), employing a complex walking program (S-shaped, figure-of-8, square- and oval-shaped paths) or a rotating platform.

Notably, the synergies responsible for maintaining a fixed body orientation in space while stepping on the rotating treadmill (45) are the same that are put in place when stepping in place while voluntarily turning (54). Prolonged stepping in place and turning produces in normal subjects an after-effect consisting in a long lasting spontaneous turning on the spot (eyes closed), likely created by adaptation to the continuously activated somatosensory channel (54, 59, 192). Time decay of angular velocity, stepping cadence, and head acceleration were remarkably similar after both conditioning procedures [voluntarily stepping-and-turning and the so-called podokinetic stimulation by the rotating treadmill (193)]. Therefore, stepping in place while voluntarily turning could be administered in lieu of the rotating platform for training the turning task in patients with PD. This treatment could be as successful as the rotating platform for the purpose of exercising the coordinated movements underpinning turning and of training curved walking. In this connection, it is worthwhile to mention that Aman et al. (194) have shown the effectiveness of proprioceptive training in improving sensorimotor function, and Elangovan et al. (195) proved that proprioceptive training is indeed effective in patients with PD.

The responses of patients with PD to vibration of postural muscles are largely normal, even if they show abnormal transient postural responses to vibration-off, a sign of impaired sensory reweighting in balance control (196). This is reminiscent of their inappropriate response to light-off when balancing on a movable platform (197). Similarly, alternate trains of postural muscle vibration promote cyclic body displacement in standing patients much as in age-matched subjects (198), showing that these patients can integrate and exploit the vibratory proprioceptive input to produce postural oscillations comparable to those occurring during walking. Moreover, vibratory stimulation of trunk muscles significantly increases stride length, cadence and velocity in both patients and healthy subjects (199). As far as curved walking is concerned, it is notable that, in healthy subjects, vibration of trunk muscles interferes with the above mentioned podokinetic aftereffect by enhancing or reducing—body rotation velocity depending on the vibrated side (59). The summation of vibration and podokinetic effect speaks for the capacity of the proprioceptive input from the trunk and from the pelvis muscles to affect steering by modulating the activity of the responsible brain centers through a common mechanism. These interventions (unilateral axial muscle vibration, stepping on the rotating treadmill, voluntarily stepping-and-turning) might be considered when planning a training protocol aimed at rehabilitating gait with emphasis on curved walking.

Overall, it seems that interventions based on the new evidence about planning, organization, and execution of curved

# REFERENCES


walking and associated postural control represent a promising rehabilitation approach. We are now ready for undertaking largescale studies that address the effects of sensory feedback and of stepping-and-turning task repetition on steering and turning capacities in older subjects and patients with PD, also considering patients with typical PD and atypical parkinsonism [see (200)].

# CONCLUSIONS

The findings summarized here suggest that the turn-related command operates by fine modulation of the phase relationships between the tightly coupled neuronal assemblies that drive motor neuron activity during walking. Aging and Parkinson's disease seem to affect steering by slowing the expression of this modulation and simplifying the neural command by coupling subroutines. Older adults and more so patients with PD are compelled to modify the gait pattern, reduce some spatiotemporal variables when facing curvilinear trajectories, and adopt en-bloc rotation of axial segments. This strategy allows the CNS to manage a less complex task, even if it does not fully protect these subjects from the risk of falling, likely because of persisting poor posturo-kinetic coordination. The cost of the new control mode, implying a newly designed integration of vestibular, proprioceptive, and visual information for equilibrium control during curved walking, turns out to be substantial in patients with PD, and would lead to freezing and frequent falls. More research is needed on rehabilitative interventions to train voluntary or treadmill-induced body rotation and on their potential of improving walking performance during curvilinear paths.

# AUTHOR CONTRIBUTIONS

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

# FUNDING

This work was partially supported by the Ricerca Corrente, funding scheme of the Ministry of Health, Italy.


Parkinson's disease with and without freezing of gait. Gait Posture. (2016) 43:54–9. doi: 10.1016/j.gaitpost.2015.10.021


turning performance in individuals with Parkinson's disease: a randomized controlled trial. Sci Rep. (2016) 6:33242. doi: 10.1038/srep33242


**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 Godi, Giardini and Schieppati. 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.

# Long-Term Effects of Whole-Body Vibration on Human Gait: A Systematic Review and Meta-Analysis

Matthieu Fischer 1,2, Thomas Vialleron1,2, Guillaume Laffaye1,2, Paul Fourcade1,2 , Tarek Hussein<sup>3</sup> , Laurence Chèze<sup>4</sup> , Paul-André Deleu<sup>4</sup> , Jean-Louis Honeine<sup>5</sup> , Eric Yiou1,2 and Arnaud Delafontaine1,2 \*

<sup>1</sup> CIAMS, Université Paris-Sud, Université Paris-Saclay, Orsay, France, <sup>2</sup> CIAMS, Université d'Orléans, Orléans, France, <sup>3</sup> ENKRE, Saint-Maurice, France, <sup>4</sup> LBMC, Université de Lyon, Lyon, France, <sup>5</sup> VEDECOM, Versailles, France

Background: Whole-body vibration is commonly used in physical medicine and neuro-rehabilitation as a clinical prevention and rehabilitation tool. The goal of this systematic review is to assess the long-term effects of whole-body vibration training on gait in different populations of patients.

#### Edited by:

Thomas Platz, University of Greifswald, Germany

#### Reviewed by:

Cristiano De Marchis, Roma Tre University, Italy Alessandro Giustini, Consultant, Arezzo, Italy

\*Correspondence: Arnaud Delafontaine arnaud.delafontaine@u-psud.fr

#### Specialty section:

This article was submitted to Neurorehabilitation, a section of the journal Frontiers in Neurology

Received: 24 January 2019 Accepted: 28 May 2019 Published: 19 June 2019

#### Citation:

Fischer M, Vialleron T, Laffaye G, Fourcade P, Hussein T, Chèze L, Deleu P-A, Honeine J-L, Yiou E and Delafontaine A (2019) Long-Term Effects of Whole-Body Vibration on Human Gait: A Systematic Review and Meta-Analysis. Front. Neurol. 10:627. doi: 10.3389/fneur.2019.00627 Methods: We conducted a literature search in PubMed, Science Direct, Springer, Sage and in study references for articles published prior to 7 December 2018. We used the keywords "vibration," "gait" and "walk" in combination with their Medical Subject Headings (MeSH) terms. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology was used. Only randomized controlled trials (RCT) published in English peer-reviewed journals were included. All patient categories were selected. The duration of Whole-Body Vibration (WBV) training had to be at least 4 weeks. The outcomes accepted could be clinical or biomechanical analysis. The selection procedure was conducted by two rehabilitation experts and disagreements were resolved by a third expert. Descriptive data regarding subjects, interventions, types of vibration, training parameters and main results on gait variables were collected and summarized in a descriptive table. The quality of selected studies was assessed using the PEDro scale. Statistical analysis was conducted to evaluate intergroup differences and changes after the WBV intervention compared to the pre-intervention status. The level of evidence was determined based on the results of meta-analysis (effect size), statistical heterogeneity (I 2 ) and methodological quality (PEDro scale).

Results: A total of 859 studies were initially identified through databases with 46 articles meeting all of the inclusion criteria and thus selected for qualitative assessment. Twenty-five studies were included in meta-analysis for quantitative synthesis. In elderly subjects, small but significant improvements in the TUG test (SMD = −0.18; 95% CI: −0.32, −0.04) and the 10MWT (SMD = −0.28; 95% CI: −0.56, −0.01) were found in the WBV groups with a strong level of evidence (I <sup>2</sup> = 7%, p = 0.38 and I <sup>2</sup> = 22%, p = 0.28, respectively; PEDro scores ≥5/10). However, WBV failed to improve the 6MWT (SMD = 0.37; 95% CI: −0.03, 0.78) and the Tinetti gait scores (SMD = 0.04; 95% CI: −0.23, 0.31)

**160**

in older adults. In stroke patients, significant improvement in the 6MWT (SMD = 0.33; 95% CI: 0.06, 0.59) was found after WBV interventions, with a strong level of evidence (I <sup>2</sup> = 0%, p = 0.58; PEDro score ≥5/10). On the other hand, there was no significant change in the TUG test despite a tendency toward improvement (SMD = −0.29; 95% CI: −0.60, 0.01). Results were inconsistent in COPD patients (I <sup>2</sup> = 66%, p = 0.03), leading to a conflicting level of evidence despite a significant improvement with a large effect size (SMD = 0.92; 95% CI: 0.32, 1.51) after WBV treatment. Similarly, the heterogeneous results in the TUG test (I <sup>2</sup> = 97%, p < 0.00001) in patients with knee osteoarthrosis make it impossible to draw a conclusion. Still, adding WBV treatment was effective in significantly improving the 6 MWT (SMD = 1.28; 95% CI: 0.57, 1.99), with a strong level of evidence (I <sup>2</sup> = 64%, p = 0.06; PEDro score ≥5/10). As in stroke, WBV failed to improve the results of the TUG test in multiple sclerosis patients (SMD = −0.11; 95% CI: −0.64, 0.43). Other outcomes presented moderate or even limited levels of evidence due to the lack of data in some studies or because only one RCT was identified in the review.

Conclusions: WBV training can be effective for improving balance and gait speed in the elderly. The intervention is also effective in improving walking performance following stroke and in patients with knee osteoarthrosis. However, no effect was found on gait quality in the elderly or on balance in stroke and multiple sclerosis patients. The results are too heterogenous in COPD to conclude on the effect of the treatment. The results must be taken with caution due to the lack of data in some studies and the methodological heterogeneity in the interventions. Further research is needed to explore the possibility of establishing a standardized protocol targeting gait ability in a wide range of populations.

Keywords: whole-body vibration, long-term effects, gait, biomechanics, randomized controlled trials, metaanalysis

# HIGHLIGHTS


# INTRODUCTION

Whole-body vibration (WBV) is a therapeutic method that exposes the entire body to mechanical oscillations while the patient stands or sits on a vibrating platform. This method was first used in the late nineteenth century by Charcot to treat gait disorders in neurological patients, especially in patients with Parkinson's disease (1). It is now commonly used in the physical medicine/neuro-rehabilitation fields as a prevention and rehabilitation tool for sarcopenia (2), osteoporosis (3), chronic low back pain (4), and fibromyalgia (5), among other conditions. WBV is also used in rehabilitation to improve muscle function (strength, power, and endurance) (6), muscle soreness (7), joint stability (8) and to reduce the risk of falling (9).

Several spinal and supraspinal mechanisms have been proposed to explain increased muscle activity during exposure to WBV. While there is currently no consensus, the most frequently cited mechanism is a reflex muscular contraction called tonic vibration reflex (TVR). This phenomenon has been shown to occur during direct and indirect vibratory musculo-tendinous stimulations that excite muscle spindles and enhance activation of Ia afferents, resulting in a higher recruitment of motor units and gradual development of muscle activity (10). In addition to these spinal reflexes, neuromuscular changes (11, 12), increased intramuscular temperature (10) and peripheral blood flow (13) may contribute at different levels to the increased muscular performance observed after WBV.

A recent review (14) reported a beneficial effect of long-term WBV training on balance control under static postural conditions. Since the literature appears to suggest a neuroanatomic (15) and a biomechanical continuum between standing posture and gait (16–18), Rogan et al. suggested that this beneficial effect could be extended to dynamic motor tasks such as gait (14). Such a continuum has been analyzed in stroke patients (19), for example. The most recent literature review focusing on the effect of WBV on gait, however, provided

only mitigated support for this assumption (20). Based on the screening of 10 randomized controlled trials (RCT), Lindberg and Carlsson concluded there was low-quality evidence for the beneficial use of long-term WBV on gait, and acknowledged there were major limitations (20), the most important being that only one of the authors reviewed the literature. Thus, no group discussions were conducted with experts to resolve possible disagreements and reach a mutual consensus. In addition, the low number of RCT included (n = 10) and the absence of meta-analysis may have limited the relevance of Lindberg and Carlsson's review. Since that review was published, WBV training has been used increasingly in physiotherapy to prevent and/or treat gait disorders. Consequently, more and more experimental studies have been conducted in this area with both healthy and pathological participants.

Hence, the aim of this article is to provide an up-to-date literature review of RCT studies on the effects of long-term WBV training on gait in both healthy subjects and pathological patients. It will contribute to provide evidence-based practice for a promising non-pharmacological rehabilitative method that is both safe and cheap, and that can be used by patients at home as part of an auto-rehabilitation program.

# MATERIALS AND METHODS

# Design and Literature Screening

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology was employed in this systematic review (**Figure 1**).

The PubMed, Science Direct, Springer and Sage databases were used for a comprehensive systematic literature search for articles published prior to 7 December 2018 with no time limit. The keywords used were: "vibration" AND (gait OR walk). More specifically, the search details specified in PubMed were: ("vibration"[MeSH Terms] OR "vibration"[All Fields]) AND (("gait"[MeSH Terms] OR "gait"[All Fields]) OR ("walking"[MeSH Terms] OR "walking"[All Fields] OR "walk"[All Fields])).

The selection procedure was conducted by two experts in rehabilitation. Disagreements were discussed with a third expert in a group until a mutual consensus was reached. First, a review was performed on all available titles obtained from the literature search with the selected keywords. All relevant or potentially relevant titles were included in the subsequent phase. Then, the abstracts were reviewed with all relevant or potential articles included in the following phase. Finally, full-text articles were reviewed to ensure that only relevant studies were included. In the same way, reference lists of all included articles were reviewed to possibly include articles through cross-referencing.

# Inclusion and Exclusion Criteria

To be included, the studies had to meet all of the following inclusion criteria: all patient categories were selected if: gait ability was measured before and after at least 4 weeks of WBV training performed on a vibration platform; the results were based on biomechanical analyses or were clinically relevant; the control group had no intervention or performed the same physical rehabilitation, resistance, balance or endurance training as the intervention group. In addition, only RCT, articles in English, and articles published in peer-reviewed journals were included. Studies were excluded if they measured only short-term effects (< 4 weeks) and if WBV was combined with non-physical training or with any intervention not provided to the control group (i.e., not only WBV effects are measured).

# Data Extraction and Main Measurements Examined

Data were extracted from the selected articles by one of the authors. The extracted data were checked by another author and disagreements were resolved with a third.

The following data were extracted for each selected article: (1) the names of the authors and the date of publication; (2) the number of subjects involved in the experiment with their characteristics and breakdown in each group; (3) WBV training details (in the following order: name of the WBV device, duration of the intervention, number of sessions, types of exercises, number of vibration sets, exposure duration per set, rest period between sets, frequency, amplitude and type of vibration) and control group details; and (4) the main outcomes related to gait with the main results (e.g., timed up-and-go test, 6-min walk test, walking speed, etc). When information could not be provided, it was indicated by a "?".

# Quality and Risk of Bias Assessment

The PEDro scale was used to assess the risk of bias, and thus the methodological quality of the selected studies (21). The scale was chosen for its ability to provide an overview of the external (criterion 1), internal (criteria 2–9) and statistical (criteria 9 and 10) validity of RCT. The scale is divided in 11 criteria, but the first criterion is not calculated in the total score. The output of each criterion could be either "yes" (y), "no" (n) or "do not know" (?). A "y" was given a score of one point, while a "n" or "?" was assigned zero points. Studies with a total score of 5– 10/10 (≥ 50%) were considered to be of high quality, and scores of 0–4/10 (<50%) as low quality (20). Two evaluators assessed the quality of the included studies independently. In the event of disagreements, a group discussion was held with a third expert to reach a mutual consensus.

# Statistical Analysis

To estimate the effect of WBV training on human gait, a metaanalysis compared the intervention groups with the control groups. Within group comparisons were added (i.e., pre vs. post intervention) when the groups were not comparable (e.g., statistical difference in outcomes at baseline or additional training in control group not provided in the intervention group). Estimations were calculated using the methodology described by Wan et al. (22) when mean and standard deviations were not reported by the authors and medians and interquartile ranges were used. The authors were contacted to request additional data when an estimation was not possible.

If no response was received, the variables were excluded from meta-analysis.

Statistical analysis and figures (i.e., forest plot to facilitate the visualization of values) were produced using a random-effect model in Review Manager software (RevMan, v 5.3, Cochrane Collaboration, Oxford UK) (23). A random-effect model was used to take into account the heterogeneity between the study effects. The effect size of the interventions was reported by standard mean difference (SMD) and their respective 95% Confidence Interval (CI). In this way, the magnitude of the overall effect can be quantified as very small (<0.2) small (0.2– 0.49), moderate (0.5–0.79) or large (≥0.8) (24, 25). Statistical heterogeneity was calculated using the I 2 and Cochrane Q statistic tests (25). Statistical significance was set at p < 0.05.

# Level of Evidence

The strength of evidence of primary outcomes was established as described by Van Tulder et al. (26) based on the results of metaanalysis (effect size), statistical heterogeneity (I 2 ) and risk of bias (PEDro scale). The level of evidence was considered strong with multiple high-quality RCT (at least two studies with a PEDro score ≥5/10) that were statistically homogenous (I <sup>2</sup> p ≥ 0.05). The level of evidence was considered moderate with multiple low-quality studies (two studies with a PEDro score <5/10) that were statistically homogenous and/or one high quality RCT. The level of evidence was considered limited when only one low quality RCT was identified. The level of evidence was conflicting when there were multiple statistically heterogenous studies (I <sup>2</sup> p < 0.05).

# RESULTS

# Included Studies

A total of 816 titles were screened in the first search stage, 43 more were included through cross-referencing, and 692 were excluded because they did not concern our research question. The main reasons for exclusion were: absence of WBV treatment (e.g., studies using local vibrations were excluded), measurement of acute effects, no value for dynamic balance, case studies and reviews. Following exclusion, 167 studies were considered for an abstract review. A further 104 were excluded in this second stage because they did not meet the inclusion criteria. Finally, 63 fulltext articles were assessed for eligibility with 17 not accepted: five because training lasted < 4 weeks, six because they were not RCT, four because there were no walking outcomes, one because it combined WBV training with non-physical therapy and one for comparing WBV training combined with another intervention not provided in the control group (meaning that not only WBV effects were measured). Thus, 46 articles were ultimately included in this systematic review (9, 27–69, 71, 72). A summary of the study selection is provided in **Table 1**.

# Characteristics of the Populations

A total of 2 029 patients took part in the 46 studies selected in this review (see **Table 1**). The sample size ranged from 14 to 159 participants, with a mean age of 60.9 ± 20.0 years, varying from 7.9 years to 83.2 years. With regard to the adult population, 16 studies evaluated the effects of WBV in the elderly (n = 59.8 ± 35.4 subjects) (9, 30–32, 36, 40, 44, 45, 50, 53, 55, 56, 59, 62, 64, 69), four in patients with Chronic Obstructive Pulmonary Disease (COPD) (n = 42.5 ± 16.7 subjects) (57, 61, 65, 71), seven in patients with stroke (n = 46.1 ± 27.2 subjects) (28, 35, 38, 51, 54, 67, 72), four in patients with osteoarthritis (OA) (n = 32.2 ± 11.9 subjects) (29, 33, 63, 68), three in postmenopausal women (n = 40.3 ± 12.5 subjects) (48, 58, 66), two in patients with multiple sclerosis (n = 29.5 ± 6.3 subjects) (34, 39) and one in patients with the following pathologies: incomplete cervical spinal injury (47), pulmonary arterial hypertension (42), lung transplantation (43), idiopathic Parkinson's disease (41), total knee arthroplasty (49) and cerebral palsy (27) (n = 30.0 ± 26.4 subjects). With regard to the child population, two studies evaluated the effects of WBV in cerebral palsy (37, 60), one in patients with osteogenesis imperfect (46) and one in patients with spastic diplegia or quadriplegia forms of cerebral palsy (52) (n = 22.5±5.9 subjects). Most of the studies included both males and females, except for nine studies that either did not mention the participants' gender or selected only males or females (including the three studies on post-menopausal women). Most of the studies clearly explained their eligibility criteria and had similar baselines (no significant differences between groups in any outcomes before the intervention) in their groups, except in 10 articles.

# Training Protocols

The duration of the WBV training interventions ranged from four to 32 weeks, with between two and five sessions per week, with a mean of 3.1 ± 0.8 (three sessions per week in 31 of the 46 selected articles). The frequency and amplitude used in the training sessions ranged from 2 to 45 Hz and from 0.44 to 20 mm, respectively. The intensity of the training sessions, by frequency and/or amplitude, was progressively increased in 30 studies, and remained unchanged in the other selected studies. Some WBV platforms delivered the vibrations alternating between the right and the left foot, while the right and left foot moved up and down at the same time in other vibration plates (70). Synchronous vibrations were delivered in 20 studies, side-alternating vibrations were used in 11 studies, while 15 studies did not mention the type of vibration in their intervention method.

For the groups that were exposed to WBV training (interventions groups), vibrations were delivered while participants stood in static positions (e.g., squat or lunge positions) in 27 studies and dynamic exercises were provided in 11 studies. In the remaining eight studies, both static and dynamic exercises were combined during the WBV training sessions. The number of WBV sets per training session ranged between 1 and 135. The duration of the vibration sets ranged from 10 s to 3 min, with a between-sets resting time ranging between 3 s and 5 min. For the groups not exposed to WBV training interventions (control groups), participants performed strengthening and balance exercises without WBV in fourteen studies, had no intervention and were asked to maintain their habitual lifestyle in sixteen studies, were exposed to a sham intervention in six studies, continued to follow their conventional physiotherapy in four studies, received relaxation exercises in four studies and performed walking training sessions in two studies.

# Gait Motor Outcomes

The "Timed Up-and-Go" (TUG) test and the "six-minute walking test" (6MWT) were the clinical outcomes most frequently used to assess gait (in 29 and 18 studies, respectively). The "ten-meter walking test" (10MWT) was used in 10 studies to assess gait velocity. Walking speed was also evaluated using biomechanical and kinematic assessments (e.g., walking on a platform or camera motion analysis) in six studies. Other temporal and spatial parameters such as time of swing phase and stance phase, stride length and step length were presented in only two studies. Gait quality was assessed using the gait score of the Tinetti test in five studies. Finally, other outcomes were used once in all 46 studies: the "functional ambulation categories test" with stroke patients, the "50-foot walking test" with knee OA patients, the "25-foot walking test" with multiple sclerosis patients, the "two-minute walking test" with knee OA patients, and the time to walk four meters in postmenopausal women. A summary of the primary outcomes related to gait is provided in **Table 1**.

# Quality Assessment

The results from the quality assessments for each of the studies for respective quality indexes are provided in **Table 2**. According to the PEDro Scale, 40 studies obtained a high-quality methodology score while six studies were rated as low quality.

The mean score was 5.8 ± 1.4 with a median of 5.5 and a range of scores from 3 to 9. The highest-quality methodology scores were found in the articles concerning stroke patients, with a mean

#### TABLE 1 | Descriptive checklist of the included studies.








WBV, Whole body vibration; TUG, Timed up and go test ; 6MWT, 6-minute walk test; 2MWT, 2-minute walk test; 50FWT, 50-feet walk test; 10MWT, 10-meter walk test.

score of 7.2 ± 1.7. The poorest methodological quality was found for postmenopausal women with a mean score of 4.7 ± 1.1.

# Studies Included for Meta-Analysis

A total of 25 studies were included in statistical analysis. Eleven studies were included for meta-analysis in the elderly (30–32, 40, 44, 45, 50, 53, 59, 64, 69), four studies for COPD patients(57, 61, 65, 71), four studies for stroke patients (35, 38, 51, 54), four studies for patients with knee OA (29, 33, 63, 68) and two studies for patients with Multiple Sclerosis (MS) (3, 34).

# Results Ranked According to Aging and Pathology

### Elderly Subjects

Sixteen studies examined the effect of WBV on elderly subjects(9, 30–32, 36, 40, 44, 45, 50, 53, 55, 56, 59, 62, 64, 69). The studies had an average PEDro score of 5.5 ± 1.0. The sample size ranged from 19 to 159 participants with a mean age of 76.5 ± 5.8 years. Most of the studies included both men and women except for three with women only (32, 55, 62) and one with only men (45). Only one study failed to mention the eligibility criteria (44) and seven studies exhibited heterogeneity in their baselines (9, 36, 40, 44, 55, 62, 64). Training duration varied from 6 weeks to 8 months. Fifteen studies had a frequency of three sessions per week (9, 30–32, 36, 40, 44, 45, 50, 53, 55, 56, 59, 64, 69) while one study involved two sessions per week (62). The frequency and amplitude of platform vibrations varied from 10 to 40 Hz and 0.5 to 8 mm, respectively. Intensity was progressively increased in 11 studies (9, 30, 32, 40, 45, 50, 53, 55, 59, 62, 64). Eight studies used synchronous vibrations (9, 30, 31, 36, 40, 50, 59, 69) while the other eight studies (32, 44, 45, 53, 55, 56, 62, 64) did not mention the type of vibrations delivered by their devices. The number of vibration bouts delivered per session varied from two to 39 sets with a period lasting between 15 sand 3 min each. Resting time was between 5 s and 5 min. In nine protocols (9, 30, 31, 36, 40, 44, 53, 56, 69), the subjects maintained a static position, while they performed dynamic exercises in three studies (32, 50, 55), or both in four studies (45, 59, 62, 64). The most frequently used outcome was TUG, found in 14 studies (9, 30–32, 36, 40, 45, 50, 53, 55, 59, 62, 64, 69). Six studies combined TUG with the Tinetti gait score (9, 30, 31, 36, 40, 64). Four studies assessed gait speed using the 10MWT (32, 45, 59, 69). Three studies assessed functional performance with the 6MWT (44, 50, 56). Two studies used the Locometrix system for biomechanical analysis (31, 36).

### **Comparisons to control groups**

Four meta analyses (9, 31, 36, 55) were conducted for the following outcomes: TUG test, 10MWT, Tinetti test and 6MWT.

For the TUG test (**Figure 2A**), 10 studies were included in meta-analysis and four studies were excluded due to a lack of data despite requests to the authors (9, 31, 36, 55). Meta-analysis showed a significant decrease in time in favor of the WBV groups (SMD = −0.18; 95% CI: −0.33, −0.04), with consistent results (I <sup>2</sup> = 7%, p = 0.38). The included studies were of high quality

#### TABLE 2 | Quality assessment with the PEDro scale.


n, criterion not fulfilled; y, criterion fulfilled; 1, eligibility criteria were specified; 2, subjects were randomly allocated to groups or to a treatment order; 3, allocation was concealed; 4, the groups were similar at baseline; 5, there was blinding of all subjects; 6, there was blinding of all therapists; 7, there was blinding of all assessors; 8, measures of at least one key outcome were obtained from more than 85% of the subjects who were initially allocated to groups; 9, intention-to-treat analysis was performed on all subjects who received the treatment or control condition as allocated; 10, the results of between-group statistical comparisons are reported for at least one key outcome; 11, the study provides both point measures and measures of variability for at least one key outcome; total score, each satisfied item (except the first) contributes 1 point to the total score, yielding a PEDro scale score that can range from 0 to 10. B, the level of evidence was B (randomized control trials that lacked double-blinding).

(mean PEDro score = 5.8 ± 1.0), so a strong level of evidence supports the positive effect of WBV training on the TUG test.

For the 10MWT (**Figure 2B**), three studies were included in meta-analysis and one study was excluded because it used a different unit of measure (i.e., m/s instead of seconds in the other studies) (69). Meta-analysis showed a significant decrease in time on the 10MWT in WBV groups (SMD = −0.28; 95% CI: −0.56, −0.01), with consistent results (I <sup>2</sup> = 22%, p = 0.28). The overall quality of the included studies was high (PEDro score = 5.0 ± 0.0). Thus, a strong level of evidence supports the positive effect of WBV training in improving gait speed on the 10MWT.

For the 6MWT (**Figure 2C**), two studies were included and one was excluded due to a lack of data despite requests to the authors (56). Meta-analysis showed no significant difference between groups (SMD = 0.37; 95% CI: −0.03, 0.78), despite a tendency toward an improvement in distance in WBV groups. Results were consistent (I <sup>2</sup> = 0%, p = 0.43) and the quality of the included studies was high (PEDro score = 5.5 ± 0.7). Thus, a strong level of evidence supports the lack of a beneficial effect of WBV training for improving performance in the 6MWT.

For the Tinetti gait score (**Figure 2D**), three studies were included in meta-analysis and three were excluded due to a lack of data despite requests to the authors (9, 31, 36). Meta-analysis showed no significant difference between groups (SMD = 0.04; 95% CI: −0.23, 0.31), with consistent results (I <sup>2</sup> = 0%, p = 0.46). The quality of the included studies was high (mean PEDro score = 7.0 ± 1.0). Thus, a strong level of evidence supports the absence of a positive effect of WBV training on the Tinetti gait score.

For biomechanical data recorded using the Locometrix system (gait speed, stride frequency, stride length, stride symmetry, stride regularity, cranio-caudal mechanic power, antero-posterior mechanic power, medio-lateral mechanic power, and counting speed), no comparison between groups could be performed due to a lack of data despite requests to the authors (31, 36). Both Beaudart et al. (31) and Buckinx et al. (36) reported no significant inter-group difference for parameters recorded by the Locometrix (p > 0.05).

## Chronic COPD Patients

Four studies examined the effect of WBV on chronic COPD patients (57, 61, 65, 71) with an average PEDro score of 5.2 ± 0.5. The sample size ranged from 28 to 62 participants with a mean age of 66.2 ± 4.3 years. Three studies included both men and women (61, 65, 71) and one included only male patients (57). All of the studies specified the eligibility criteria and had similar baselines. The training duration varied from 6 weeks to 3 months. In two studies (57, 61), subjects performed three WBV sessions per week, while patients had only two sessions per week in the other two studies (65, 71). The frequency and amplitude of the platform vibrations varied from 6 to 35 Hz and 2 to 6 mm, respectively. Intensity was progressively increased in two studies (65, 71). Half of the studies used sidealternating vibrations (65, 71) while the two other studies used synchronous vibrations (57, 61). The number of vibration bouts delivered per session varied from three to eight sets with a period lasting between 30 s and 2 min for each. Resting time was 60 s to 2 min. In two protocols (57, 71), the subjects maintained a static position, while they performed dynamic exercises in the other studies (61, 65). Only the 6MWT methodology was used to test gait.

### **Comparisons to control groups (Figure 3A)**

For the meta-analysis, two studies were included and two were excluded because the control groups were intervention groups with additional exercises not provided in the WBV group (i.e., not only WBV effects are measured) (61, 65). Meta-analysis showed no significant difference between groups (SMD = 1.66; 95% CI: −0.17, 3.49) with heterogeneous results (I <sup>2</sup> = 91%, p = 0.0008). Thus, the level of evidence was conflicting for the 6MWT outcome in COPD.

For the excluded studies, Salhi et al. (61) showed that there was no significant difference between WBV training and conventional resistance training for improving 6MWT scores (SMD = −0.24; 95% CI: −0.79, 0.31). Similar results were found by Spielmanns et al. (65), where no significant difference was shown between the WBV intervention and the calisthenics group (SMD = 0.54; 95% CI: −0.23, 1.32).

#### **Comparison to pre-intervention (Figure 3B)**

A second meta-analysis was conducted to include the four studies. Meta-analysis demonstrated a significant improvement in the distance walked during the 6MWT after WBV treatment (SMD = 0.92; 95% CI: 0.32, 1.51). Again, because there were heterogeneous results (I <sup>2</sup> = 66%, p = 0.03), the level of evidence was conflicting for the 6MWT outcome.

#### Stroke Patients

Seven studies examined the effect of WBV on stroke patients (28, 35, 38, 51, 54, 67, 72) with an average PEDro score of 7.2 ± 1.7. The sample size ranged from 21 to 84 participants with a mean age of 58.3 ± 4.5 years. All of the studies included both men and women. All explained the eligibility criteria. Two studies (35, 38, 51, 72) found significant differences between groups for some outcomes at baseline. The training duration varied from 4 to 8 weeks. In four studies, subjects performed three sessions per week (28, 51, 54, 72), while patients had five sessions per week in two studies (38, 67), and two sessions per week in one study (35). The frequency and amplitude of platform vibrations varied from 20 to 40 Hz and 0.44 to 5 mm, respectively. The intensity was progressively increased in two studies (51, 72). Three studies used side-alternating vibrations (38, 67, 72), three synchronous vibrations (35, 51, 54), while one did not mention the type of vibrations (28). The number of vibration bouts delivered per session varied from 2 to 135 sets with a period lasting from 10 to 150 s each. Resting time was between 3 and 60 s. In four protocols (28, 35, 38, 67), the subjects maintained a static position, performed dynamic exercises in two studies (51, 72) and both types of exercises in one study (54). The TUG test was assessed in three studies (35, 38, 54), the 6MWT in three studies (35, 51, 54) and the 10MWT in two (28, 51). Only one study used a biomechanical methodology to assess gait function (72) and one study used the Functional Ambulation Categories (FAC) scale (67).

#### **Comparisons to control groups (Figures 4A and 4C)**

Two meta-analyses were conducted for the TUG test and the 6MWT.

For the TUG test, two studies were included and one study was excluded because the groups were statistically different at baseline (35). Meta-analysis demonstrated no significant difference between groups (SMD = −0.21; 95% CI: −0.55, 0.13), with consistent results (I <sup>2</sup> = 0%, p = 0.83). The quality of the study was high (mean PEDro score = 8.0 ± 0.0). Thus, a strong level of evidence supports the absence of effect of WBV training on the TUG test in stroke patients.

compared to the pre-intervention status (B).

For the 6MWT, two studies were included and one study was excluded because the groups were statistically different at baseline (35). Meta-analysis demonstrated no significant difference between the groups (SMD = −0.09; 95% CI: −0.37, 0.19), with consistent results (I <sup>2</sup> = 0%, p = 0.70). The quality of the study was high (mean PEDro score = 6.0 ± 2.8). Thus, a strong level of evidence supports the absence of effect of WBV training on the 6MWT test in stroke patients.

For biomechanical data, Choi et al. (72) demonstrated no significant difference between groups for stride length (SMD = 0.50; 95% CI: −0.23, 1.23) and walking speed (SMD = 0.32; 95% CI: −0.40, 1.04). Similarly, walking speed assessed by the 10MWT (51) was not different between groups (SMD = 0.39; 95% CI: −0.05, 0.83). Finally, the Functional Ambulation categories scale (67) was not different between groups (SMD = 0.00; 95% CI: −0.54, 0.54) after the interventions. All studies were of high quality RCT (Perdro scores ≥ 5/10). Thus, the level of evidence for each outcome was considered moderate.

#### **Comparisons to pre-intervention (Figures 4B and 4D)**

Two additional meta-analyses were conducted to include the two studies excluded for group comparisons for the TUG test and the 6MWT outcomes.

For the TUG test, meta-analysis showed a tendency but no significant improvement after the WBV treatment (SMD = −0.29; 95% CI: −0.60, 0.01) with consistent results (I <sup>2</sup> = 0%, p = 0.89). The overall quality of the included studies was high (mean PEDro score= 7.0 ± 2.6). Thus, a strong level of evidence supports the absence of effect of WBV treatment on the TUG test in stoke patients.

For the 6MWT, meta-analysis showed a significant improvement after WBV treatment (SMD = −0.33; 95% CI: 0.06, 0.59) with consistent results (I <sup>2</sup> = 0%, p = 0.58). The overall quality of the included studies was high (mean PEDro score = 8.3 ± 0.5). Thus, a strong level of evidence supports the positive effect of WBV treatment to improve the distance walked during the 6MWT test in stroke patients.

#### Knee Osteoarthritis

Four studies examined the effect of WBV on patients suffering from knee osteoarthritis (29, 33, 63, 68). The studies had an average PEDro score of 6.5 ± 1.2. The sample size ranged from 21 to 49 subjects with a mean age of 65.1 ± 9.2 years. Two studies included both men and women (33, 68), while two studies did not mention the gender of the patients (29, 63). All of the studies specified the eligibility criteria and had similar baselines. The training duration ranged from 8 to 24 weeks. Three studies had a frequency of three sessions per week (29, 33, 63) while the other had five (68). The frequency and amplitude of the platform vibrations varied from 25 to 40 Hz and 2 to 6 mm, respectively. The intensity was progressively increased in all studies. Two studies used synchronous vibrations (33, 63) while two did not mention the type of vibrations of the devices (29, 68). The number of vibration bouts delivered per session varied from six to 30 sets with a period lasting 20 to 70 s. Resting time was between 20 and 70 seconds. In three protocols (29, 33, 68), the subjects maintained a static position, but performed static and dynamic exercises in the other study (63). Three studies used the TUG test (29, 33, 68), three used the 6MWT (29, 63, 68) and one combined the 2MWT and the 50FWT with the TUG (33).

#### **Comparisons to control groups**

For the TUG test (**Figure 5B**), two studies were included and one was excluded due to a lack of data (33). Meta-analysis showed no significant difference between groups (SMD = −1.54; 95% CI: −4.65, 1.56) with heterogeneous results (I <sup>2</sup> = 97%, p < 0.00001). Thus, the level of evidence was conflicting.

For the 6MWT (**Figure 5A**), meta-analysis showed a significant difference in favor of the WBV group (SMD = 1.28;

95% CI: 0.57, 1.99), with consistent results (I <sup>2</sup> = 64%, p = 0.06). The quality of the studies was high (mean PEDro score = 6.6 ± 1.5). Thus, a strong level of evidence supports the positive effect of adding WBV to improve the 6MWT in patients with knee OA.

### Postmenopausal Women

Three studies examined the effect of WBV on postmenopausal patients (48, 58, 66). These studies had an average PEDro score of 4.6 ± 1.1. The sample size ranged from 27 to 52 participants with a mean age of 65.8±8.4 years. All of the studies specified eligibility and had similar baselines. The training durations ranged from 4 weeks to 8 months. One study had a frequency of five sessions per week (66), another of three sessions per week (58), while the last one did not specify the number of sessions per week (48). The frequency of the platform vibrations varied from 6 to 35 Hz and the amplitude was indicated in only one study (6 mm). The intensity of the sessions was progressively increased in two studies during training duration (58, 66). One study used synchronous vibrations (66) and the other two studies did not mention the type of vibrations (48, 58). The vibration bouts were delivered from 30 to 60 s with two to six sets. Resting time was 60 s in two studies (58, 66) and was not indicated in the third (48). In all protocols, the subjects maintained a static standing position. Two studies used the TUG (48, 66) and one combined it with a 10MWT (48). The third study measured walking speed along a four-meter pathway (58). Meta-analysis could not be

performed for the TUG test due to a lack of post-intervention data in all three studies, despite requests to the authors. Two studies reported significant improvement of the 10MWT after WBV training (p < 0.05 and p = 0.006) (48, 58). Sucuoglu et al. (66) showed a significant improvement of the TUG test post treatment (p < 0.005), whereas Iwamoto et al. (48) found no significant difference between groups (p > 0.05).

#### Multiples Sclerosis

Two studies examined the effect of WBV on patients with multiple sclerosis (34, 39). The studies had an average PEDro score of 6.0 ± 1.4. The sample size ranged from 25 to 34 participants with a mean age of 43.4 ± 6.3 years. Both studies included both men and women, specified the eligibility criteria and had similar baselines. The training duration was 10 and 20 weeks. In one study, patients underwent three sessions per week (39), while in the other they performed an average of 2.5 sessions per week (34). The frequency and amplitude of the platform vibrations varied from 2 to 45 Hz and 2 to 2.5 mm, respectively. The intensity was progressively increased in both studies. One study used synchronous vibrations (34) while the other did not mention the type of vibrations (39). The number of vibration bouts delivered per session varied from 2 to 15 sets with a period lasting 30 to 120 s. Resting time was between 30 and 120 s. In one protocol (39), the subjects maintained a static position, while they performed static and dynamic exercises in the other (34). Both studies used the TUG test. One study combined it with the 10MWT and the 6MWT (39), while the other used the TUG with the 2MWT and the 25-foot walk test (34).

#### **Comparisons to control groups**

In one study, no meta-analysis was conducted for between-group comparisons because the groups were statistically different at baseline for the TUG test and 2MWT (34).

Ebrahimi et al. (39) found no significant difference between groups for the TUG test (SMD = −0.47; 95% CI: −1.20, 0.26). However, they did observe significant improvement in the WBV group for the 10MWT (SMD = −1.05; 95% CI: −1.82, −0.28) and the 6MWT (SMD = 1.22; 95% CI: 0.43, 2.01). The level of evidence was high (PEDro score = 5/10). Thus, the level of evidence was considered moderate for each outcome.

#### **Comparison to pre-intervention**

Meta-analysis (**Figure 6**) showed no significant improvement in the TUG test after WBV training (SMD = −0.11; 95% CI−0.64, 0.43) with consistent results (I <sup>2</sup> = 0%, p = 0.71). The overall quality of the included studies was high (mean PEDro score = 6.0 ± 1.4). Thus, there is a strong level of evidence to conclude that WBV treatment had no impact on the TUG test in patients with multiple sclerosis.

#### Other Pathologies in Adults

Six studies reported results on different pathologies in adult patients(27, 41–43, 47, 49): incomplete cervical spinal injury (47), pulmonary arterial hypertension (42), lung transplantation (43), idiopathic Parkinson's disease (41), total knee arthroplasty (49) and cerebral palsy (27). The average PEDro score was 5.3 ± 1.7. The sample size ranged from 14 to 83 subjects with a mean age of 54.6 ± 14.1 years. All of the studies included both men and women. All specified the eligibility criteria and had similar baselines except for one study where patients were statically different at baseline for certain outcomes (49). Training duration varied from 4 to 8 weeks. In four studies, subjects performed three WBV sessions per week (27, 41, 43, 49), while in one study they had four sessions per week (42), and five in another (47). The frequency and amplitude of the platform vibrations varied from 6 to 40 Hz and 2 to 20 mm, respectively. The frequency or amplitude of the vibrations was progressively increased in four studies (27, 43, 47, 49). Two studies used side-alternating vibrations (42, 43), one study used synchronous vibrations (47) and the other three studies did not mention the type of vibrations (27, 41, 49). The number of vibration bouts

delivered per session varied from 1 to 18 sets with a period lasting 30 to 120 s. Resting time was between 15 and 240 s. In three protocols (27, 41, 47), the subjects maintained a static position, while they performed dynamic exercises in two others (42, 43) and combined both in the last study (49). Four studies used the TUG test (27, 41, 47, 49) and three used the 6MWT (27, 42, 43) (one study combined both). One study used biomechanical analysis combined with the clinical TUG test (41). No metaanalysis was conducted due to the heterogeneity of patients in this subgroup.

### **Comparisons to control groups**

In patients with incomplete cervical spinal injury, In et al. (47) found no significant difference between the WBV group and control group for the TUG test (SMD = −0.64; 95 CI: −1.40, 0.13) and the 10MWT (SMD = −0.23; 95% CI: −0.97, 0.52). The quality of the study was high (PEDro score = 8/10). Thus, the level of evidence was moderate.

In patients diagnosed with idiopathic Parkinson's disease, Gaßner et al. (41) observed no significant difference between the WBV group and the placebo group for the TUG test (SMD = −0.37; 95% CI: −1.34, 0.59), gait velocity (SMD = −0.21; 95% CI: −1.17, 0.74) and step length (SMD = 0.14; 95% CI: −0.81, 1.09). The quality of the study was high (PEDro score = 5/10). Thus, the level of evidence was moderate.

After total knee arthroplasty, Johnson et al. (49) reported no significant difference for the TUG test between a WBV group and a resistance training group (SMD = −0.59; 95% CI: −1.59, 0.42). The quality of the study was low (PEDro score=3/10). Thus, the level of evidence was limited.

In patients with cerebral palsy, Ahlborg et al. (27) found no significant difference between a WBV group and a resistance training group for the TUG test (SMD = 0.28; 95% CI: −0.77, 1.34). The quality of the study was high (PEDro score=6/10). Thus, the level of evidence was moderate.

SMD could not be reported in the study of Gerhardt et al. (42) and Gloeckl et al. (43) due to the lack of post-intervention data despite a request to the authors. The authors of the first study indicated that WBV was associated with a significant improvement of the 6MWD vs. baseline of +38.6 ± 6.6 m (p < 0.001) (42). The authors of the second study reported a significant between-group difference of 28 m (95% CI: 3, 54; p = 0.029) in favor of WBV (73).

#### **Comparisons to pre-intervention**

Ahlborg et al. (27) reported no significant difference after WBV (SMD = 0.14; 95% CI: −0.91, 1.19) in patients with cerebral palsy, with a level of evidence considered moderate (PEDro score = 6/10). Similarly, Johnson et al. (49) observed no significant improvement of the TUG test after WBV (SMD = 1.02; 95% CI: −0.04, 2.09) in patients with total knee arthroplasty, with a limited level of evidence (PEDro score = 3/10).

### Other Pathologies in Children

Four studies examined the effect of WBV in children(37, 46, 52, 60): two evaluated the effects of WBV in cerebral palsy (37, 60), one in patients with osteogenesis imperfect (46) and one in patients with spastic diplegia or quadriplegia forms of cerebral palsy (52). The studies had an average PEDro score of 6.0 ± 1.4. The sample size ranged from 16 to 30 participants with a mean age of 8.7 ± 0.8 years. All of the studies included both boys and girls, specified the eligibility criteria and had similar baselines. The training duration varied between 8 and 24 weeks. One study had a frequency of two sessions per week (46), one had three sessions per week (52), one had five sessions per week (60), while the number of sessions was not specified in the last study (37). The frequency and amplitude of the platform vibrations varied from 5 to 25 Hz and 1 to 9 mm, respectively. The frequency or amplitude of the vibrations was progressively increased in three studies (46, 52, 60). Two studies used synchronous vibrations (37, 60) and the other two studies used side-alternating vibrations (46, 52). The number of vibration bouts delivered per session varied from three to six sets with periods lasting 3 min each. Resting time was also 3 min. In two protocols (37, 52), the subjects maintained a static position, while they performed static and dynamic exercises in the other studies(46, 60). Two studies used the 6MWT (37, 46), and one study combined it with the TUG test (37). One study used biomechanical analysis including gait speed, stride length and cycle time (52). One study used the 10MWT (for gait speed) (60). Three studies reported significant improvement in gait parameters following WBV treatment (37, 52, 60), whereas one study reported that the 6MWT remained unchanged (46). No meta-analysis was conducted due to the heterogeneity of patients in this subgroup. Additionally, data post interventions were not reported in three studies (37, 46, 60).

#### **Comparisons to control groups**

In children with clinically mild to moderate osteogenesis imperfecta, Högler et al. reported no significant difference between groups for the 6MWT (p = 0.278) (46).

In children with cerebral palsy, Cheng et al. reported a significant difference between the treatment and control conditions for the 6MWT (p = 0.005) (37). Ruck et al. reported a significant improvement in the 10MWT in favor of WBV (p = 0.03) (60).

Finally, in patients with either spastic diplegia or quadriplegia forms of cerebral palsy, Lee and Chon found significant improvement in favor of the WBV group for gait speed (SMD = 1.41; 95% CI: 0.60, 2.22) and stride length (SMD = 0.91; 95% CI: 0.15, 1.67) (52), with a level of evidence considered moderate (PEDro score = 8/10).

# DISCUSSION

The aim of this systematic review was to determine the changes in gait outcomes after WBV training in healthy adults and various patient categories. We found a strong level of evidence for a positive effect of WBV training on the TUG test and the 10MWT in the elderly. The same level of evidence was found in favor of a significant improvement of the 6MWT in stroke patients and patients with knee OA. In contrast, there is no change in the 6MWT and the Tinetti gait score in the elderly, and the TUG test was not improved in stroke or multiple sclerosis patients. Conflicting results were found in COPD patients despite significant improvements in the 6MWT. Other outcomes showed a moderate or limited level of evidence, due to a lack of data or because only one RCT was identified.

As mentioned in a prior review (20), the major obstacle in conducting meta-analysis and establishing strong evidence on the effects of WBV on gait is the heterogeneity in study methodologies. Intervention regimes, settings, combined interventions and control groups varied greatly. As in any training protocol, numerous factors can affect the results of the program (e.g., the duration of the intervention; the frequency or volume of the sessions; the type, frequency and amplitude of the vibrations and the exercises performed on the platform). Because the studies used different protocols, a random-effects model was used. In the presence of heterogeneity, a random-effects metaanalysis weights the studies relatively more equally than a fixedeffect analysis (25). Because control groups varied a great deal in terms of interventions (i.e., exercises, physical therapy, sham, no interventions etc.), intergroup comparisons were not always possible, so we added within group comparisons (i.e., pre vs. post WB) to estimate the effect of the treatment. Additionally, some groups were statistically different at baseline for certain outcomes (35, 40, 62), making between groups comparisons impossible after intervention. The results of each patient category are discussed below.

On the one hand, the results in the elderly showed significant improvements in the TUG test and the 10MWT after WBV intervention. These results are in favor of better dynamic stability and gait performance as both outcomes are related to balance and gait speed, respectively (45, 59).

However, the effect sizes were small (−0.18 and−0.28, respectively) and the 6MWT was not significantly modified by treatment despite a tendency toward improvement in favor of WBV (SMD = 0.37; 95% CI: −0.03, 0.78). Our findings on physical improvements partially corroborate earlier reports. In a recent scoping review, Park et al. (74) concluded that WBV training could be effective in increasing lean mass, muscular strength and cardiovascular health (74). Positive changes in body composition and fitness induced by WBV training may explain the improvement in gait performance. After the age of 50, muscle mass decreases approximately 2% every year and muscle strength decreases 15% every 10 years (75). These agerelated changes impact functional mobility, including gait speed, static dynamic balance, and the risk of falling. As a resistance training exercise, WBV appears efficient in attenuating the loss of muscle mass and muscle strength. In order to combat the effects of aging, it should be recommended that older adults perform WBV 2 or 3 days per week, as suggested for resistance training (76, 77). In addition, because both offer numerous benefits, these interventions could be combined, depending on feasibility and patient motivation. These recommendations are also valuable for patients with reduced mobility and who need to improve their autonomy at home, as only the TUG test and 10MWT (shortdistance walking tests) were improved but not the 6MWT (a long-distance walking test).

On the other hand, the results support that gait performance can be improved with no improvement in qualitative aspects of the locomotion pattern. In fact, the quality of gait, assessed by the Tinetti test, was not changed. The outcome is classically divided in two parts. One assesses static balance, while the second asses dynamic balance (78, 89). Because gait was the main outcome of the present research, we did not include the total score in meta-analysis. All 28 points are necessary to assess the whole balance score, and thus, the risk of falling. Additionally, the Tinetti total score has been recently demonstrated as being related to muscle mass and strength (79). As previously discussed, improvements in muscle mass, strength, and performance are demonstrated after WBV training. Thus, changes in the Tinetti total score can be expected and further investigation on this outcome is warranted.

The results in stroke patients are more mitigated. On the one hand, between-group comparisons showed that the 6MWT was not modified by WBV training. However, two studies had to be excluded from this analysis and a second analysis was performed to include them in a pre vs. post comparison. This time, the results showed a significant increase in distance after a WBV intervention. On the other hand, no significant changes were found for the TUG test in both comparisons, despite a tendency toward an improvement. The 6MWT is commonly used to assess aspects of walking performance in stroke survivor studies (80). It evaluates the global responses of all the systems involved during exercise, including the pulmonary and cardiovascular systems, systemic circulation, peripheral circulation, blood, neuromuscular units, and muscle metabolism (90). The results concerning the 6MWT confirm the findings discussed previously in favor of functional improvements in elderly and disease populations (74). Conflicting results emerged from the TUG test. Balance, assessed by the TUG test, appeared less modified by the WBV treatment in the treatment of neuropathologic subjects than in the elderly population. This was confirmed by the poor results in multiple sclerosis patients included in meta-analysis for the TUG outcome (34, 39).

Conflicting results were also found for the 6MWT in COPD patients. The pooled studies demonstrated a significant improvement of the distance walked but with inconsistent results. The size of the overall effect was large (SMD = 0.92; 95% CI:0.32, 1.51) and seems to corroborate the functional improvement observed after WBV. However, the results must be taken with caution because of their heterogeneity and the small sample of high quality studies revealed by the present review (i.e., four studies) (57, 61, 65, 71). Heterogeneity was found in both intergroup comparison and comparison with pre intervention, and was difficult to explain as the studies had similar populations (i.e., older adults with COPD) and settings (i.e., frequency and amplitude of the vibrations).

In patients with knee OA, a strong level of evidence supports the beneficial effect of WBV training in improving the 6MWT, with a large effect size (SMD=1.28; 95% CI: 0.57, 1.99). Interestingly, Wang et al. showed that adding WBV to quadriceps resistance training was more efficient that resistance training alone (SMD = 1.68; 95% CI: 1.22, 2.14) (68). A recent review showed that, in patients with knee OA, a resistance training program is effective for improving knee extensor strength but has limited effect on pain and disability if the gains are <30% (81). WBV interventions combined with strength training may help achieve this gain necessary for beneficial effects on pain and functional performance. Again, for the TUG test, Wang et al. (68) demonstrated the value of combining WBV and resistance training to improve gait performance (SMD = −3.11; 95% CI: −3.71, −2.52). However, because of the absence of effect in the second study (29), the level of evidence was conflicting. Heterogeneity might be explained by the two major differences between the studies, which were the addition of quadriceps resistance training and the doubled duration of the training (i.e., 24 vs. 12 weeks) in Wang et al. (68).

Although some significant gait improvements after WBV were reported, it is important to stress that significant statistical changes are not always linked to significant clinical improvements for patients. For example, in elderly patients, Bogaerts et al. (32) reported a significant improvement of the time required to perform the TUG test after WBV intervention. While the time was decreased from 13.1 s to 11.19 s (SMD = −0.71; 95%CI: −1.10, −0.32) with an effect size considered moderate, this difference may not correspond to major changes in the patients' daily activities. Thus, the benefit of a non-functional intervention such as WBV should always be questioned with regard to each patient's goals.

However, WBV appears to be a time-efficient and easy-touse intervention that is both relatively inexpensive and safe for patients with balance deficits. Vibration plates are readily available at all rehabilitation hospitals/centers. Moreover, it might be interesting to complete certain conventional rehabilitation programs like resistance or balance training with WBV training that may offer the same results. For example, for COPD patients, it has been found that both the WBV and resistance training groups significantly improved in the 6MWT (61, 65, 71) with no significant difference between the groups. Additionally, no control group including any form of training was significantly superior to WBV training in improving gait.

Most of the studies included in this review (27/43 articles) reported a drop-out rate of <15% during their interventions. Considering that most of the subjects were patients with diseases or physical disorders, it is logical to assume that they would have stopped treatment had they experienced any harmful or adverse side effect. This might support the hypothesis that the patients tolerated vibration training well. Moreover, WBV training has been reported to be appreciated and considered a safe training method. The fact that participants could perform either dynamic or static exercises while holding a bar increases safety and would be beneficial for the weakest populations such as elderly persons with balance impairments.

The vibration type may impact the training response. sidealternating WBV has been shown to increase heart rate higher than synchronous vibrations in young sedentary women during 20-min sessions (82). This illustrates the potential of WBV to improve fitness capacity, particularly in less active populations. Additionally, higher electromyography of knee extensor and plantar flexor activities were observed with a side-alternating vibration platform compared to synchronous vibrations (83). Although these results are in favor of side-alternating vibrations, our review showed heterogeneous results regarding the vibration type when it was mentioned. Significant improvements in gait parameters were found in 13 studies that used synchronous vibrations (9, 30, 33, 37, 40, 47, 51, 57, 59–61, 63, 66) and in seven studies that used side-alternating vibrations (38, 42, 43, 48, 52, 65, 72). Again, because of the lack of consistency in the protocols and results, it is difficult to reach a consensus on specific WBV training to improve human locomotion.

We chose to select only studies on long-term effects because they are better correlated to conventional physiotherapies that often last many weeks. Moreover, long-term effects have been studied more than short-term or even immediate effects. We found that a wide range of protocols lasted 6 weeks or more and a few lasted 4 weeks. However, we can add that only a few RCT focused on the acute effects of WBV training on gait parameters (84–86). In the future, it might be interesting to compare different WBV protocols (i.e., with different WBV frequencies) in order to evaluate the effect of high vs. low WBV frequency on balance and gait within a single session.

Finally, most studies used clinical assessments instead of biomechanical analysis (41, 52, 72). Since it might be a more objective measure, future studies should integrate this kind of outcome more often in order to compare it with functional assessments.

# LIMITATIONS OF THE STUDY

Results of meta-analysis must be taken with caution as some studies could not be included in the comparisons due to a lack of complete data, notably for the TUG test and Tinetti gait score in the elderly, despite requests to the authors.

The Cochrane Qualitative and Implementation Methods Group recommends the application of Grades of Recommendation, Assessment, Development, and Evaluation (GRADE) in the Evidence from Qualitative Reviews to assess confidence in qualitative synthesized findings (87). However, the GRADE necessitates assessing the risk of publication bias with a funnel plot, determining its asymmetry, which can be performed with at least ten studies (88). Because most of the statistical analyses were conducted on few RCT, we decided to implement other guidelines described by a Cochrane collaboration group to assess the level of evidence (26). Because this method includes fewer criteria, our confidence in the results must be taken with caution.

# CONCLUSION

While WBV training appears to be a useful and relatively successful tool in improving gait and walking abilities, it remains unclear whether the treatment could be generalized to all patients. Some populations have been studied more than others with varying degrees of consistency. In the elderly, there is a strong level of evidence that WBV can improve mobility by improving the TUG test, and gait speed by improving the 10MWT. The results also showed significant improvements to functional performance in stroke patients and patients with knee OA by improving the 6MWT. However, the treatment

# REFERENCES


was inefficient in changing the TUG test in stroke and multiple sclerosis patients, and conflicting results were obtained for the 6MWT in COPD. Finally, other outcomes were studied less and the level of evidence was moderate or even limited depending of the quality of the study. The transferability of this kind of training to daily activities remains unclear and the use of vibration training to replace functional rehabilitation must always be questioned. Further research is needed to explore the possibility of finding a standardized protocol targeting gait ability in a wide range of populations.

# AUTHOR CONTRIBUTIONS

MF, TV, GL, PF, TH, LC, J-LH, EY, P-AD, and AD designed the study, collected, analyzed, and interpreted the data, drafted and revised the manuscript, tables and figures, and gave final approval.

# FUNDING

This article was funded by the Agence Régionale de Santé d'Ilede-France (ARS-IDF).

Exp Physiol. (2009) 94:103–16. doi: 10.1113/expphysiol.2008.0 42689


transplantation: a randomized, controlled trial. J Heart Lung Transpl. (2015) 34:1455–61. doi: 10.1016/j.healun.2015.07.002


training on the muscle mass and functional capacity of elderly women. J Strength Cond Res. (2015) 29:1863–1869. doi: 10.1519/JSC.0000000000000830


**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 Fischer, Vialleron, Laffaye, Fourcade, Hussein, Chèze, Deleu, Honeine, Yiou and Delafontaine. 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 Functioning, Muscle Power and Reactive Balance Are Major Contributors to Gait Adaptability in People With Parkinson's Disease

Maria Joana D. Caetano1,2 , Stephen R. Lord2,3\*, Natalie E. Allen<sup>4</sup> , Jooeun Song<sup>4</sup> , Serene S. Paul <sup>4</sup> , Colleen G. Canning<sup>4</sup> and Jasmine C. C. Menant 2,3

1 Independent Researcher, São Carlos City Hall, São Carlos, Brazil, <sup>2</sup>Neuroscience Research Australia (NeuRA), Sydney, NSW, Australia, <sup>3</sup>School of Public Health & Community Medicine, University of New South Wales, Sydney, NSW, Australia, <sup>4</sup>Faculty of Health Sciences, The University of Sydney, Sydney, NSW, Australia

Background and Aim: The ability to adapt gait when negotiating unexpected hazards is crucial to maintain stability and avoid falling. This study investigated cognitive, physical and psychological factors associated with gait adaptability required for obstacle and stepping target negotiation in people with Parkinson's disease (PD).

#### Edited by:

Eric Yiou, Université Paris-Sud, France

#### Reviewed by:

Rahul Goel, Baylor College of Medicine, United States Arnaud Delafontaine, Staps Orsay, CIAMS Paris-Saclay, France

> \*Correspondence: Stephen R. Lord s.lord@neura.edu.au

Received: 02 April 2019 Accepted: 11 June 2019 Published: 28 June 2019

#### Citation:

Caetano MJD, Lord SR, Allen NE, Song J, Paul SS, Canning CG and Menant JCC (2019) Executive Functioning, Muscle Power and Reactive Balance Are Major Contributors to Gait Adaptability in People With Parkinson's Disease. Front. Aging Neurosci. 11:154. doi: 10.3389/fnagi.2019.00154 Methods: Fifty-four people with PD were instructed to either: (a) avoid an obstacle at usual step distance; or (b) step onto a target at either a short or long step distance projected on a walkway two heel strikes ahead and then continue walking. Participants also completed clinical [Hoehn & Yahr rating scale; Movement Disorders Society version of the Unified Parkinson's Disease Rating Scale motor section (MDS-UPDRS-III)], cognitive [simple reaction time, Trail Making and Stroop stepping (difference between incongruent and standard Choice Stepping Reaction Time, CSRT) tests], physical [hip abductor muscle power and reactive balance (pull test from the MDS-UPDRS-III)] and psychological (Fall Efficacy Scale–International) assessments.

Results: Discriminant function analysis revealed Stroop stepping test (inhibitory control) performance was the best predictor of stepping errors across the Gait Adaptability Test (GAT) conditions. Poorer executive function [Trail Making Test (TMT)] and reactive balance predicted poorer stepping accuracy in the short target condition; poorer reactive balance predicted increased number of steps taken to approach the obstacle and the long target; and poorer executive function predicted obstacle avoidance. Weaker hip abductor muscle power, poorer reactive balance, slower reaction time, poorer executive function and higher concern about falling were significant predictors of shorter step length while negotiating the obstacle/targets.

Conclusion: Superior executive function, effective reactive balance and good muscle power were associated with successful gait adaptability. Executive function and reactive balance appear particularly important for precise foot placements; and cognitive capacity for step length adjustments for avoiding obstacles. These findings suggest that impaired inhibitory control contributes to stepping errors and may increase fall risk in people with PD. These findings help elucidate mechanisms for why people with PD fall and may facilitate fall risk assessments and fall prevention strategies for this group.

Keywords: Parkinson's disease, gait adaptability, obstacle avoidance, cognition, choice stepping reaction time, stroop stepping test

# INTRODUCTION

The incidence of falls in people with Parkinson's disease (PD) is higher than in the healthy older population. Prospective studies indicate that between 45%–68% of people with PD fall at least once a year (Wood et al., 2002; Pickering et al., 2007; Latt et al., 2009; Paul et al., 2013; Lamont et al., 2017), with a large proportion (39%) falling recurrently (Allen et al., 2013). Most falls occur when people with PD are walking (Mak and Pang, 2010) and when they are optimally medicated (Gray and Hildebrand, 2000; Bloem et al., 2001; Lamont et al., 2017). It is possible that declines in the ability to adapt gait behavior, particularly under challenging environmental conditions contribute to trips; which are a frequently reported cause of falls in people with PD (Mak and Pang, 2010; Stack and Roberts, 2013; Gazibara et al., 2014).

Several studies have identified spatiotemporal gait alterations in people with PD walking at self-selected comfortable speed whilst optimally medicated. Compared with controls, people with PD walk slower with shorter stride length (Lewis et al., 2000; Sofuwa et al., 2005; Yang et al., 2008; Caetano et al., 2009) and slower cadence (Morris et al., 1994), present an increased double support duration (Caetano et al., 2009), more variable stride time (Hausdorff et al., 1998; Lord et al., 2013) and reduced foot clearance (Alcock et al., 2016). PD-related gait alterations interfere with the performance of daily activities, particularly in challenging conditions requiring modification of the walking pattern to deal with environmental changes or other task demands. Indeed, there is evidence that the ability to make gait adjustments in response to upcoming environmental changes is impaired in PD (Galna et al., 2010, 2013; Vitório et al., 2010, 2016; Stegemoller et al., 2012; Pieruccini-Faria et al., 2013; Geerse et al., 2018).

Several studies have used obstacle avoidance and stepping target tasks to assess gait adaptability in people with PD. This work has shown that compared with control participants, people with PD walk slower and take shorter steps throughout the approach, crossing and recovery steps of obstacle crossing (Galna et al., 2010; Vitório et al., 2010; Stegemoller et al., 2012; Pieruccini-Faria et al., 2013). Further, people with PD have reduced foot clearances (shorter vertical foot-obstacle distance), poorer balance control (increased speed and sway of the center of mass; Galna et al., 2013) during the obstacle crossing and place the lead foot closer to the obstacle after crossing it (Galna et al., 2010; Vitório et al., 2010; Stegemoller et al., 2012). People with PD also exhibit impaired foot placement accuracy in a walking task involving fixed stepping targets (Vitório et al., 2016; Geerse et al., 2018) and display lower obstacle-avoidance success rates and smaller obstacle-foot distances when adapting their walking to suddenly appearing obstacles (Geerse et al., 2018).

We recently designed a gait adaptability task to simulate the motor and cognitive challenges of daily walking activities. This task required people to either step onto a target or avoid an obstacle appearing at short notice while walking on a pathway in the laboratory. A decision-making component was incorporated into the walking task by using two stimulus colors (pink and green) that triggered different responses: a pink stimulus required an avoidance strategy (obstacle) whereas a green stimulus required a stepping strategy (target). We found that people with PD had more difficulty adapting their gait in response to targets (poorer stepping accuracy) and obstacles (increased number of steps) appearing at short notice on a walkway in comparison with healthy control participants (Caetano et al., 2018a). We noted such gait impairments were related to PD symptoms (Caetano et al., 2018a), but did not investigate whether gait adaptability deficits were also related to specific physical and cognition related impairments.

In the current study, we aimed to identify cognitive, physical and psychological factors associated with gait adaptability in people with PD. Seven variables related to falls in PD (Latt et al., 2009; Allen et al., 2010; Kerr et al., 2010; Paul et al., 2013, 2014) were examined: (i) freezing of gait; (ii) concern about falling; (iii) reactive balance (retropulsion test); (iv) lower limb muscle power; (v) simple reaction time; (vi) set-shifting of executive function; and (vii) inhibitory control of executive function. Our primary hypothesis was that cognitive capacities would discriminate between PD participants who do and do not make errors in the Gait Adaptability Test (GAT). Our secondary hypothesis was that a combination of cognitive and physical factors would predict stepping parameters in the experimental gait adaptability trials.

# MATERIALS AND METHODS

# Participants and Ethics Approval

The sample comprised 54 people with PD who were recruited from metropolitan Sydney, Australia through the research team's research volunteer databases and through Parkinson's NSW newsletters and support groups. PD volunteers were recruited for a training study (ACTRN12613000688785) and their data were collected as part of the baseline assessments. Participants were included if they were living in the community, able to walk unaided for ≥30 m and cognitively capable of following all instructions (MOCA scores ≥20; Nasreddine et al., 2005). Participants were required to have been on the same PD medication for at least 2 weeks. Volunteers were excluded if they had any medical conditions which would preclude or interfere with the physical assessment (e.g., physician diagnosed dementia, acute or terminal illness, progressive neurodegenerative diseases (other than PD), major psychiatric illnesses, color-blindness or visual impairments that could not be corrected). The University of Sydney Human Research Ethics Committee approved this study and all participants gave informed consent prior to study participation.

All measurements were conducted while participants were ''on'' their usual PD medication. Researchers experienced in working with people with PD and trained in the Movement Disorders Society Unified Parkinson's Disease Rating Scale (MDS-UPDRS) administered section III of the scale (motor examination; Goetz et al., 2012), the Hoehn & Yahr rating scale (H&Y; Hoehn and Yahr, 1967) and The New Freezing of Gait Questionnaire (NFOG-Q, part I: dichotomous item in which individuals were classified as a freezer or a non-freezer if they had experienced freezing of gait episodes during the past month; Nieuwboer et al., 2009). Psychological assessment regarding participants' concern about falling was determined with the Fall Efficacy Scale–International (Yardley et al., 2005).

# Protocol

Participants performed the GAT as well as brief cognitive and physical capacity assessments.

### Gait Adaptability Test (GAT)

Participants wore their own comfortable flat shoes and were required to walk at their self-selected speed over an obstacle-free path (baseline condition). They were then instructed about the GAT. As described elsewhere (Caetano et al., 2017), the GAT required participants to complete walking trials in four experimental conditions: (i) avoid stepping on a pink stimulus appearing two steps ahead (obstacle avoidance); (ii) stepping onto a green stimulus appearing slightly short of two steps ahead (short target); (iii) stepping onto a green stimulus appearing slightly further than two steps ahead (long target); and (iv) walking with no stimulus appearing on the pathway (walk-through). Walk-through trials were included to encourage participants to walk naturally. Trials were presented in a randomized order for a total of three trials per condition. At least one practice trial per condition (baseline, walk-through, obstacle avoidance, short target and long target) was performed until the task was understood, before data acquisition. Participants were offered rest breaks throughout the test.

The equipment and set-up have also been described in detail previously (Caetano et al., 2017). In brief, the targets and obstacle consisted of a colored light stimulus projected onto an area on the walkway (23 × 23 cm), presented on the third heel strike following gait initiation and appearing two steps ahead of the participant (**Figure 1**). Participants were instructed to step in the middle of the targets (green light) and to avoid stepping on the obstacle (pink light) but not step off the mat, using any avoidance strategy. Participants were asked to start walking with the right foot in all conditions. Distance to the obstacle/target was personalized for each individual. The starting position was adjusted to align the obstacle with the fifth-foot landing location based on the average foot placement from the baseline walking trials. During the experimental trials, obstacle and targets were presented on the third heel strike following gait initiation. In the trials where participants maintained their gait pattern similar to the baseline, the obstacle/targets appeared at two-step distance. However, it was a common behavior that participants adjusted their gait parameters after the appearance of the obstacle/targets. Considering our aims were to identify gait adaptation strategies toward suddenly appearing obstacles and targets, we consciously decided to not adjust the obstacle/target location for each experimental trial, but rather included the number of steps taken while approaching the obstacle/target as a dependent variable.

For the purpose of understanding stepping strategies, the step that hit or avoided the stimulus was named ''target/obstacle step'' and the preceding step was named ''previous step.'' Step length values were normalized by height of the participants. The average of the successful trials per condition was used in the analysis. Gait adaptability performance outcome measures were: (i) GAT errors—number of participants who made at least one error (stepping on an obstacle or missing a target); (ii) stepping accuracy for short and long target conditions (distance between the center of the target and the center of the foot); (iii) number of steps taken to approach the target or obstacle (during interval between the appearance of the stimulus and the target or obstacle step); and (iv) length of the two steps preceding the target or obstacle. An electronic walkway (4 m-long ZenoMetricsrmat/PKMAS software, v2011–2013, Havertown, PA, USA) recorded the temporal and spatial gait parameters. Position coordinates of the foot with reference to the target or obstacle coordinates extracted from the electronic walkway were used to determine GAT variables using a Matlab routine (MathWorks, Natick, MA, USA).

### Cognitive Assessments

Cognitive capacity was assessed using a test of simple reaction time (Lord et al., 2003), the Trail Making (Lezak et al., 2004) and the Stroop Stepping (Schoene et al., 2011, 2014) tests. For the assessment of simple reaction time (processing speed), participants were seated at a table and asked to press a button of a modified computer mouse using the index finger of their dominant hand as quickly as possible when a light stimulus appeared (Lord et al., 2003). Five practice trials were undertaken, followed by 10 experimental trials, with the average time of the experimental trials calculated in milliseconds.

The Trail Making Test (TMT) evaluates the cognitive flexibility/set-shifting of executive function [difference in execution time between parts B and A (TMT score)]. Participants were instructed to connect consecutive circled numbers for the TMT—part A and to connect numbers and letters in an alternating sequence for the TMT—part B, as quickly as possible without lifting the pen from the article. If participants made an error they were informed immediately and allowed to correct it. The total time to complete each part was measured in seconds with the test time capped at 5 min.

shoe and connected to a wireless transmitter attached to the participant's ankle triggered the light projection on the third heel strike following gait initiation (2).

Inhibitory control of executive function was measured with the Choice Stepping Reaction Time (CSRT) and the Stroop Stepping tests (Schoene et al., 2011, 2014) using a custom-made step mat (**Figure 2**). Descriptions of the apparatus, procedures and test-retest reliability for these tests are reported elsewhere (Schoene et al., 2011, 2014; Caetano et al., 2018a). In brief, the CSRT test assesses the ability to take a rapid step forward, backward or sidewards with either leg in response to randomly presented stimuli (Schoene et al., 2011). Six practice trials and 18 test trials were administered for this test. The Stroop stepping test combines stepping and response inhibition requiring rapid responses to incongruent stimuli (Schoene et al., 2014). A random sequence of four practice trials and 20 test trials in which the directions of word and orientation never matched was administered. In both tests, the average response time (i.e., stimulus presentation to step-on the target) was measured in milliseconds (ms). In the present study, the CSRT test was used as a proxy for a congruent Stroop Stepping test and the difference in execution time between both tests (Stroop stepping score) was used as a measure of inhibitory control.

#### Physical and Balance Assessments

Physical and balance assessments included tests of hip abductor muscle power and reactive balance. Hip abductor muscle power was measured in Watts for each leg using pneumatic variable resistance equipment (Keiser A420, Keiser Sports Health Equipment, Fresno, CA, USA). Muscle power was measured by having the participant abduct the hip as fast as possible against a low load (35 N—equivalent to 30% of the one repetition maximum on average in people with PD; Paul et al., 2012). Muscle power was recorded as the average of both legs.

Reactive balance was measured using the retropulsion test from the MDS-UPDRS Rating Scale (item 3.12; Goetz et al., 2012). The test examines the response to sudden body displacement produced by a quick, forceful pull on the shoulders while the patient is standing erect with eyes open and feet comfortably apart and parallel to each other. Performance was rated with a 0 (Normal: No problems, recovers with one or two steps), 1 (Slight: 3–5 steps, but subject recovers unaided), 2 (Mild: More than 5 steps, but subject recovers unaided), 3 (Moderate: Stands safely, but with absence of postural response; falls if not caught by examiner) or 4 (Severe: Very unstable, tends to lose balance spontaneously or with just a gentle pull on the shoulders).

# Statistical Analysis

Data normality was confirmed using the Skewness test. Discriminant function analysis was used to determine which cognitive, physical and psychological variables discriminated between the PD participants who made one or more mistakes in

the GAT from those who did not; Only those variables that were statistically and independently associated with gait adaptability errors were retained in the final model. Discriminant function analysis was chosen because our outcome was dichotomous and all the putative predictor variables were continuously scaled. Pearson's correlations were computed to examine associations between the cognitive, physical and psychological measures and the gait adaptability parameters. Stepwise linear regression analyses were then performed to identify independent and significant cognitive, physical and psychological explanatory variables for stepping accuracy in the short and long target conditions; the number of steps in the short target, long target and obstacle avoidance conditions and step length of the previous and target/obstacle steps for all conditions. The cognitive, physical and psychological variables that showed the strongest significant correlations with the gait adaptability measures were entered as standardized z-scores into the stepwise linear regression with a limitation of one predictor variable per 10 cases. All statistical analyses were performed using SPSS (Version 25 for Windows, SPSS Science, Chicago, IL, USA), with significance set at p < 0.05.

# RESULTS

# Gait Adaptability Performance

**Table 1** shows demographic, clinical, cognitive, physical and psychological measures for the sample, and **Table 2** presents participants' performance data for the gait adaptability variables TABLE 1 | Anthropometric, clinical, cognitive and physical characteristics of participants with Parkinson's disease (n = 54).


Data are mean (SD) or number (%). MoCA, Montreal Cognitive Assessment; MDS-UPDRS, Movement Disorders Society version of the Unified Parkinson's Disease Rating Scale; NFOG-Q, New freezing of gait questionnaire (part I, yes/no); FESI, Fall Efficacy Scale–International; TMT score: Trail Making Test, time difference between part B and part A; Stroop stepping score: time difference between Choice Stepping Reaction Time and Stroop Stepping tests. Note: higher scores in concern about falling, TMT, Stroop stepping, simple reaction time and reactive balance tests mean worse performance, while higher scores in MoCA and muscle power mean better performance.

for each test condition. Fourteen participants with PD (26%) made at least one error in the experimental conditions, totaling 18 incorrect responses of which 13 were commission errors (step on the obstacle) and five were omission errors (avoid the target—all in the short target condition).



Data presented are mean (SD) unless stated otherwise. <sup>a</sup>Number of participants (%) who made at least one mistake in the gait adaptability test.bDistance between the center of the target and the center of the foot; high values mean worse performance. <sup>c</sup>Number of steps taken to approach the target or obstacle (during interval between the appearance of the stimulus and the target or obstacle step). <sup>d</sup>Step that hit or avoided the stimulus was named "target/obstacle step" and the preceding step was named "previous step."

# Predictors of Impaired Gait Adaptability

Discriminant function analysis identified Stroop stepping performance as the only independent and significant predictor of stepping errors across the GAT conditions (Wilk's lambda: 0.822, p < 0.003; canonical correlation: 0.422).

Several cognitive, physical and psychological factors were weakly to moderately correlate with many gait adaptability performance measures (**Table 3**). The stepwise linear regression models revealed: (i) poorer reactive balance and executive function (TMT performance) were independent and significant predictors of poorer stepping accuracy in the short target condition; and (ii) poorer reactive balance (long target and obstacle avoidance conditions) and executive function (obstacle avoidance condition) were independent and significant predictors of increased number of steps taken to approach the target/obstacle (**Table 4**).

Further, independent and significant physical predictors of shorter step length were weaker hip abductor muscle power (baseline, walk-through and previous and target steps in the short target condition) and poorer reactive balance (walkthrough, previous step in the short/long target conditions and target step in the long target condition). Independent and significant cognitive predictors were slower simple reaction time (previous step in the long target condition), poorer Stroop stepping (target step in the short target condition and previous and obstacle steps in the obstacle avoidance condition) and TMT performances (previous step in the obstacle avoidance condition). Finally, greater concern about falling was an independent and significant predictor of shorter step length (baseline and previous step in the obstacle avoidance condition; **Table 4**).

# DISCUSSION

This study examined associations between cognitive, physical and psychological factors and gait adaptability in people with PD. Our hypotheses were supported by the findings of significant associations between successful gait adaptability and intact cognition as assessed with the trail making, Stroop stepping and simple reaction time tests in addition to better hip abductor muscle power and reactive balance.

# Cognitive Correlates

Inhibitory control of executive function (Stroop stepping score) was identified as the variable that best discriminated between participants who did and did not make mistakes (i.e., failing to hit the stepping targets or avoid the obstacle) in the GAT. Stroop stepping performance was also a predictor of target step length in the short target condition and the most powerful predictor for the obstacle avoidance condition (previous and obstacle step lengths). These results are in line with our previous study in healthy older adults (Caetano et al., 2017) and confirm the importance of inhibitory control of executive function for gait adaptability, particularly in people with PD. Our findings are also consistent with a previous study that found people with PD, compared with controls, make more errors in the incongruent trials of the finger-tapping Stroop task (Vandenbossche et al., 2012), and evidence that attentional control deficits lead to less effective behavioral responses in people with PD (Cools et al., 2010).

The identification of Stroop stepping performance as a predictor for both short target and obstacle avoidance conditions reflects the inhibitory component of the walking task that required participants to select the appropriate response while suppressing a dominant one (Caetano et al., 2017). Inhibitory control is an important discriminator between fallers and non-fallers in healthy older people (Anstey et al., 2009; Mirelman et al., 2012), and gait adaptability mistakes are more prevalent among older people at high risk of falling (Caetano et al., 2018b). Thus, our results suggest that impaired inhibitory control of executive function contributes to poor gait adaptability and consequently increased fall risk among people with PD.

Cognitive flexibility of executive function (TMT score) predicted stepping accuracy in the short target condition. Considering people with PD are more dependent on visual feedback to make target steps (Vitório et al., 2016), the short target condition appears to have been more challenging than the long target condition due to the shorter response time available to identify the target and plan and execute the appropriate gait adjustments. TMT performance was significantly associated with the number of steps taken to approach the obstacle and previous step length in the obstacle condition, suggesting that gait adjustments for obstacle avoidance require earlier and/or additional cognitive processing than stepping onto a target, consistent with findings in healthy older adults (Caetano et al., 2017).

Further, slow reaction time was significantly associated with shorter previous step length in the long target condition. Taken together, the above findings suggest that intact cognition is


TABLE 3 |Correlations coefficients among gait, clinical, cognitive, physical and psychological variables for each condition: baseline, walk-through, short target, long target and obstacle avoidance.

Note: higher scores in TMT, Stroop stepping, simple reaction time and reactive balance tests mean worse performance, while higher scores in muscle power mean better performance. Increased stepping accuracy values indicate poorer performance. aTrail Making Test, time difference between part B and part A. <sup>b</sup>Time difference between Choice Stepping Reaction Time and Stroop Stepping tests. cNew freezing of gait questionnaire (NFOG-Q, Part 1, yes/no). ∗Significant correlation (p<0.05),∗∗Significant correlation (p<0.01).


TABLE 4 | Stepwise linear regression models for predicting stepping accuracy in the short target and long target conditions, number of steps in the short target, long target and obstacle avoidance conditions and step length in the baseline, walk-through, short target, long target and obstacle avoidance conditions.

important for safe and precise foot placement for fall avoidance and may elucidate why reduced cognitive performance has been strongly associated with falls in people with PD (Paul et al., 2014).

# Muscle Power, Reactive Balance and Concern About Falling

We found reduced muscle power was associated with short step length in the baseline, walk-through and short target conditions; findings that build on previous work that has shown people with PD have shorter step lengths during usual (Sofuwa et al., 2005) and adaptive (Vitório et al., 2010) gait as well as reduced lower limb muscle strength and power than healthy controls (Allen et al., 2009). Robichaud et al posited that disease-related changes in the organization of the basal ganglia-thalamo-cortical circuit produce abnormal force generation patterns (shorter agonist burst duration and delayed antagonist activation; Robichaud et al., 2002) and subsequent shorter steps, and this may be exacerbated in a walking condition requiring gait adaptability. Thus, a multiple short step strategy in our gait adaptability protocol may have been adopted to compensate for motor deficits (Jankovic, 2008) or indicate difficulty in lengthening the step (Morris and Iansek, 1996).

Reactive balance performance was associated with several gait adaptability measures: stepping accuracy in the short target condition, number of steps taken to approach the target/obstacle and step length in the walk-through and short/long target conditions. This indicates balance control plays an important role in gait adaptability, particularly for accurate foot placement in a walking task requiring stepping adjustments. Our findings corroborate previous work that has demonstrated the importance of adequate postural adjustments during gait initiation with the goal to clear an obstacle in young adults (Yiou et al., 2016a,b). Previous work has also shown people with PD display inappropriate postural adjustments (reduced anteroposterior center of mass motion and smaller distance between the center of pressure and center of mass) during obstacle crossing (Stegemoller et al., 2012) and prior to step initiation when performing a concomitant attentional task (Tard et al., 2014).

Our finding that reactive balance predicts step length in the walk-through condition suggests participants whose reactive balance is compromised adopt a cautious walking pattern in a cognitively demanding situation that may require step adjustments to negotiate hazards (Caetano et al., 2017). Furthermore, a higher concern about falling was associated with many gait adaptability parameters in the univariate analyses and was an independent predictor of step length at baseline and previous step in the obstacle condition which suggests psychological factors also influence gait performance in situations involving environmental hazards (van Schooten et al., 2019).

Finally, it is surprising that overall freezing of gait was generally not related to the gait adaptability parameters. Participants were tested ''on'' their usual PD medication thus they are less likely to demonstrate freezing of gait. Also, the task was highly attention demanding with some predictability about when the target or obstacle would appear. People with freezing of gait are known to perform better when external cueing are provided (Ginis et al., 2018), thus the GAT may have worked as a visual cue to the freezers.

# Practical Implications

Our findings demonstrate the importance of considering the context under which walking performance is assessed. A complex interplay of sensorimotor and cognitive abilities are likely required for people with PD to meet everyday challenges associated with walking, such as crossing streets, moving in crowds, etc. Thus, the GAT provides a novel way to explore sensorimotor and cognitive mechanisms involved in complex walking tasks. Our findings also suggest that impaired inhibitory control contributes to stepping errors and may increase fall risk in people with PD. This information may elucidate mechanism as to why people with PD fall and facilitate fall risk assessments and fall prevention strategies for this group. These strategies could involve adding decision-making tasks (e.g., obstacle avoidance) to walking adaptability training. Future studies could also investigate whether gait adaptability measures are associated with prospective falls and whether rehabilitation interventions aimed at improving gait adaptability can prevent falls in people with PD.

# Limitations

We acknowledge certain study limitations. First, our study sample was relatively small, and we conducted multiple comparisons relating to the different outcome measures. It is possible, therefore, that some of the associations uncovered are due to chance. However, similar associations were evident between step length and particular gait adaptability parameters, and are also in line with previous findings in healthy older people (Caetano et al., 2017). Second, the calculation of the Stroop stepping test score used a standard version of the CSRT and not a congruent word/arrow direction as the simpler test condition. It may therefore not represent a ''pure'' measure of inhibitory control of stepping.

# CONCLUSION

In conclusion, superior executive function, effective reactive balance and good muscle power were associated with successful

# REFERENCES


gait adaptability in people with PD. Executive function and reactive balance appear particularly important for precise foot placements, and cognitive capacity for step length adjustments for avoiding obstacles.

# 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 University of Sydney Human Research Ethics Committee 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 Human Research Ethics Committee at the University of Sydney (Project Number: 2013/207).

# AUTHOR CONTRIBUTIONS

MC, SL and JM conceived the study objectives and designed the study. MC, NA, JS, SP and CC acquired the data. MC, SL and JM analyzed and interpreted the data. MC drafted the manuscript. All authors were involved with preparation of the manuscript.

# FUNDING

This work was supported by the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq —200748/2012-2) to MC; National Health and Medical Research Council to SL; Parkinson's New South Wales and The University of Canberra grant to NA.

# ACKNOWLEDGMENTS

We would like to acknowledge Matthew Brodie for developing the MatLab code to process the raw data. We also thank all participants for their voluntary participation.


gait adaptability and fall risk in older people. Gait Posture 59, 188–192. doi: 10.1016/j.gaitpost.2017.10.017


a five-year prospective study link fall risk to cognition. PLoS One 7:e40297. doi: 10.1371/journal.pone.0040297


of different height and distance made under reaction-time and self-initiated instructions. Front. Hum. Neurosci. 10:449. doi: 10.3389/fnhum.2016.00449

Yiou, E., Fourcade, P., Artico, R., and Caderby, T. (2016b). Influence of temporal pressure constraint on the biomechanical organization of gait initiation made with or without an obstacle to clear. Biomaterials 234, 1363–1375. doi: 10.1007/s00221-015-4319-4

**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 Caetano, Lord, Allen, Song, Paul, Canning and Menant. 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.

# Prefrontal Cortex Activation During Dual Task With Increasing Cognitive Load in Subacute Stroke Patients: A Pilot Study

Eric Hermand<sup>1</sup> , Bertrand Tapie1,2, Olivier Dupuy <sup>3</sup> , Sarah Fraser <sup>4</sup> , Maxence Compagnat 1,2 , Jean Yves Salle1,2, Jean Christophe Daviet 1,2 and Anaick Perrochon<sup>1</sup> \*

<sup>1</sup> Laboratoire HAVAE, EA6310, Université de Limoges, Limoges, France, <sup>2</sup> Médecine Physique et de Réadaptation, Centre Hospitalier Universitaire, Limoges, France, <sup>3</sup> Laboratoire Move, EA6314, Poitiers University, Poitiers, France, <sup>4</sup> Faculty of Health Sciences, Interdisciplinary School of Health Sciences, University of Ottawa, Ottawa, ON, Canada

Stroke patients often exhibit difficulties performing a cognitive task while walking, defined as a dual task (DT). Their prefrontal cortex (PFC) activity is higher in DT than in single task (ST). The effects of an increasing load on PFC activity during DT in subacute stroke patients remains unexplored. Our objective was to assess the effects of N-back tasks (low/high load) on cerebral activity, gait parameters, and cognitive performances. Eleven subacute stroke patients (days post-stroke 45.8 ± 31.6) participated in this pilot study (71.4 ± 10 years, BMI 26.7 ± 4.8 kg.m−<sup>2</sup> , Barthel index 81.8 ± 11.0). Patients completed a STwalk, and 4 conditions with 1-back (low load) and 2-back (high load): STlow, SThigh, DTlow, and DThigh. Overground walking was performed at a comfortable pace and -N-back conditions were carried out verbally. Both gait (speed, stride variability) and cognitive (rate of correct answers) performances were recorded. Changes in PFC oxyhemoglobin (1O2Hb) and deoxyhemoglobin (1HHb) were measured by functional near infrared spectroscopy (fNIRS). Results showed an increase of 1O2Hb while walking, which was not augmented by cognitive loads in DT. Walking speed was reduced by low and high cognitive loads in DT compared to STwalk (P < 0.05), but was not different between DTlow and DThigh. Cognitive performances were negatively impacted by both walking (P < 0.05) and cognitive load (between "low" and "high," P < 0.001). These data highlight a "ceiling" effect in 1O2Hb levels while walking, leaving no available resources for simultaneous cognitive tasks, during the early recovery period following stroke. In these patients, cognitive, but not motor, performances declined with a higher cognitive load.

Keywords: functional near-infrared spectroscopy, prefrontal cortex, stroke, dual task, gait, cognition

# INTRODUCTION

Stroke patients with brain lesions may exhibit impaired cognitive functions, altered walking capacity, or both (Grotta, 2016). These declines in cognitive and walking performances are accentuated when they are performed simultaneously in dual task (DT) such as walking while performing a cognitive task (Al-Yahya et al., 2011), which illustrates a cognitive-motor interference in these patients (Plummer et al., 2013).

#### Edited by:

Helena Blumen, Albert Einstein College of Medicine, United States

#### Reviewed by:

Laura Lorenzo-López, University of a Coruña, Spain Jennifer Yuan, New York University, United States

#### \*Correspondence: Anaick Perrochon

anaick.perrochon@unilim.fr

Received: 20 March 2019 Accepted: 12 June 2019 Published: 02 July 2019

#### Citation:

Hermand E, Tapie B, Dupuy O, Fraser S, Compagnat M, Salle JY, Daviet JC and Perrochon A (2019) Prefrontal Cortex Activation During Dual Task With Increasing Cognitive Load in Subacute Stroke Patients: A Pilot Study.

> Front. Aging Neurosci. 11:160. doi: 10.3389/fnagi.2019.00160

It is not possible yet to study gait directly in a scanner environment, such as functional magnetic resonance imagery (fMRI), although a few studies proposed that the imaging of ankle dorsi-flexion, a component movement of gait, may provide a useful marker for gait recovery (Johansen-Berg, 2007). Functional near infrared spectroscopy (fNIRS) has proven to be an effective tool for acquiring brain activity during human walking (Perrey, 2014) and in DT (Holtzer et al., 2014; Leone et al., 2017). fNIRS studies that have assessed changes in oxygenated hemoglobin (1O2Hb) levels in stroke patients, have reported greater changes in the prefrontal cortex (PFC) during DT than in single task (ST) (Al-Yahya et al., 2016; Hawkins et al., 2018; Liu et al., 2018). This change in PFC activation is similar to what has been observed with fMRI in stroke patients performing simulated motor activities (Al-Yahya et al., 2016) and to healthy older adults performing a dual-task in the fMRI scanner (Erickson et al., 2007). This implies that walking in DT requires additional attentional resources. The observed decline in performances during DT are greater in cognition than in gait, suggesting that stroke patients prioritize walking over cognition during DT, contrary to age-matched controls (Mori et al., 2018). These studies included only chronic stroke patients (Al-Yahya et al., 2016; Hawkins et al., 2018; Liu et al., 2018; Mori et al., 2018), in which rehabilitation potentially allowed a recovering of equilibrium reflexes and stepping to a greater extent than patients in subacute phase (Stinear, 2017). Moreover, the effects of DT walking with increasing cognitive load have demonstrated decrements in cognitive performances in healthy older adults (Fraser et al., 2016) but this has yet to be investigated in stroke patients.

In this study, our main objective was to investigate the effects of increasing cognitive load on bilateral, affected and unaffected PFC activation during DT, in subacute stroke patients. The second objective was to assess the cost of DT on gait and cognitive performances and their associations with PFC activation.

# PARTICIPANTS AND METHODS

### Participants

Eleven subacute stroke patients (6 men, 5 women) participated in this pilot study, in the center of Physical Medicine and Rehabilitation (University Hospital, Limoges). Inclusion criteria for post-stroke patients included: acute (<2 weeks after stroke) or early subacute stroke (between 2 weeks and 3 months) (HAS, 2012; Ammann et al., 2014), first stroke located in left or right middle cerebral artery and being able to walk 10 meters with or without assistance (orthotics, crutch) and corrected hearing/vision. Exclusion criteria included previous neurological disease limiting gait, aphasia, pre-existing cognitive disorders (including mild cognitive dementia, Alzheimer and Parkinson diseases), cardiovascular or pulmonary diseases. Functional ambulation category (Holden et al., 1984) for each patient was evaluated on test day, from 0 (non-walking) to 5 (walking alone, stairs included) (FAC, **Table 1**). Level of education assessed according to the International Standard Classification of Education (Schneider, 2013), from 0 (pre-primary) to 8 (Ph.D. or equivalent).

The study was approved by national ethic committee (CPP, registration number 2017-A01883-50) and patients gave their informed consent.

# Design Protocol

Patients performed 3 randomly ordered phases successively: cognitive single tasks (i.e., STlow and SThigh), walking single task (STwalk) and DT including simultaneous cognitive and walking conditions (i.e., DTlow and DThigh). Cognitive tasks for ST and DT followed a modified N-back from Fraser et al. (2016) in which the stimuli were presented aurally by the experimenter: "low" was associated with the 1-back condition and "high" with the 2 back condition. The "low" condition was performed before "high" condition, separated by a 4–5 min rest. During the N-back test, the experimenter, facing the patient at a distance of 1 m during ST or walking 1 m behind him/her during DT, read aloud and clearly a series of 20 fixed random numbers, between 0 and 10, evenly spaced in a 30-s interval. Responses were recorded with a voice recorder. One practice trial for each cognitive task was conducted prior to experimental testing to ensure proper hearing/vision and a good understanding of each task. In walking conditions, patients walked in an open space at a comfortable pace for 30 s and in DT they were asked to focus equally on walking and cognitive tasks.

# Gait and Cognitive Performances

Patients performed walking conditions (i.e., ST and DT) on an 8-m GAITRite walkway (GAITRite <sup>R</sup> - CIR Systems, Inc., Sparta, NJ, USA), which provided spatio-temporal gait parameters, such as speed ( −→V , cm.s−<sup>1</sup> ), stride variability (tVar, n.u.), and stride asymmetry ( length of left stride length of right stride , n.u.). In cognitive tasks (STlow, SThigh, DTlow, and DThigh), the percentage of correct answers was compiled for each condition, as missing or incorrect answers were accounted for as errors. The DT costs on cognition and gait were calculated by:

Cognitive costs:

$$\begin{aligned} \csc DT\_{\text{low}} &= \frac{\text{Score } DT\_{\text{low}} - \text{Score } ST\_{\text{low}}}{\text{Score } ST\_{\text{low}}} \; and \\ \csc DT\_{\text{high}} &= \frac{\text{Score } DT\_{\text{high}} - \text{Score } ST\_{\text{high}}}{\text{Score } ST\_{\text{high}}} \end{aligned}$$

where Score DTlow/high is the cognitive performance (percentage of correct answers); and

Gait costs:

$$\text{cgDT}\_{\text{low}} = \frac{\left\| \begin{array}{l} \overrightarrow{V}\_{\text{DT}\_{\text{low}}} \; \left\| \begin{array}{l} - \\ \end{array} \right\| \overrightarrow{V}\_{\text{ST}\_{\text{walk}}} \; \right\|}{\left\| \overrightarrow{V}\_{\text{ST}\_{\text{walk}}} \; \right\|} \; \text{and} \\\\ \text{cgDT}\_{\text{high}} = \frac{\left\| \begin{array}{l} \overrightarrow{V}\_{\text{DT}\_{\text{high}}} \; \left\| \begin{array}{l} - \\ \end{array} \right\| \overrightarrow{V}\_{\text{ST}\_{\text{walk}}} \; \end{array} \right\|}{\left\| \overrightarrow{V}\_{\text{ST}\_{\text{walk}}} \; \right\|} \end{array}$$


TABLE 1 | Clinical characteristics of patients.

I, Ischemic; H, Hemorrhagic; R, Right; L, Left.

The cognitive and gait costs from DTlow to DThigh are, respectively given by the equations:

### fNIRS Acquisition and Analysis

Cerebral oxygenation was measured using a fNIRS system (Portalite, Artinis Medical, Netherlands). Two optodes were placed on symmetrical prefrontal sites Fp1 and Fp2 according to the EEG 10/20 system. Acquisition was made through the Oxysoft software (version 3.0.97.1). Differential Pathlength Factor was set on 5 as its calculation formula does not apply to patients' age 50 years and older (Duncan et al., 1996). In each condition, after a 30 s rest for baseline, patients performed the 30 s test, before a final 30 s rest phase. A 0.1 Hz low-pass filter was applied to the fNIRS signal to remove physiological and instrumental noise, and motion artifacts were corrected using Matlab-based scripts when needed (Fishburn et al., 2019). The relative concentrations in O2Hb and HHb (1O2Hb and 1HHb, respectively, µmol.L−<sup>1</sup> ) in the test interval (i.e., the last 20 s) were then normalized by subtracting to them the mean value of the last 10 s of baseline, immediately before the beginning of the task, i.e., seated for STlow and SThigh, and standing for STwalk, DTlow, and DThigh. From these data were extracted the hemoglobin difference (1HbDiff = 1O2Hb − 1HHb) and the laterality index (LI) defined as the ratio:

$$LI = \frac{\Delta O\_2Hb \text{ (affected hemisphere)} - \Delta O\_2Hb \text{ (щаб }h\text{-misphere)}}{\Delta O\_2Hb \text{ (affected hemisphere)} + \Delta O\_2Hb \text{ (щаб }h\text{-misphere)}}.$$

### Statistical Analysis

A Shapiro-Wilk test confirmed the non-normality of data. Friedman and Wilcoxon tests were then conducted to compare and assess the respective effects of walking (i.e., sit:STlow and SThigh vs. walk: STwalk, DTlow, and DThigh) and cognitive load (i.e., none: STwalk vs. low: STlow and DTlow vs. high: SThigh and DThigh) on cerebral activity (1O2Hb, 1HHb, 1HbDiff) and gait parameters (speed, stride variability). A Spearman correlation test was then conducted to establish potential correlations between PFC activity and gait/cognitive performances. The statistical significance was set at P < 0.05.

# RESULTS

Individual patients' characteristics and gait/cognitive performances are presented on **Tables 1**, **2**, respectively.

# PFC Activation and Its Correlations With Gait/Cognitive Performance

There was a main effect of walking with an increase of 1O2Hb (P < 0.01, **Figure 1A**) and 1HbDiff (P < 0.05, **Figure 1B**) in bilateral PFC, but there was no difference between the different walking conditions (STwalk vs. DTlow vs. DThigh, P > 0.05) or between cognitive conditions. No effect of cognitive load was observed for other oxygenation parameters, such as 1HHb.

Taken separately, we observed a similar increase of 1O2Hb in the respective affected and unaffected hemispheres (P < 0.01) whereas LI was not modified (**Figure 1C**).

Finally, there were no significant correlations between PFC activation (for total, affected or unaffected hemispheres) and gait/cognitive performances (raw values or DT costs, P > 0.05).

# ST vs. DT

Speed decreased and gait variability increased in DTlow and in DThigh compared to STwalk (P < 0.05, respectively **Figures 1D,E**). Walking did not affect the percentage of correct answers (**Figure 1F**).

The gait cost of cognitive load in "high" condition (i.e., cgDThigh) was superior to cgDTlow (respectively −19 ± 18% and −15 ± 19%) but walking did not significantly influence the cognitive cost (i.e., ccDThigh and ccDTlow).

#### TABLE 2 | Gait and cognitive parameters and costs in percentage.


STwalk → DThigh −19 ± 18\*% DTlow → DThigh

$$-6 \pm 14\%$$

p significant (0.05) is presented in bold print\*.

# DTlow vs. DThigh

No difference in speed or stride variability were found between DTlow and DThigh, Cognitive performances were negatively impacted by cognitive load (percentage of correct answers, P < 0.01, **Figure 1F**).

The cognitive cost of heavier cognitive load during DT (i.e., ccDTlow→high) was −58 ± 31% (P < 0.01) but cgDTlow→high was negligible.

# DISCUSSION

To our knowledge, this is the first study on DT with an increasing cognitive load in subacute stroke patients. Our first finding relates to the resources required for walking in these patients: 1O2Hb is drastically increased during walking and is not further augmented by any additional cognitive load, low or high. In parallel, we observed a decline in gait performances in DT compared to ST, but no difference between the two DT conditions. As for the cognitive performances, the percentage of correct answers was not decreased by locomotion during DT compared to ST, but was negatively impacted by a higher cognitive load.

Our findings are similar to previous work (Hawkins et al., 2018) in which cerebral oxygenation reached a "ceiling" in chronic stroke patients while walking in ST, leaving no attentional resources for other simultaneous cognitive tasks. This cerebral overactivation triggered by a simple motor task such as walking that PFC activation is consistent with previous work reporting a prioritization of resources on locomotor requirements over cognitive ones in stroke patients, contrary to healthy participants (Mori et al., 2018). Unlike other studies (Al-Yahya et al., 2016; Liu et al., 2018), the cognitive load did not significantly impact cerebral oxygenation (**Figure 1A**). We assume that this difference could be explained by the difference between subacute and chronic stroke patients. The latter may have recovered a substantial proportion of their walking abilities (i.e., walking speed), therefore leaving some unused cerebral resources for other simultaneous cognitive tasks. Subacute stroke patients, in the early period following stroke, may still be in the cerebral recovery process to regain gait automaticity (Skilbeck et al., 1983), and therefore allocate most of available resources to locomotor needs.

Our data could not highlight a compensatory cerebral reorganization between ipsi- and contralesional hemispheres in ST or DT during the subacute phase: when an increase of O2Hb was observed in STwalk or DT, as observed in fMRI research (Al-Yahya et al., 2016), PFC activities were similarly augmented in both affected and unaffected hemispheres, and LI was therefore not modified. These results are consistent with observations in chronic stroke patients (Hawkins et al., 2018), whereas other works show different activities between hemispheres in various motor or cognitive conditions (Liu et al., 2018; Mori et al., 2018). In particular, Liu et al. showed the key role of other brain areas, such as the correlation between unaffected supplementary motor areas (SMA) and gait parameters during DT, in patients with less motor abilities (Liu et al., 2018). Similarly, Mori et al. showed a correlation between motor DT cost and right PFC activation in stroke patients but did not find analogous correlation between brain activation and cognitive performance (Mori et al., 2018). However, they performed a post-analysis showing that ipsilesional PFC activation was also negatively correlated with gait DT cost, implying a prioritization of locomotion over cognition by the affected hemisphere, which we could not support with our data.

The phenomenon of overactivation may be put into perspective with changes in gait parameters: in DT, the simultaneous cognitive tasks were demanded to an already overactivated system, therefore the speed decrease between "none" and "low"/"high" cognitive load was expected (**Figure 1D**). This is illustrated by the gait cost in "high" condition (cgDThigh = −19%), as observed in other studies in stroke patients (Hawkins et al., 2018; Liu et al., 2018) and older participants (Hawkins et al., 2018). However, we did not notice any further gait impairment between "low" and "high" cognitive loads (DTlow vs. DThigh, **Figure 1D**). This would imply, contrary to 1O2Hb ceiling, the potential existence of a minimal walking speed in participants or patients submitted to a "saturation" of brain activation. Further studies including multiple gradual cognitive loads should be conducted to test this hypothesis.

This is the first study to examine the relationship between brain overactivation and cognitive performance, from "low" to "high" condition during DT in stroke patients. In our study, as expected, cognitive performances were diminished by a higher load during DT, as was previously demonstrated in dual-task walking research with younger and older healthy participants (Fraser et al., 2016). We did not find any correlation between 1O2Hb and cognitive performances. Research with younger participants, found negative correlations between 1O2Hb and cognitive performance during DT, but not ST (Mirelman et al., 2014). This further illustrates the remaining attentional resources in healthy participants (Hawkins et al., 2018), although the reduction in cognitive performances remains greater in older adults (Voelcker-Rehage et al., 2006).

The very limited number of patients included in this study was due to the difficulty of performing both walking and cognitive tasks in subacute phase. In parallel, it is important to note that fNIRS studies during walking show technical and methodological risks which can modify results in study findings if they are not controlled (Vitorio et al., 2017). While the current study used filtering techniques to minimize physiological noise and motion artifact in the signal, we could not control for potential contributions from the scalp to cerebral oxygenation, as this would require a multichannel device with short separation channels. Future research should test the same protocol with short separation channels in order to minimize contributions from the scalp effectively. Also, further studies will need to include a larger sample size and an age-matched control population to confirm the present findings. In addition, existing imaging studies (Johansen-Berg, 2007) have demonstrated the involvement of motor and other areas of the brain that would likely also exhibit changes during STwalk and DT: future studies could provide more information on cerebral changes during dual-task walking with stroke by including the level of cerebral O<sup>2</sup> in additional regions (i.e., premotor cortex, SMA, primary motor cortex).

In conclusion, our study on subacute stroke patients highlights a "ceiling" in PFC oxygenation that is already reached during walking, requiring most of attentional resources in the early stages after stroke. This partly confirms previous findings that demonstrate large decrements in gait and cognitive performances during DT, regardless of cognitive load, which were not associated with changes in cerebral oxygenation in the PFC.

# DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the supplementary files.

# REFERENCES


# ETHICS STATEMENT

This study was carried out in accordance with the recommendations of national ethic committee (CPP, registration number 2017-A01883-50) with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki.

# AUTHOR CONTRIBUTIONS

AP and JD: study concept and design. BT and EH: acquisition of data. EH, OD, and AP: analysis and interpretation of data. EH, AP, and JD: drafting of the manuscript. SF, MC, and JS: critical revision of the manuscript for important intellectual content.


**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 Hermand, Tapie, Dupuy, Fraser, Compagnat, Salle, Daviet and Perrochon. 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.

# Perturbation-Induced Stepping Post-stroke: A Pilot Study Demonstrating Altered Strategies of Both Legs

Katherine M. Martinez <sup>1</sup> \*, Mark W. Rogers <sup>2</sup> , Mary T. Blackinton<sup>3</sup> , M. Samuel Cheng<sup>4</sup> and Marie-Laure Mille1,5,6

<sup>1</sup> Department of Physical Therapy and Human Movement Sciences, Northwestern University Feinberg School of Medicine, Chicago, IL, United States, <sup>2</sup> Department of Physical Therapy and Rehabilitation Science, University of Maryland School of Medicine, Baltimore, MD, United States, <sup>3</sup> Physical Therapy Program, Nova Southeastern University, Tampa, FL, United States, <sup>4</sup> Physical Therapy Program, Nova Southeastern University, Fort-Lauderdale, FL, United States, <sup>5</sup> UFR-STAPS, Université de Toulon, La Garde, France, <sup>6</sup> Institut des Sciences du Mouvement (ISM UMR 7287), Aix Marseille Université and CNRS, Marseille, France

#### Edited by:

Helena Blumen, Albert Einstein College of Medicine, United States

#### Reviewed by:

Paolo Tonin, Sant'Anna Institute, Italy Domenico Antonio Restivo, Ospedale Garibaldi, Italy

\*Correspondence: Katherine M. Martinez k-martinez@northwestern.edu

#### Specialty section:

This article was submitted to Neurorehabilitation, a section of the journal Frontiers in Neurology

Received: 28 January 2019 Accepted: 17 June 2019 Published: 03 July 2019

#### Citation:

Martinez KM, Rogers MW, Blackinton MT, Cheng MS and Mille M-L (2019) Perturbation-Induced Stepping Post-stroke: A Pilot Study Demonstrating Altered Strategies of Both Legs. Front. Neurol. 10:711. doi: 10.3389/fneur.2019.00711 Introduction: Asymmetrical sensorimotor function after stroke creates unique challenges for bipedal tasks such as walking or perturbation-induced reactive stepping. Preference for initiating steps with the less-involved (preferred) leg after a perturbation has been reported with limited information on the stepping response of the more-involved (non-preferred) leg. Understanding the capacity of both legs to respond to a perturbation would enhance the design of future treatment approaches. This pilot study investigated the difference in perturbation-induced stepping between legs in stroke participant and non-impaired controls. We hypothesized that stepping performance will be different between groups as well as between legs for post-stroke participants.

Methods: Thirty-six participants (20 persons post-stroke, 16 age matched controls) were given an anterior perturbation from three stance positions: symmetrical (SS), preferred asymmetrical (PAS−70% body weight on the preferred leg), and non-preferred asymmetrical (N-PAS−70% body weight on the non-preferred leg). Kinematic and kinetic data were collected to measure anticipatory postural adjustment (APA), characteristics of the first step (onset, length, height, duration), number of steps, and velocity of the body at heel strike. Group differences were tested using the Mann-Whitney U-test and differences between legs tested using the Wilcoxon signed-rank test with an alpha level of 0.05.

Results: Stepping with the more-involved leg increased from 11.5% of trials in SS and N-PAS up to 46% in PAS stance position for participants post-stroke. Post-stroke participants had an earlier APA and always took more steps than controls to regain balance. However, differences between post-stroke and control participants were mainly found when stance position was modified. Compare to controls, steps with the preferred leg (N-PAS) were earlier and shorter (in time and length), whereas steps with the non-preferred leg (PAS) were also shorter but took longer. For post-stroke participants, step duration was longer and utilized more steps when stepping with the more-involved leg compared to the less-involved leg.

Conclusions: Stepping with the more-involved leg can be facilitated by unweighting the leg. The differences between groups, and legs in post-stroke participants illustrate the simultaneous bipedal role (support and stepping) both legs have in reactive stepping and should be considered for reactive balance training.

Keywords: postural control, reactive balance, compensatory stepping, rehabilitation, stroke

# INTRODUCTION

Sensorimotor dysfunction is a common outcome after a stroke that contributes to deficits in gait and voluntary stepping as well as impaired postural control (1–6). Stroke survivors initiate gait predominantly with their more-involved leg and display asymmetrical cadence and step kinematics (7–9). When balance is challenged, stepping is a common strategy used to maintain upright control (10, 11). Previous studies have shown that older adults rely more on stepping to recover their balance and this induced stepping strategy is less effective for those who have diminished sensorimotor function in their lower limbs (11–14). Dynamic balance is also impaired in stroke survivors compared to controls as seen in number of falls, greater sway and altered ground reaction forces after a lateral perturbation and differences during automatic and voluntary weight shifts (15–17). Whether stepping voluntarily as in gait or reactively in response to a perturbation, the body must coordinate the use of both lower limbs, one for support and one for stepping. When presented with an asymmetrical impairment of the lower limb after stroke, the question becomes which leg should be more effectively and safely used for support vs. stepping and whether it possible to induce the use of a specific leg?

Several studies on perturbation stepping in persons poststroke report a predilection for initiating steps with the lessinvolved leg, even with lateral perturbations (18–22). Two studies involving steps with the paretic (more-involved) leg with cuing found no difference in step onset or number of steps between paretic and non-paretic (less-involved) legs during these cued reactive stepping tasks (23, 24). Our previous study using a forward-diagonal perturbation method to induce stepping with paretic leg in stroke survivors found slower and earlier steps for the paretic leg and that induced steps were faster and earlier compared to voluntary stepping (20). However, studies on reactive stepping report no difference in the incidence of falls when initiating a step with paretic or non-paretic leg (19, 21, 23).

The asymmetry in stepping preference and sensorimotor function creates challenges in investigating the bipedal nature of induced reactive stepping. Comparing perturbation induced reactive stepping responses between individuals' post-stroke and controls would provide some insight into the alterations in inter-limb control of reactive stepping. To our knowledge, only one study has compared persons post-stroke and controls and only during a posterior perturbation. In that study, poststroke participants initiated fewer backward steps, took more steps when they did initiate stepping, and were less stable with shorter step length and delayed onset compared to controls (25). However, the extent to which these changes in stepping performance affect balance recovery after stroke for other directions of stepping remains to be determined.

Differences between the inter-limb control of stepping for different directions of imbalance in unimpaired controls and persons post-stroke are important to identify given the bipedal nature of the induced stepping task and the direction-dependent neuromotor requirements for recovering balance (13). Incomplete understanding of the inter-limb step characteristics during induced stepping after stroke limits the design of effective treatment interventions to improve balance function and reactive postural control. For example, additional information about the use of the more-involved and less-involved legs prior to and during stepping as well as after step landing will help delineate the effectiveness of each leg in the different phases of balance recovery.

To further address the foregoing issues, the purpose of this pilot study was to investigate potential differences in reactive perturbation-induced stepping characteristics between: (1) post-stroke survivors and healthy control participants, and (2) between legs in the post-stroke participants. Using a simple method of modifying the initial stance symmetry, we hypothesized that for perturbation induced reactive steps in the anterior direction: (1) stance asymmetry will alter the selection of the stepping leg, (2) stepping performance would be different between post-stroke participants and age-matched controls for both legs, and (3) the more-involved leg would be less efficient in reactive step performance than the less-involved leg.

# METHODS

Data sharing is fully appreciated. The raw data supporting the conclusions of this manuscript will be made available by the corresponding author, without undue reservation, to any qualified researcher upon reasonable request.

# Study Population

Thirty-six individuals (20 persons post-stroke and 16 controls) participated in this study. The post-stroke participants were recruited from the Clinical Neuroscience Research Registry. They all met the following criteria: history of a single unilateral non-cerebellar stroke at least 1 year prior to the study, ambulated independently in the community and did not use a wheelchair or a long leg brace. Additional selection criteria included: no history of other neurological and orthopedic diseases or surgery to the lower extremities, and not currently receiving occupational or physical therapy. In addition, post-stroke participants were physically screened for the ability to stand and walk independently 10 feet without an assistive device or orthosis, unilateral paresis, ability to give consent. They were excluded if they scored below a five on the Functional Ambulation Categories (26, 27). Controls were recruited from the Northwestern University Buehler Center Aging Research Registry and from flyers posted on campus. They were age (± 3years) matched without neurological or orthopedic impairments. All participants gave written inform consent for the study which was approved by the Institutional Review Board of Northwestern University and Nova Southeastern University and performed in agreement with the ethical principles of the Declaration of Helsinki. Participant demographics characteristics are presented in **Table 1**.

# Protocol

# Clinical Assessments

The clinical tests were selected to give a fuller picture of the participants' characteristics (**Table 1**). The Activity-specific Balance Confidence scale (ABC) questionnaire was completed to assess balance confidence in performing 16 functional activities using a rating of 0% (no confidence at all) to 100% (complete confidence) (28–30). Scores above 80 indicated high balance confidence. The Timed Up and Go test (TUG) was used to assess balance and ambulation mobility (31, 32). Participants began seated and walk three meters at their normal pace, turn around, walk back, and sat down. The Step Test (ST) assesses stance stability and dynamic balance in persons post-stroke (33, 34). The participants started in standing and place their whole foot up on a step 7.5 centimeters high as many times as they could in 15 s without loss of balance, first with their more-involved leg and then with their less-involved leg. The score for each leg was the number of times the participant touches the step in 15 s.

Sensorimotor measures included sensation on the plantar surface of feel and extension control in standing. Deep pressure sensation in the feet were measured using Semmes-Weinstein aesthesiometer (35, 36). The aesthesiometer filaments were placed perpendicular to the plantar surface of the first ray. The lowest gram filament perceived when touched to the foot was recorded. Testing strength in persons post-stroke is complicated due to the presence of synergy patterns, altered motor recruitment patterns, and spasticity in some patients. The Upright Motor Control (UMC) extension test provides information on the ability to weight bear and extend the leg in person post-stroke (37–39). In this test the participant bend both knees approximately 30◦ with light upper extremity support for balance and then lifts their leg off the ground. In this single leg support position, the ability of the involved knee to extend is graded strong (scored as 1) if able to fully extend, moderate (scored as 2) if able to support on flexed knee, and weak (scored as 3) if unable to support on flexed knee. The closed chain position is similar to the leg extension position needed for the landing phase of the induced step.

# Perturbation-Induced Stepping

Participants stood with feet shoulder-width apart, each foot on a separate force platform (**Figure 1**). They were placed in a safety harness attached by straps to an overhead rigid beam that minimized potential falls but allowed them to move freely. The placement of the feet was traced to allow the subject to return to the same foot position for each trial. A belt linked to a cable connected to a perturbation device was secured to the subject's waist, and the cable height was adjusted to the level of the umbilicus. Stepping was induced using a mechanical weight drop system hidden behind a screen that delivered an anterior perturbation equal to 10% of the individual's body weight (BW). A monitor provided visual feedback to the subjects on their body weight distribution and the pull was triggered when the subject maintained the required weight distribution (see below) on the force plate for 250–1,000 ms. Participants were instructed to react naturally and not resist the waist pull perturbation.

Three different body weight distributions were used to encourage stepping with both legs (**Figure 2**). In the symmetrical stance (SS) condition, participants placed 50% (±3%) of their weight on each foot. In the preferred asymmetrical stance (PAS) condition, they placed 70% (±3%) of BW on their preferred supporting leg (i.e., the one that controls naturally selected for single limb stance and the less-involved limb for the post-stroke participants). In the non-preferred asymmetrical stance (N-PAS) condition, they were asked to put 70% (±3%) of BW on the non-preferred supporting leg (i.e., the more-involved limb for the post-stroke participants). These values were based on the abilities of post-stroke participants to shift weight onto their involved leg in prior studies and in pilot testing (40, 41).

After three practice trials of increasing weight bearing to the predetermined level, 10 trials at each of the three weight distribution conditions and five catch trials (i.e., perturbation at only 2% BW) were presented in the same predetermined standard randomized order. Participants were given a seated rest after 18 trials or as needed. A research personnel was always standing next to the participants for safety.

# Data Recording

Two force platforms (AMTI, OR6-6, Newton MA, USA) recorded the forces under each foot and allowed the calculation of the total center of pressure (CoP) position. Reflective markers were attached bilaterally over key landmarks of the leg and trunk (on the medial and lateral malleolus, calcaneus, lateral first and fifth metatarsal head, top of second metatarsal head, medial and lateral femoral condyles, and bilateral acromion) to record foot, shank, and trunk movements using an eight-camera motion analysis system (QTM-Qualisys Tracking Manager, Qualisys, Gothenburg, Sweden). Perturbation, kinetic, and kinematic data was collected in Qualisys at 100 Hz for 15 s, beginning at least 1 s prior to perturbation. Kinematic data was filtered using a second order Butterworth filter with 1 bidirectional pass resulting in a fourth order filer. The cutoff frequency was set at 10 Hz (42, 43).

# Dependent Variables

Following the perturbation, pre-step postural activity, referred to as anticipatory postural adjustment (APA), is often characterized TABLE 1 | Participants characteristics.


ABC, Activity-specific Balance Confidence Scale; TUG, Timed Up and Go test; PSL, preferred stance leg = less-involved leg (without paresis) for participants post-stroke; UMC, Upright Motor Control test; \* indicates significant difference between groups (Man-Whitney U-test), #indicates significant difference between more-involved (non-PSL or paresis side) and less-involved legs (PSL) in stroke (p<0.05) (Wilcoxon signed-rank test).

by an initial displacement of the CoP toward the stepping leg that initiates weight transfer toward the upcoming single stance leg (12). Its characteristics were derived from the net mediolateral CoP displacement. The APA onset was defined as the beginning of the CoP displacement (i.e., when the first derivative becomes continuously >0). The APA amplitude was the maximum displacement of the net CoP toward the stepping side, and the APA duration was from the onset of the CoP displacement to the APA Peak.

For each trial, the execution of the step was evaluated by determining the number of steps, first step onset timing, first step length, width, height, and duration. These characteristics were identified from the ankle markers of the stepping side. The beginning and end of the step were identified from the vertical velocity of the marker in order to determine the step duration and the mediolateral and anteroposterior displacement of the foot (i.e., step width and length). The step clearance was defined as the maximal vertical excursion of the step ankle marker above the ground. The onset time of stepping was calculated relative to the onset of the perturbation.

To quantify the termination of the whole-body movement which influences the maintenance of balance and the necessity of a second step, the velocity of the body at the end of the first step (first heel strike) was calculated using the midpoint of the bilateral acromion markers. The velocity was multiplied by the subjects' mass to calculate the body momentum as the difference in momentum between first and second heel strike.

# Statistics

Descriptive statistics were used to describe the data for both groups and are reported as median (Mdn) and interquartile range (25 and 75th percentile). The Shapiro-Wilk and Levene tests indicated that the reactive step measures were not normally distributed, and that the variance was different between groups. In addition, an unequal number of stepping responses between groups and legs were observed (**Figure 2**), thus requiring nonparametric analyses. Non-parametric tests were also used for the clinical measures.

Based on the number of participants who took steps with each leg (**Figure 2**), the difference between groups was tested separately for each stepping leg using the Mann-Whitney U-test. We compared the groups when stepping with the preferred/lessinvolved leg for the N-PAS and SS conditions, and with the nonpreferred/more-involved leg for the PAS condition. Differences between legs were assessed using a Wilcoxon signed-rank test in SS conditions in controls, whereas for post-stroke participants, we compared stepping performance between the legs for the same body weight distribution condition. Thus, steps with the lessinvolved leg in the N-PAS condition (i.e., less weight on the stepping leg) were compared to steps with the more-involved leg in the PAS condition. The IBM SPSS 23 was used for all statistical analyses. An alpha level of 0.05 was used to test statistical significance.

# RESULTS

# Behavioral Responses to Perturbations

Across all trials, one control participant resisted the pull and did not step on the first trial, otherwise all participants stepped in all trials. Two participants (one individual post-stroke and one control) needed external assistance to regain their balance during one trial and those two trials were dropped from the analyses.

The control group stepped more often with their nonpreferred stance leg (68.5%) in the SS condition (5 persons used that leg only); whereas in the asymmetrical conditions, they stepped more often with the leg initially supporting less weight, which was the preferred leg in the N-PAS condition (79.1% - 6 persons used that leg only) and the non-preferred leg in the PAS condition (92.4% - 12 participants used that leg only) (**Figure 3**). The post-stroke participants stepped with their moreinvolved/non-preferred leg only 11.5% of the trials in the SS

condition (13 participants never used that leg) and only 6.1% in the N-PAS condition (16 participants never used that leg). However, in the PAS condition (i.e., when standing with more weight on the less-involved / preferred stance leg) they increased the use of the opposite more-involved leg to step (46.2%) and only 5 patients never stepped with that leg (**Figure 3**).

Z = −4.568, p < 0.001; PAS: U = 46.5, Z = 3.613, p < 0.001).

A significant difference was observed between groups for the number of steps taken when stepping with the preferred leg in both the SS (U = 48, Z = −2.560, p = 0.009) and N-PAS condition (U = 60, Z = −3.0, p = 0.002). Post-stroke participants took more steps (SS: Mdn = 2.40 [2.11 3]; N-PAS: Mdn = 2.53 [2 2.9]) than controls (SS: Mdn = 2 [2 2.33]; N-PAS: Mdn = 2 [1.89 2]) to regain balance. They also took significantly (U = 31, Z = −3.518, p < 0.001) more steps (Mdn = 2.78 [2 3]) than controls (Mdn = 2.00 [1.27 2]) when stepping with the non-preferred/more-involved leg in the PAS conditions.

# Difference Between Groups APA Characteristics

Postural activity before stepping (i.e., APA) was observed in most trials for both groups of participants (96% for controls and 94% for post-stroke participants). APA was absent in 54 trials: 17 trials (3.5%) involving 7 controls and 37 trials (6.2%) involving 15 post-stroke participants.

**Figure 4** shows the characteristics of the APA for both groups of participants in the analyzed conditions. When stepping with

anticipatory postural adjustment's characteristics is presented for each group and each condition in which they were compared. Boxes represent 25 and 75th percentile. Bars represent min and max values. \*Indicates a difference between groups at p < 0.05.

the preferred stance leg, there was no significant difference between groups for APA duration (U = 86, Z = 0.616, p = 0.538) or amplitude (U = 98, Z = −0.088, p = 0.930) in SS condition or in N-PAS condition (duration: U = 112, Z = 1.267, p = 0.205; amplitude: U = 121, Z = 0.967, p = 0.334). This indicates that, when post-stroke participants stepped with their less-involved leg, the step preparation had the same characteristics as controls. However, APA onset occurred significantly earlier in post-stroke participants compared to controls in both the SS conditions (U = 26, Z = 3.256, p = 0.001) and N-PAS (U = 76, Z = 2.467, p = 0.014).

When stepping with the non-preferred leg in the PAS condition, there was also no significant difference in APA duration (U = 97, Z = −0.909, p = 0.363) or amplitude (U = 116, Z = −0.158, p = 0.874). However, APA onset occurred earlier post-stroke compared to controls (U = 63, Z = 2.253, p = 0.024).

#### First Step Characteristics

**Figure 5** shows the main characteristics of the first step for both groups of participants in the analyzed conditions. When stepping with the preferred stance leg, step characteristics were not significantly different between groups in the SS condition for most of the step variables (onset: U = 73, Z = 1.528, p = 0.127; duration: U = 61, Z = 1.716, p = 0.086; height: U = 87, Z = 0.950, p = 0.342). However, the step length (U = 59, Z = −2.241, p = 0.025) and step width (U = 59, Z = 2.11, p = 0.035) were different between groups showing that post-stroke participants took shorter and slightly more lateral (Mn = 1.52 cm [−0.05 2.96]) steps than controls (Mn = −0.05 cm [−1.51 1.25]). In the N-PAS condition, step onset was earlier (U = 77, Z = 2.433, p = 0.015), step length (U = 52, Z = −3.267, p = 0.001) and step duration (U = 79, Z = 2.367, p = 0.018) were shorter for the post-stroke participants compared to controls. No significant difference was found for step height (U = 124, Z = 0.950, p = 0.342) or width (U = 124, Z = 0.867, p = 0.386).

When stepping with the non-preferred leg in the PAS condition, step duration was significantly longer (U = 60, Z = −2.372, p = 0.018) and step length was significantly shorter (U = 59, Z = −2.411, p = 0.016) for post-stroke participants compared to controls. No differences were found for other step characteristics in this condition (onset: U = 71, Z = 1.937, p = 0.053; height: U = 78, Z = 1.660, p = 0.097; or width: U = 91, Z = 1.146, p = 0.252).

#### Landing Characteristics

**Figure 6** shows the main characteristics of the landing for both groups of participants in the analyzed conditions. When stepping with the preferred stance leg in the SS condition, there was no difference between groups either in the body's velocity at first heel strike (U = 68, Z = 1.57, p = 0.116) or in the change in momentum between the first and the second step (U = 73, Z = −1.356, p = 0.175).

Asymmetrical conditions uncovered differences between groups. When stepping with the preferred leg in the N-PAS condition, the velocity at the end of the step was smaller for poststroke participants compared to controls (U = 66, Z = 2.65, p = 0.008) and the change of momentum between first and second steps (U = 92, Z = −1.752, p = 0.08) tended to be smaller for post-stroke participants.

When stepping with the non-preferred leg in PAS condition, the velocity at the end of the step was smaller (U = 45, Z = 2.122, p = 0.034) and there was a smaller change of momentum (U = 34, Z = −2.665, p = 0.008) between first and second steps for post-stroke participants compared to controls indicating that post-stroke participants were less efficient arresting the body momentum.

### Difference Between Legs

No significant difference between legs was found for controls in postural activity preceding the step (p > 0.38), for stepping measures (p > 0.53) or the landing characteristics (p > 0.15). The only almost significant difference concerned the step width (Z = 1.87, p = 0.062). Controls tended to place their foot slightly more medially when stepping with the preferred leg (Mdn = 1.51 cm

[−0.6 2.96]) than when stepping with the non-preferred leg (Mdn = −0.20 cm [−1.29 2.03]).

and max values. \*Indicates a difference between groups at p < 0.05.

For post-stroke participants, the APA onset (Z = 0.45; p = 0.65) and amplitude (Z = 1.19; p = 0.23) before the step were the same between the legs but APA duration tended to be longer (Z = 1.76, p = 0.08) when stepping with the more-involved leg compared with the less-involved leg (**Figure 7**). No significant difference was found between legs for step onset (Z = 0.91, p = 0.363), step width (Z = 1.59, p = 0.112), step height (Z = 1.13, p = 0.256) or step length (Z = 0.51, p = 0.609). However,

a significant difference was found in number of steps (Z = 2.45, p = 0.014) and for step duration (Z = 3.07, p = 0.002). Poststroke participants took more steps (Mdn = 2.78 [2 3]) when stepping with the more-involved leg than when stepping with the less-involved leg (Mdn = 2.3 [2 2.8]) and step duration was longer with the more-involved leg compared to the less-involved one (**Figure 8**). Finally, although the whole-body velocity at first heel strike was the same (Z = 0.863, p = 0.388), the change in momentum was significantly smaller (Z = 2.589, p = 0.010) when stepping with the more-involved leg compared to the lessinvolved side in participants post-stroke. Thus, although the body was moving at the same speed at the end of the first step, post-stroke participants had difficulties in reducing the body momentum (**Figure 9**).

# DISCUSSION

This pilot study was possibly the first to examine the potential differences during induced reactive stepping in the anterior direction between post-stroke and control participants, and between legs in the post-stroke participants. As hypothesized, differences were found between post-stroke participants and controls for the non-preferred leg: steps were longer in duration and the change in body momentum following the first step was smaller despite a smaller velocity of the body at heel strike

thus requiring additional steps. However, differences were also found between groups for the preferred leg: APA onset and step onset were earlier and step length was shorter. Difference in selection of the stepping leg was also seen in asymmetrical conditions allowing for a better comparison between legs in post-stroke participants. Steps with the more-involved leg were longer in duration compared to the less-involved leg and also displayed a smaller change in body momentum, thus requiring more additional steps. The findings provide several new insights regarding the stepping strategies of persons post-stroke in response to a balance perturbation and illustrate the complexity of perturbation-induced stepping performance with either leg after a stroke.

# Stepping With the Less-Involved Leg: A Preset Strategy

Stepping with the less-involved (preferred) leg was the strategy observed most often among the post-stroke group as previously reported (18, 19), particularly during symmetrical stance or when more weight was initially placed on the more-involved leg. In these cases, post-stroke participants reacted faster to the postural perturbation as illustrated by an earlier APA and step onset. This is consistent with previous studies that also showed faster initiation timing for reflex-like induced stepping in healthy elderly individuals with past falls (12, 44). It may reflect instability to the perturbation and/or anxiety about falling that possibly heightens a fear potentiated triggering of an earlier stepping response. This would thus correspond to a predetermined strategy to step earlier at the time of the perturbation, rather than waiting for a specific evaluation of the evolving conditions of instability based on sensory information. The lower ABC score illustrating lower balance confidence found in the post-stroke group compared to the controls is in line with this hypothesis. In this group of chronic stroke survivors, the decision to step could also be indicative of a learned behavior that stepping in response to the perturbation is functionally more effective than a feet-in-place response, similar to that found in healthy older adults (45). Other studies in older adults have also shown that reactive induced steps are often taken well before the limits of stability exceed the BOS (43, 45–48).

In the asymmetrical stance condition, a shorter step was observed in the post-stroke group compared to controls when they stepped with the less-involved leg and more weight was on the more-involved leg. Standing with greater weight on the moreinvolved leg is often challenging for people post-stroke and can be observed clinically. The shorter step length in this condition may reflect not only a desire to return to bipedal stance as soon as possible, but possibly lateral instability on the more-involved stance side during the step. In this case, the generation of hip abduction torque is an important contributor to the regulation of mediolateral stability during the single limb stance phase of stepping as well as for ongoing gait (49). Deficits in hip abduction torque associated with hip extension after stroke would possibly contribute to the reduction of step length (50).

Although we anticipated that stepping with the less-involved leg in post-stroke participants would be similar to that of controls, it was not the case. Despite a smaller velocity of the body at the end of the first step and similar change in momentum, post-stroke participants took more than two steps to stop their forward progression when using the lessinvolved leg whereas the controls stopped generally with <2 steps. Thus, the need for additional steps might be related to instability arising at the end of the second step when the moreinvolved leg contacts the ground, creating the necessity for additional steps.

# Stepping With the More Involved Leg: Issues at All Phases

Stepping with the more-involved leg was achieved by changing the initial weight distribution and initially reducing the

represent min and max values.

amount of weight on that leg. The longer step duration and shorter step length of the more-involved leg in post-stroke participants compared to controls is not surprising given the greater sensorimotor impairments of hemiparetic side. The lack of differences between groups and legs for the other step characteristics when stepping with the more involved leg was unexpected. Given that the perturbation was standardized and the velocity at landing was not significantly different between legs, the problem for stepping with the more-involved leg was likely not directly related to the execution phase of the reactive step. Instead, the smaller change in momentum between heel strikes for the post-stroke participants suggested that they were less able to slow the momentum of the body between the first step with the more-involved leg and the second step, and thus that the control of the landing phase of the first step was problematic. This could be influenced by decreased extension strength or control in the more involved leg (51, 52) that is needed at step landing. This diminished extension control is noted in the differences between groups and between legs in the Upright Motor Control test which assessed the ability to perform leg extension from a flexed standing position.

When using the more-involved leg, no group differences in the APA duration and amplitude were observed. However, the comparison was made in the asymmetrical condition when the stepping leg supported only 30% of the body weight and thus when the need for weight transfer postural requirements were minimized. Despite the lack of differences between the groups, a deficit in pre-step mediolateral postural control when stepping with the more-involved leg cannot be ruled out. Specifically, when comparing the legs in post-stroke participants for the same weight distribution condition, the pre-step postural adjustment tended to be longer for steps with the more-involved leg compared to steps with the less-involved leg. Even though the legs were partially unweighted, additional time might be needed to complete the postural adjustment before a step with the more-involved leg thus illustrating deficits in postural control. This may result from the larger variability in the amplitude of the APA. In contrast with controls or less-involved leg steps, several of the post-stroke participants displayed multiple peaks in their APA performance before reaching their maximum amplitude prior to releasing a step with the more-involved leg. This observation resembled that of previous studies showing differences in postural adjustments during gait initiation between post-stroke and control participants (53) and between the more and less involved legs (2). It is quite possible that the difference in APA duration is related to challenge of the mediolateral postural adjustment prior to the release of the step.

Post-stroke participants took more steps when stepping with the more-involved leg than with the less-involved one. The need for more than two reactive steps when stepping with the more-involved leg may be less about the step execution and more about the postural challenge before the step and/or at landing. The smaller change in momentum (see **Figure 9**) observed between the first step with the more-involved leg and second step tend to support this hypothesis. Thus, the use of additional steps appeared to be related to difficulties in

slowing the body when the more-involved leg lands on the ground. This underscores the need to consider all phases of perturbation induced stepping in the design of intervention

programs targeting the specific deficits in lower limb motor control problems that occur following stroke and the need to facilitate the interactive use of both legs for balance recovery through stepping.

# Using Asymmetrical Stance for Enhancing Performance of the More-Involved Leg

The increased use of the more-involved leg for stepping when initially bearing 30% of body weight support illustrates the difficulty of releasing a step when more weight is on the leg and may reflect an underlying issue with the postural demands prior to stepping. It is thus important to implement situations facilitating the use of the more-involved leg by varying (reducing or increasing) postural demands prior to the step. In a prior study using diagonal anterior waist-pull perturbations, a pull toward the less-involved leg appeared to force increased stepping with the more-involved leg (20). This diagonal pull assisted with the weight shift off of the moreinvolved leg prior to stepping and facilitated increased use of the limb.

As a therapeutic approach, induced step training has been used in other populations and has shown improvement in step initiation timing in healthy older and younger adults for both reactive and voluntary stepping (54), and increased in step length in Parkinson's (55). After stroke, induced step training has been found to decrease the step completion time in one acute stroke survivor (56), and increase use of the more-involved in chronic post-stroke participants (57). Based on these results, it is conceivable that perturbation-induced step training impacts the simultaneous roles (support and stepping) of both legs in the ambulatory post-stroke individuals. In this regard, one intervention strategy should focus on increasing initiation of steps with the more-involved leg which will not only challenges the postural adjustment prior to stepping but also encourage development of strategies to control the momentum of the body at landing and subsequent steps. This could be accomplished by systematically modifying the initial stance symmetry as seen in this study or through use of cueing to step with the moreinvolved leg as done in other studies (23, 24, 57). Another intervention strategy should focus on training the less-involved leg to take a larger step which would facilitate weight shift and a longer single limb stance phase on the more-involved leg to enhance mediolateral stability. This could be done with visual targets or cues for foot placement. These approaches would address the bipedal nature of perturbation induced stepping and could potentially enhance the effectiveness of stepping with either leg.

# Limitations of the Study

This pilot study results are limited to community dwelling ambulatory chronic post-stroke survivor's reaction to sudden anterior perturbations of balance in controlled conditions. Due to the unrestrained responses of the participants, differences between legs for all stance symmetry posture could not be analyzed. Although the laboratory setting tries to mimic real life perturbation situations, only the onset of the pull was unknown whereas the perturbation direction was predictable. Lastly, specific neuromuscular measures were not taken so we can only speculate on the precise underlying neuromotor factors that may contribute to these findings. The findings however illustrate several issues with reactive stepping that involve both legs post-stroke.

# CONCLUSIONS

These results highlight an altered stepping strategy involving both legs in persons post-stroke and illustrate the complexity of perturbation-induced stepping with either leg. Stepping with the more-involved leg can be facilitated by unweighting the leg which constitutes a simple intervention to encourage stroke survivors to use their weaker limb. The predilection for initiating a step with the less-involved leg may be, at least in part, a learned behavior as stepping with either leg after a perturbation appears to be challenging for ambulatory stroke survivors. Therefore, consideration of the simultaneous roles (support and stepping) of both legs during reactive stepping is important for reactive balance training and should be included when designing rehabilitation approaches to enhance balance function.

# AUTHOR CONTRIBUTIONS

KM designed and collected the data, conducted the data analysis, and drafted the initial manuscript. M-LM assisted with design, data analysis, and manuscript revisions. MR, MB, and MC assisted with results interpretation and gave critical comments for the manuscript. All authors read and approved the final manuscript.

# REFERENCES


# FUNDING

Financial support for the subjects was provided by Northwestern University Department of Physical Therapy & Human Movement Sciences.

# ACKNOWLEDGMENTS

The authors acknowledge and thank the participants, Doctoral Physical Therapy students, Nicholas Reimold, M Wu, and Keith Gordon for their assistance with this project. This study was part of a Ph.D. dissertation requirement for Nova Southeastern University.


function and perceived health status after stroke. Arch Phys Med Rehabil. (2006) 87:364–70. doi: 10.1016/j.apmr.2005.11.017


**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 Martinez, Rogers, Blackinton, Cheng and Mille. 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.

# Effects of Virtual Reality-Based Physical and Cognitive Training on Executive Function and Dual-Task Gait Performance in Older Adults With Mild Cognitive Impairment: A Randomized Control Trial

Ying-Yi Liao<sup>1</sup> , I-Hsuan Chen<sup>2</sup> , Yi-Jia Lin<sup>3</sup> , Yue Chen<sup>3</sup> and Wei-Chun Hsu<sup>3</sup> \*

<sup>1</sup>Department of Gerontological Health Care, National Taipei University of Nursing and Health Sciences, Taipei, Taiwan, <sup>2</sup>Department of Physical Therapy, Fooyin University, Kaohsiung, Taiwan, <sup>3</sup>Graduate Institute of Biomedical Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan

#### Edited by:

Helena Blumen, Albert Einstein College of Medicine, United States

#### Reviewed by:

Markus A. Hobert, University of Tübingen, Germany Mark Speechley, University of Western Ontario, Canada

> \*Correspondence: Wei-Chun Hsu wchsu@mail.ntust.edu.tw

Received: 01 April 2019 Accepted: 12 June 2019 Published: 16 July 2019

#### Citation:

Liao Y-Y, Chen I-H, Lin Y-J, Chen Y and Hsu W-C (2019) Effects of Virtual Reality-Based Physical and Cognitive Training on Executive Function and Dual-Task Gait Performance in Older Adults With Mild Cognitive Impairment: A Randomized Control Trial. Front. Aging Neurosci. 11:162. doi: 10.3389/fnagi.2019.00162 Background: Walking while performing cognitive and motor tasks simultaneously interferes with gait performance and may lead to falls in older adults with mild cognitive impairment (MCI). Executive function, which seems to play a key role in dual-task gait performance, can be improved by combined physical and cognitive training. Virtual reality (VR) has the potential to assist rehabilitation, and its effect on physical and cognitive function requires further investigation. The purpose of this study was to assess the effects of VR-based physical and cognitive training on executive function and dual-task gait performance in older adults with MCI, as well as to compare VR-based physical and cognitive training with traditional combined physical and cognitive training.

Method: Thirty-four community-dwelling older adults with MCI were randomly assigned into either a VR-based physical and cognitive training (VR) group or a combined traditional physical and cognitive training (CPC) group for 36 sessions over 12 weeks. Outcome measures included executive function [Stroop Color and Word Test (SCWT) and trail making test (TMT) A and B], gait performance (gait speed, stride length, and cadence) and dual-task costs (DTCs). Walking tasks were performed during single-task walking, walking while performing serial subtraction (cognitive dual task), and walking while carrying a tray (motor dual task). The GAIT Up system was used to evaluate gait parameters including speed, stride length, cadence and DTCs. DTC were defined as 100

∗ (single-task gait parameters − dual-task gait parameters)/single-task gait parameters.

Results: Both groups showed significant improvements in the SCWT and single-task and motor dual-task gait performance measures. However, only the VR group showed improvements in cognitive dual-task gait performance and the DTC of cadence. Moreover, the VR group showed more improvements than the CPC group in the TMT-B and DTC of cadence with borderline significances.

Conclusion: A 12-week VR-based physical and cognitive training program led to significant improvements in dual-task gait performance in older adults with MCI, which may be attributed to improvements in executive function.

Keywords: dual-task gait, MCI, virtual reality, executive function, combined physical and cognitive training

# INTRODUCTION

Performing activities of daily living (ADLs) requires the ability to perform multiple cognitive and physical tasks simultaneously. Dual tasking destabilizes gait performance, especially for those with impaired cognitive function, which may lead to falls (Springer et al., 2006). A previous study reported that older adults with mild cognitive impairment (MCI) showed significant decreases in gait velocity, increases in stride time and increases in stride time variability when changing from a single to a dual task (Muir et al., 2012). Meta-analysis has also revealed some differences between MCI patients and cognitively normal controls in several gait parameters, including velocity, stride length and stride time, during either a single or dual task (Bahureksa et al., 2017). In addition, dual-task gait performance has been associated with progression to dementia in patients with MCI. Dual-task gait testing may be used by clinicians to assess the risk of cognitive decline (Montero-Odasso et al., 2017; Rosso et al., 2019).

Executive function is defined as a set of cognitive skills necessary for planning, monitoring and executing a sequence of goal-directed complex actions (Diamond, 2013). Executive dysfunction, such as inadequate divided attention and selective attention, are more pronounced in older adults with MCI than in controls (Johns et al., 2012; Kirova et al., 2015). Executive function can modulate competition interference between two attention-demanding tasks and has been suggested to be associated with spatial and temporal characteristics of dual-task gait performance (Woollacott and Shumway-Cook, 2002; de Bruin and Schmidt, 2010). For example, Doi et al.'s (2014) study showed significant correlation between executive function and dual task gait speed in 389 older adults with MCI. Consequently, individuals with poor executive function have reduced gait speed and higher levels of fall risk and functional disability (Johnson et al., 2007; Herman et al., 2010). Improving executive function and minimizing dual-task interference may, therefore, have clinical utility in avoiding falls occurring in older adults with MCI (Bahureksa et al., 2017).

Either cognitive or physical training has proven to be an effective intervention in enhancing cognitive functions in older adults with MCI (Simon et al., 2012; Suzuki et al., 2012, 2013). Therefore, some studies have investigated the combined effects of physical and cognitive training (Barnes et al., 2013; Anderson-Hanley et al., 2018; Damirchi et al., 2018). Studies have found that older adults with MCI showed greater improvements in various cognitive functions after receiving combined therapy than after receiving cognitive or physical therapy alone (Barnes et al., 2013). In addition to improved cognitive function, dual-task gait interference may be decreased through the repeated practice of dual tasking (i.e., combined physical and cognitive interventions) in accordance with the principles of task-specific training (Plummer et al., 2015). However, the amount of evidence on the effects of combined physical and cognitive exercises on dual-task gait performance in older adults with MCI is limited. Tay et al.'s (2016) finding showed that combined physical and cognitive training improved dual-task walking performance. More evidence is required to investigate the training effects in older adults with MCI.

Virtual reality (VR) is a computer-generated technology that enables interactions between the user and virtual environments. The advantages of using VR interventions include enhancing accessibility and cost-effectiveness, creating an immersive experience, and providing immediate feedback based on an individual's performance. Previous articles have shown positive effects of these VR interventions on attention and visual and verbal memory, as well as executive functioning, in older adults with MCI (Coyle et al., 2015; Mrakic-Sposta et al., 2018). Due to the advantages of VR, integrating physical and cognitive training into VR appears to be a good intervention approach. However, most articles using VR training include either physical or cognitive training (García-Betances et al., 2015). Studies on the effects of combining both physical and cognitive training in the VR context are lacking. In addition, instrumental ADLs (IADLs) comprise many dual-task activities and require high demands on executive functioning. Integrating IADL tasks into the VR context represents an innovative approach to improve executive function in older adults with MCI. Whether these innovative interventions can reduce dual-task interference and transfer to improvements in dual-task gait performance requires investigation. Therefore, the purpose of the current study was to assess the effects of VR-based physical and cognitive training on executive function and dual-task gait performance in older adults with MCI, as well as compare the VR-based physical and cognitive training with traditional combined physical and cognitive training.

# METHODS

# Study Design and Protocol

This study was a single-blinded (assessor) randomized controlled trial. Participants were randomly assigned to either the VR training group or the combined physical and cognitive training (CPC) group via a sealed envelope. Subjects in the VR group participated in a 60-min, VR-based physical and cognitive training each visit, three times a week for 12 weeks. Those in the CPC group participated in combined physical and cognitive training for 60 min each visit, three times a week for 12 weeks. An experienced physical therapist supervised exercise training in a small group for both the VR and CPC groups. The assessor, who was always blinded to the group assignments, measured the outcomes at baseline and after completing the 36 sessions. This trial was approved by the Institutional Review Board of National Yang-Ming University, and consent forms were obtained from all participants at the beginning of the experimental procedures. The trial was registered in the Thai Clinical Trials Registry<sup>1</sup> , and the approval number is ''TCTR20180531001.''

# Participants

All participants were recruited from communities and day care centers of Taipei, Taiwan. The inclusion criteria were: (1) aged 65 years and over; (2) able to walk more than 10 m without walking aids; (3) had a Montreal Cognitive Assessment (MoCA) score lower than 26 (Tsai et al., 2012); (4) had self-reported memory complaints; and (5) had the ability to perform ADLs. The exclusion criteria included: (1) dementia; (2) a history of malignant tumors with life expectancy less than 3 months; (3) the presence of an unstable neurological or orthopedic disease interfering with participation in the study; and (4) an education level less than 6 years (elementary school).

# Intervention

# Combined Physical and Cognitive Training (CPC) Group

Our CPC program contains both physical and cognitive elements of training. The physical training regimen comprised resistance, aerobic and balance exercises that meet the standards of the American College of Sports Medicine for seniors (Chodzko-Zajko et al., 2009; Garber et al., 2011). Our physical training was set to reach 50%–75% of the maximal heart rate (calculated as 220- age) with the exertion perceived by the participants as ''somewhat hard'' (scored 13–14). Specifically, Therabands were applied to assist the training of both the upper and lower extremities during the resistance exercise. A series of whole-body aerobic exercises, for example, stepping while in the seated and standing positions, as well as on and off of a stool, was performed. Balance exercises included standing on a steady foam mat in various postures and walking forward and backward with eyes open and closed. Other functional tasks simulated ADLs and were designed to enhance motor performance and were integrated into the CPC program. Samples of functional tasks included asking participants to climb stairs, cross obstacles while reaching for objects, and turning and rising from a chair. In addition to the functional tasks, training that targeted cognitive abilities was also integrated into the physical training program. Training scenarios included walking while reciting poems, naming flowers and animals while crossing obstacles, solving math questions during the resistance training, drawing a circle in the air in the clockwise or counterclockwise direction with the right or left hand, respectively, and searching for the prefix and roots of a Chinese character at moments when they repetitively stand up from a chair.

## Virtual Reality-Based Physical and Cognitive Training (VR) Group

Scenes from the VR-based physical and cognitive training program are shown in **Figure 1**.

FIGURE 1 | Scenes from the VR-based training program. (A) Take the MRT. (B) Kitchen chef. (C) Convenience store clerk. (D) Tai Chi. (E) Football (running and stepping). (F) Subject wearing VR glasses and performing VR tasks. (B) and (C) are derived from "job simulator" created by Owlchemy Labs. Both the subject and trainer have provided written informed consent allowing the publication of this image.

# **VR-Based Physical Exercise Program**

We used the Kinect system (Microsoft Corporation, Redmond, WA, USA) to capture the limb motions and create a full-body 3D virtual map. The physical elements of the VR training were developed by the well-established and widely used Tano and LongGood programs. We adopted programs, including a simplified 24-form Yang-style Tai Chi, resistance exercise, aerobic exercise, and functional tasks in the forms of window cleaning, goldfish scooping and other tasks relevant to daily activities, to improve upper and lower extremity balance, stability, strength and endurance (**Figures 1D,E**). In the VR context, participants would imitate the virtual character and adjust their movements based on the simultaneous visual and auditory feedback.

# **VR-Based Cognitive Training Program**

The cognitive training required wearing the VR glasses on their heads with a motor controller in both hands to execute the training tasks (**Figure 1F**). Our laboratory invented most of the cognitive training VR games, while others were derived from the ''Job Simulator'' software developed by Owlchemy Labs. The concept of the cognitive programs was inspired by simulated IADL tasks. For example, in the taking mass rapid transit (MRT) game, participants took the MRT in a familiar VR context where station gates, ticket vending machines, and ATMs were located in the usual places. To complete the task, a participant needed to be aware of their present location and the designated stations. They also needed to gather enough coins based on the fare chart to obtain a ticket. In the store finder game, a big red cross sign

<sup>1</sup>http://www.clinicaltrials.in.th/

appeared as an indicator when something was going wrong. A participant needed to virtually walk to the store noted on a map in less than 3 min. If the participants failed to get closer to the targeted store in 2 min, directional marks in red popped up to guide their way. In the kitchen chef game, a participant found herself/himself in a well-equipped kitchen surrounded by numerous utensils available for use to prepare an ordered dish. Once he or she was able to complete a simple meal, a more complicated dish requiring more ingredients and utensils to complete followed. The last game was convenience store clerk; participants were responsible for gathering items from the to-do list and checking them out. Some of the listed items were easy to find, while others could not be located as easily (**Figures 1A–C**).

# Outcome Measures

These outcomes were all secondary outcomes of our project. We wanted to explore how our secondary outcomes are responded to the intervention. Therefore, this is an exploratory study of mechanisms.

# Executive Function

### Trail Making Test (TMT)

The TMT has been hypothesized to reflect components of executive function, such as visual attention and task switching (Arnett and Labovitz, 1995). The current study used TMT-A and the Chinese version of TMT-B (Wang et al., 2018b). The TMT-A was composed of 25 consecutive Arabic numbers, and participants connected the numbers while following the numerical sequence. The TMT-B was composed of 12 consecutive Arabic numbers and 12 Chinese characters that represented a Chinese zodiac sign. Participants drew lines to connect the circles in ascending order while following an additional rule of alternating between numbers and animal signs (i.e., 1 - rat - 2 - ox - 3 - tiger, etc.). The time to complete each test was recorded. Delta TMT (TMT B subtract TMT A) was also recorded as our TMT outcome.

### Stroop Color and Word Test (SCWT)

The Stroop Color and Word Test (SCWT) has been used to assess the ability of inhibition in executive function (Scarpina and Tagini, 2017). The current study used the Chinese version of the Stroop (Wang et al., 2018b), which was composed of four characters and four colors. The incongruous conditions were used in the present study. In this condition, the colors of the characters were printed in an inconsistent color ink (e.g., the character ''blue'' in red ink). Participants had to name the color of the ink rather than state the word/character as quickly as they could in a limited time. The number of correct answers in 45 s (SCWT number) and time to name 45 characters (SCWT time) were our outcomes.

# Gait Performance

Gait performance was measured in three conditions: (1) walking at their preferred walking speed (single task); (2) walking while executing a serial subtraction by three task, starting from a randomized 3-digit number (e.g., 100, 97, 94. . .; cognitive dual task); and (3) walking while carrying a tray with glasses of water (motor dual task). Our primary task is waking, and the secondary tasks are executing a serial subtraction or carrying a tray. Participants were asked to focus on their walking task and walk three trials under each condition. The trial intervals were 1 min. Data were averaged from the three trials.

The GAIT Up system (Gait Up, Lausanne, Switzerland), a wearable device with good validity and reliability, was used to evaluate gait parameters for the above testing conditions (Mariani et al., 2010; Dadashi et al., 2014). The dimensions of the GAIT UP sensors were 50 mm × 40 mm × 16 mm, and the weight was 36 g. Two wireless inertial sensors with tri-axial accelerometers were fixed on the upper part of the shoe with an elastic strap. In the present study, spatiotemporal parameters recorded during each trial included speed (m/s), stride length (cm) and cadence (step/min). Dual-task interference was quantified by calculating the dual-task costs (DTCs) according to the customary formula (McDowd, 1986). For example, DTC-speed [%] = 100 <sup>∗</sup> (single-task walking speed − dual-task walking speed)/single-task walking speed.

# Data Analysis

Demographic and behavioral data analyses were performed using SPSS 20.0 software (SPSS Inc., Chicago, IL, USA). Descriptive statistics were generated for all variables, and the distributions of the variables were expressed as the means ± standard deviations or as the numbers (%). Intergroup differences in the baseline characteristics were analyzed using independent t-tests or chi-square tests. Two-way analysis of variance (ANOVA) with repeated measures was used to determine the effects of the intervention on executive function and gait performance. The model effects were the groups (VR and CPC), the times (pre and post), and their interaction. The intergroup comparison was across groups, and the intragroup comparison was the change over time. A post hoc Tukey's test was used for variables with group × time interaction effects.

# RESULTS

As shown in the flowchart, 42 participants were recruited and randomly assigned to either the VR group (n = 21) or the CPC group (n = 21). Three participants in the VR group and five participants in the CPC group dropped out due to low motivation. A total of 34 participants (18 in the VR group and 16 in the CPC group) completed all the assessments (**Figure 2**). No adverse events were reported throughout the study period. Participant characteristics are shown in **Tables 1**–**3**. Baseline demographic characteristics and outcome measures at the pre-intervention are similar between two groups.

The results of the executive function assessments are shown in **Table 2**. Of the six within group p-values for the TMT, VR group shows two significant values (TMT-B. delta TMT). None of the outcomes of TMT were found to have group × time interactions except for the TMT-B (a borderline significant p = 0.032). Of the four within group p-values for SCWT, both VR and CPC are significant (SWCT-numbers, SWCTtime); neither interaction is significant. The results of the single and dual task gait performance are shown in **Table 3**. For



MMSE, Mini-Mental State Examination; MOCA, Montreal Cognitive Assessment. The data are presented as the means ± SDs or numbers.

single-task gait, both groups have two significant p-values of 12 within group p-values (VR group: gait speed, stride length; CPC group: gait speed, cadence) and no interaction for between groups. For motor dual task gait, VR group has two significant within group p-values (gait speed, stride length), CPC group has three significant within group p-values (gait speed, stride length, cadence) of the 12 within group p-values, and none of the six interactions are significant. For cognitive dual task, VR group has three significant p-values of the 12 within group p-values (gait speed, stride length, DTCs of cadence), CPC group has no significant within group p-value. None of the single and dual task gait outcomes were found to have group × time interactions except for the cognitive DTCs of cadence (a borderline significant p = 0.018).

# DISCUSSION

The goal of the current study was to assess the effects of VR-based cognitive and physical training on executive function and dual-task gait performance in older adults with MCI, as well as compare VR training with CPC training. In this study, we found significant improvements in executive function (SCWT), single-task gait performance and motor dual-task gait performance in both groups. However, only the VR group showed improvements in cognitive dual-task gait performance

VR on Executive-Function and Gait

TABLE 2 | Comparisonof executive function in the virtual reality (VR) training group and the combined physical and cognitive training group.


TMT-A, Trail Making Test, part A; TMT-B, Trail Making Test, part B; Delta TMT, TMT-B minus TMT-A; SCWT, Stroop Color and Word Test; SCWT-number, The number of correct answers in 45 s; SCWT-time, time to name 45 characters. The data are presented as the means±SDs or numbers.∗Significance level<0.05.

TABLE 3 | Comparisonsof single-task and dual-task gait performance in the VR training group and the combined physical and cognitive training group.


Dual task costs [%]= 100 \* (single-task gait parameters−dual-task gait parameters)/single-task gait parameters. The data are presented as the means±SDs.∗Significance level <0.05. and the DTC of cadence after training. Moreover, the VR group showed more improvements than the CPC group in the TMT-B and cognitive DTC of cadence.

In our present study, the SCWT was used to assess selective attention or inhibition. Selective attention (inhibition) is the ability to control/inhibit impulsive responses and create responses by using attention and reasoning. This cognitive ability is one of our executive functions and contributes to anticipation, planning, and goal setting. Therefore, the stronger the inhibition ability is, the less interference will be produced by a new stimulus (Stroop, 1935). Falbo et al. (2016) state that physical and cognitive dual-task training effectively increased inhibitory performance in older adults. In the present dual-task intervention, subjects in both groups learned how to complete the task with multiple stimuli under the mutual interference of the cognitive and physical training. After repetitive practice, interference from new stimuli can be reduced, as reflected by improvements in inhibition shown by the SCWT in both groups.

The TMT-B assesses divided attention, which is the ability to attend to two different stimuli at the same time and respond to the multiple demands of the surroundings. The delta TMT assesses cognitive flexibility, which is an important aspect for dual tasking and prioritization during gait (Hobert et al., 2011, 2017). Divided attention and cognitive flexibility are also aspects of executive function, which allow us to process information from different sources and successfully carry out multiple tasks at a time. The effect of VR-based training on enhancing divided attention assessed via the TMT had been proved in individuals with Parkinson's disease (Mirelman et al., 2011). In the present study, significant within- and betweengroup differences on the TMT-B were observed in older adults individuals with MCI who received VR training. We suggest that the improvements on the TMT-B might be attributed to the interaction of the VR programs. Our IADL-based VR programs effectively facilitated complex executive function, especially visual attention, as participants repetitively practiced these functional tasks during the 12-week intervention. For example, the kitchen chef game was specifically designed to train for planning and task switching, while participants prepared food as ordered with available kitchenware and ingredients. The convenience store clerk game trained for working memory, orientation and attention as the participants retrieved and calculated the prices of the checkout items. To optimize their performance, participants needed to increase both motor and cognitive capacity. We believe performing these functional tasks with internalized real-time feedback by VR may have a greater effect on various executive functions. Another explanation is that the enjoyment and attractiveness of the VR characteristics may have increased motivation and led to more extensive training effects on executive function in the VR group than the CPC group. Overall, the traditional combined physical and cognitive training program could not offer these critical features.

Both groups showed similar time effects on gait speed and stride length in the single-task and motor dual-task gait tests. These improvements imply the nonspecific exercise benefits of these two different interventions (Liberman et al., 2017). However, only the VR group showed an improvement in cognitive dual-task gait performance after the intervention. This gait improvement may have transferred from improvements in executive functions; in particular, we observed between-group differences in the TMT-B improvement, which represents visual attention and task switching. In fact, the role of executive function in walking when combined with a secondary task has been previously shown (Hausdorff et al., 2005; Schweiger et al., 2008). Furthermore, Parikh and Shah (2017) showed a relationship between the TMT-B and dual-task gait performance among older adults. Older adults with MCI have an increased fall risk compared to cognitively normal older adults (Liu-Ambrose et al., 2008). A gait velocity less than 1.0 m/s has also been associated with an increased fall risk among community-dwelling older adults (Abellan van Kan et al., 2009). Although both of our groups had improved cognitive and motor dual-task gait velocities after the intervention, the post-intervention gait velocities were still below 1.0 m/s in both cognitive and motor dual-task walking conditions. We suggest that sustained cognitive and physical training and fall prevention education are required in routine daily treatment programs.

The DTC quantifies dual-task interference, which is the relative change in performance associated with dual tasking. Individuals with cognitive deficits may be particularly susceptible to dual-task interference because there are fewer attentional resources available for the simultaneous performance of secondary tasks. We observed intragroup and intergroup improvements in the cognitive DTC of cadence after our VR training. The main cause may have been a specific effect of VR. VR not only provides a more complex environment but also creates many opportunities to train visual attention. From the concept of capacity-sharing theory, if two attention-demanding tasks are performed at the same time in a condition of limited attentional resources, performance of at least one of the tasks will deteriorate (Ruthruff et al., 2001; Yogev-Seligmann et al., 2008). We speculate that our VR training, including many IADL-based cognitive training components in augmented scenarios, might have improved cognitive capacity and reduced the attention needed to perform the cognitive task, thereby permitting greater attention to be shifted toward performing another concurrent task (e.g., walking). Therefore, the DTC of cadence were reduced from 11% to 6% after training.

Previous studies have stated that normal older adults can improve their cognitive dual-task gait performance after combined physical and cognitive training (Eggenberger et al., 2015; Falbo et al., 2016; Wang et al., 2018a). Contrary to our expectations, our CPC group showed no time effect on the cognitive dual-task walking performance. We suggest that for older adults with MCI, improvements in SCWT can transfer to improvements in motor dual-task performance but are not strong enough to transfer to improvements in cognitive dual-task performance (walking during serial subtraction). In fact, walking while doing serial subtraction is a highly demanding task that requires alternating momentary processing capacity and filtering out all signals that are irrelevant to counting itself. Hunter et al. (2018) proposed a framework for the secondary gait testing incorporating cognitive and motor tasks in older adults with MCI. According to this framework, the difficulty level of the cognitive dual task (serial subtractions) is higher than that of the motor dual task (carrying glass of water on tray) and more easily increases the cognitive costs in older adults with MCI (Hunter et al., 2018). Makizako et al. (2012) stated that applying multicomponent exercises had no significant effect on cognitive dual-task gait performance in older adults with MCI. Our findings are in agreement with these findings, although we added cognitive training to the physical exercise program. A higher intensity, duration and challenge might be required to transfer the cognitive dual-task skills to the gait performance in our CPC group.

To our knowledge, this study is the first to examine the effects of IADL-based VR training on dual-task gait performance in older adults with MCI. Limitations of this study include the lack of an actual control group such as placebo treatment or no intervention at all, which made the mechanisms underlying our results unclear. Second, the motor task we chose was too simple to create a challenging dual-task condition, so the training effects of the motor dual-task gait were similar to the single-task gait in both groups. Third, more gait parameters, such as gait variability and stride time, may be required in future studies, which might help us to clarify the mechanism of the improvements. Fourth, the dual-task program in the VR group was performed sequentially, and the dual-task program in the CPC group was performed simultaneously. Although most articles have stated that simultaneous and sequential dual-task training were both effective in improving cognition and led to similar effects (Strouwen et al., 2017; Bruderer-Hofstetter et al., 2018), whether a different combination of methods leads to different training intensities may require further investigation. Fifth, because of the large number of statistical tests and the small sample, the p-values cannot be interpreted in conventional terms as estimates of Type I and TYpe II error probability. They should be confirmed in a subsequent properly powered trial.

# REFERENCES


# CONCLUSION

The current results suggest that executive function and motor dual-task performance could benefit from both VR-based and traditional combined physical and cognitive training in older adults with MCI. However, VR-based physical and cognitive training showed more improvements in divided attention and the cognitive DTC than traditional combined physical and cognitive training. Our physical and cognitive program, derived from IADLs, may constitute a reference for the VR training effect in older adults with MCI.

# 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

The study protocol was approved by the Institutional Human Research Ethics Committee of National Yang-Ming University and has been registered at http://www.clinicaltrials.in.th/ (TCTR20180531001 on 1-October-2018). Written informed consent was obtained.

# AUTHOR CONTRIBUTIONS

Y-YL conceived and designed the experiments. Y-YL, YC, and W-CH performed the experiments. Y-YL, I-HC, and W-CH analyzed the data. Y-YL, I-HC, and Y-JL wrote the article. All authors reviewed the manuscript.

# ACKNOWLEDGMENTS

This work was supported by grants from the Ministry of Science and Technology (MOST 106-2314-B-227-008, MOST 108-2622- H-011-001-CC3 and MOST 104-2628-E-011-006-MY3).


**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 Liao, Chen, Lin, Chen and Hsu. 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.

# Motor-Cognitive Neural Network Communication Underlies Walking Speed in Community-Dwelling Older Adults

Victoria N. Poole1,2,3,4\*, On-Yee Lo1,2,3 , Thomas Wooten<sup>4</sup> , Ikechukwu Iloputaife1,2,3 , Lewis A. Lipsitz 1,2,3 and Michael Esterman4,5,6

<sup>1</sup>Center for Translational Research in Mobility & Falls, Hinda and Arthur Marcus Institute for Aging Research, Hebrew SeniorLife, Boston, MA, United States, <sup>2</sup>Beth Israel Deaconess Medical Center, Boston, MA, United States, <sup>3</sup>Harvard Medical School, Boston, MA, United States, <sup>4</sup>Neuroimaging Research for Veterans (NeRVe) Center, VA Boston Healthcare System, Boston, MA, United States, <sup>5</sup>Geriatric Research, Education, and Clinical Center (GRECC), VA Boston Healthcare System, Boston, MA, United States, <sup>6</sup>Department of Psychiatry, Boston University, Boston, MA, United States

#### Edited by:

Helena Blumen, Albert Einstein College of Medicine, United States

#### Reviewed by:

Michele Linda Callisaya, University of Tasmania, Australia Jessie VanSwearingen, University of Pittsburgh, United States

\*Correspondence:

Victoria N. Poole victoriapoole@hsl.harvard.edu

> Received: 03 April 2019 Accepted: 12 June 2019 Published: 16 July 2019

#### Citation:

Poole VN, Lo O-Y, Wooten T, Iloputaife I, Lipsitz LA and Esterman M (2019) Motor-Cognitive Neural Network Communication Underlies Walking Speed in Community-Dwelling Older Adults. Front. Aging Neurosci. 11:159. doi: 10.3389/fnagi.2019.00159 While walking was once thought to be a highly automated process, it requires higher-level cognition with older age. Like other cognitive tasks, it also becomes further challenged with increased cognitive load (e.g., the addition of an unrelated dual task) and often results in poorer performance (e.g., slower speed). It is not well known, however, how intrinsic neural network communication relates to walking speed, nor to this "cost" to gait performance; i.e., "dual-task cost (DTC)." The current study investigates the relationship between network connectivity, using resting-state functional MRI (rs-fMRI), and individual differences in older adult walking speed. Fifty participants (35 females; 84 ± 4.5 years) from the MOBILIZE Boston Study cohort underwent an MRI protocol and completed a gait assessment during two conditions: walking quietly at a preferred pace and while concurrently performing a serial subtraction task. Within and between neural network connectivity measures were calculated from rs-fMRI and were correlated with walking speeds and the DTC (i.e., the percent change in speed between conditions). Among the rs-fMRI correlates, faster walking was associated with increased connectivity between motor and cognitive networks and decreased connectivity between limbic and cognitive networks. Smaller DTC was associated with increased connectivity within the motor network and increased connectivity between the ventral attention and executive networks. These findings support the importance of both motor network integrity as well as inter-network connectivity amongst higher-level cognitive networks in older adults' ability to maintain mobility, particularly under dual-task (DT) conditions.

Keywords: older adults, gait speed, dual-task cost, resting-state fMRI, functional connectivity

# INTRODUCTION

Walking speed is now widely accepted as a clinically meaningful marker of general function and wellbeing in older adults (Cesari et al., 2005; Afilalo et al., 2010; Verghese et al., 2011). For this reason, it is frequently measured within the geriatric clinic and has been referred to as the ''sixth vital sign'' (Fritz and Lusardi, 2009; Middleton et al., 2015). However, many factors contribute to declines in walking speed as we age (Tiedemann et al., 2005). For instance, although walking slows with musculoskeletal and peripheral nervous system disorders, it also slows with central nervous system disorders (Hajjar et al., 2009). Previous studies support the premise that walking engages several neurocognitive and neurovascular processes, including gait planning and initiation, sensory-motor integration, memory, the ability to detect and accommodate altered peripheral neuromusculoskeletal function and neurovascular coupling (Halliday et al., 1998; Jahn et al., 2004, 2008; Seidler et al., 2010; Sorond et al., 2011; Takakusaki, 2017).

The cognitive nature of walking is particularly important for ''real-world'' mobility. With or without physical limitations, successful mobility in living environments requires higher-level cognitive processes like attention (Woollacott and Shumway-Cook, 2002), executive functioning (Yogev-Seligmann et al., 2008), self-reference (Sholl, 2001), motor control (Winter, 2009), and caution (Donoghue et al., 2013). A person must constantly process the information within his or her environment to navigate within it, often whilst performing additional tasks, like talking or texting. When attempting to perform these additional tasks, the brain must shift and prioritize attention (Yogev-Seligmann et al., 2008). This typically results in a ''cost'' of slower speed (or poorer task performance) and has been associated with an increased risk of falls (Beauchet et al., 2008), accidents (Neider et al., 2011), and other adverse outcomes (Montero-Odasso et al., 2017). To study this real-world phenomenon, both normal walking and dual-task (DT) paradigms have been implemented within the research laboratory (Verhaeghen et al., 2003) and several studies have uncovered relationships between abnormal walking characteristics and poorer cognition (Li et al., 2018), as measured by cognitive and neuropsychological assessments.

Although these findings provide a clear basis for the premise that higher levels of neurocognitive function underlie normal gait, limitations exist in the actual study of brain function. Most studies have either utilized wearable techniques, like functional near-infrared spectroscopy (fNIRS), or have employed gait imagery paradigms (Hamacher et al., 2015) to study gait, since it is impossible to directly assess walking as a task within the MRI scanner. In addition to brain areas traditionally associated with locomotion (e.g., M1, SMA, premotor cortex, and cerebellum), several studies implicate higher-order and frontoparietal areas in walking (Hanakawa et al., 1999; Allali et al., 2013; Doi et al., 2013; Blumen et al., 2014; Jor'dan et al., 2017). However, far fewer studies have used fMRI to relate stable properties of brain function and communication to individual differences in walking speed. Yuan et al. (2015) were the first to investigate resting-state fMRI (rs-fMRI) associations and found that the connectivity within the sensorimotor, visual, vestibular, and left frontoparietal (i.e., executive control) networks were associated with normal walking and walking while talking. Additionally, Lo et al. (2017) have shown walking relationships within the frontoparietal and attention networks in cognitively impaired older adults.

More studies are needed that explicitly characterize how individual differences in connectivity within and across brain networks relate to ''normal'' walking, DT walking, and associated dual-task costs (DTCs). To address this need, we have utilized rs-fMRI in community-dwelling older adults from the MOBILIZE Boston Study to investigate normal walking and walking during serial subtraction. We hypothesized that greater within and between sensorimotor network connectivity would be associated with faster preferred walking speed. DT walking (and cost), however, would be further associated with the brain network interactions that are related to executive function and attention, as they are more engaged by walking while performing the dual serial subtraction cognitive task.

# MATERIALS AND METHODS

# Participants

Seventy-six older adults (50 females, 84.5 ± 4.3 years) were recruited from the Maintenance of Balance, Independent Living, Intellect and Zest in the Elderly of Boston (MOBILIZE Boston Study; MBS) cohort to undergo a gait assessment and MRI protocol. Original recruits to the MBS were Boston area residents 70 + years of age (or >65 if living with an already enrolled participant), able to walk 20 feet without personal assistance, had no history of neurological, mental illness, or stroke, and at least a 12th grade education. Mini-Mental State Examination scores collected two years prior to MRI ranged from 19 to 30. A full description of the greater MBS protocol is provided elsewhere (Leveille et al., 2008). To participate in the current study, all participants were further required to perform a gait assessment while simultaneously meeting eligibility criteria for a 60-min MRI. The Hebrew SeniorLife and VA Boston Healthcare System institutional review boards approved this protocol and written consent was required prior to study participation.

# Study Design

Participants took part in two visits. The first visit to the Hebrew SeniorLife Clinical Research Center included a medical history evaluation and gait assessment, where participants were asked to walk over a 16-foot GaitRite (CIR Systems Inc., Havertown, PA, USA) mat during each of two conditions: quietly and during a serial subtraction task. For the preferred walking condition, participants made six separate passes at their preferred pace without interruption, starting and ending approximately 4 feet from the mat. For the DT condition, participants walked while verbally subtracting threes from a randomly given 3-digit number. Walking speed for both conditions was derived from the GaitRite-measured timing and location of individual steps in meters per second (m/s). The primary outcomes were the mean gait speed for each condition calculated by averaging across passes and normalizing by participant height. DTC was calculated as the percent change in speed relative to the quiet condition, such that the higher the cost, the slower the serial subtraction walk. Serial-subtraction task accuracy was not assessed.

Approximately 10 days later, participants completed an MRI protocol at the VA Boston Healthcare System. During this session, participants performed a resting-state scan, during which they were asked to keep their eyes open. A gradient-echo echo-planar sequence was performed with the following parameters: TR = 3,000 ms, TE = 26 ms, flip angle = 90◦ , 34 slices at 1.5 mm, 64 × 64 matrix, and 120 volumes. A T1-weighted MPRAGE scan (T1 = 1,000 ms, TR = 2.73 ms TE = 3.31 ms, flip angle = 7◦ , 128 slices at 1.3 mm thickness, 256 × 256 matrix) was also collected for whole-brain high-resolution anatomy. These neuroimaging sessions were performed with two 3T Siemens MRI scanners (TIM Trio, n = 18; PrismaFit , n = 32) using a 12-channel head coil. To account for potential interscanner differences, scanner was included as a covariate in statistical models.

# Functional MRI

Functional data were processed using AFNI (Cox, 1996). Pre-processing steps included: the removal of the first three volumes, time-shifting, volume registration, alignment to high-resolution anatomy, warping into Talairach space, 8-mm kernel smoothing, resampling to 3 × 3 × 3 mm resolution, and scaling to a percentage of the mean. Data were then band-pass filtered from 0.01 to 0.08 Hz and entered into a general linear model to remove the effects of 6◦ of motion, their derivatives, nuisance CSF, white matter, and global signal. Time points were censored and participants were excluded for excessive motion if they demonstrated greater than 0.5 mm in sudden movement for more than 20% of the scan.

A previously-defined cortical parcellation was then applied to the whole-brain GLM residuals, representing the ''cleaned'' time series. Briefly, an atlas from Schaefer et al. (2018) was used to parse the cortex into 100 regions that were co-registered with the seven functionally-connected cortical networks identified by Yeo et al. (2011). The seven included the visual (VIS; 17 regions), sensorimotor (SOM; 14 regions), dorsal attention (DAN; 15 regions), ventral attention (VAN; 12 regions), limbic (LIM; 5 regions), executive control (ECN; 13 regions), and default mode (DMN; 24 regions) networks. The average time series were extracted from each brain region and correlated in pairs for a total of 4,950 possible pairwise correlations. To calculate both within and between functional connectivity measures at the network-level, the above estimates were Fisher's z-transformed, grouped, and averaged according to their within- and between-network pairs. This resulted in a total of seven within- and 21 between-network estimates for use in linear regression analyses.

# Statistical Analyses

A matched-pairs t-test was conducted to evaluate the change in walking speed between the normal and DT conditions. Non-neural characteristics (e.g., age, sex, body mass index (BMI), type 2 diabetes, hypertension, and arthritis) suspected to influence walking outcomes were evaluated using simple linear regression and rank sums tests. The significant covariates were then included in multiple linear regression models to predict: normal walking speed, DT speed, and the resulting cost from the average network pairs, along with scanner assignment.

Since we performed 28 network-pair analyses for each of the three conditions, we then utilized a permutation procedure to determine the probability of our observed number of significant gait-brain associations. This was done by randomization of individual walking speeds and clinical characteristics, conducting the multiple linear model for each network pair, and determining the number of chance ''significant'' (i.e., p = 0.05) associations. This was repeated for a total of 10,000 iterations for each outcome. All analyses were performed using MATLAB (2014a; Mathworks, Natick, MA, USA) and R (R Core Team, 2018).

# RESULTS

Of the 76 participants that completed the MRI protocol, 50 were included in the analyses. Reasons for exclusion included excessive motion (n = 8), missing data or incomplete scan (n = 5), reported stroke or incidental findings (n = 6), and a score of less than 25 on their most recent MMSE administration (n = 7). Participant demographics, clinical characteristics, and gait measurements for the final subset are listed in **Table 1**.

# Gait Assessment

Though participants walked relatively fast at their preferred pace (1.1 ± 0.3 m/s), they walked significantly slower during the serial subtraction task (0.9 ± 0.3 m/s; matched pairs t(49) = 13.6, p < 0.0001). This resulted in a DTC of 19 ± 11%.

Walking outcomes were then associated with clinical characteristics for covariate consideration. As suspected, participant age was associated with both preferred (β = −0.45, p = 0.001) and DT walking speeds (β = −0.44, p < 0.002) and tended to be associated with DTC (β = 0.26, p < 0.07). No other


Data are expressed in mean ± standard deviation or percentage. †Most recent scores were collected within approximately 2 years of gait and MRI assessment. ‡Preferred (Pref) and dual-task (DT) walking outcomes are reported as mean ± SD (range), with and without normalizing by height (inches−<sup>1</sup> ).

significant associations with clinical variables (see ''Materials and Methods'' section) were found.

# Resting-State fMRI

Multiple uni-network linear regression analyses were then performed to assess walking associations with within-network (n = 7 networks) and between-network (n = 21 network pairs) connectivity, after adjusting for the effects of scanner and participant age. In two outcomes, the number of observed significant gait-brain associations exceeded the probability of chance based on the randomization procedure: 9 of 28 network-pair models were associated with preferred walking (p = 0.0027) and 10 were associated with DT walking (p = 0.0011) speeds. Only two models were associated with DTC (p = 0.36), i.e., 36% of randomized iterations had two or more significant correlations by chance. These models are characterized below.

### Within-Network Associations

Normal walking was positively associated with sensorimotor (SOM; βadj = 0.31, p = 0.02) and dorsal attention (DAN; βadj = 0.29, p = 0.03) network connectivity, such that the greater the within-network connectivity, the faster the participants walked at preferred pace. However, preferred walking speed was negatively associated with the visual network (VIS; βadj = −0.31, p = 0.02). Walking while simultaneously performing the serial subtraction task (i.e., dual task) was positively associated with within-network SOM (βadj = 0.43, p = 0.003), DAN (βadj = 0.34, p = 0.03), and ventral attention (VAN; βadj = 0.34, p = 0.03) networks. DTC was associated with the SOM (βadj = −0.31, p = 0.03) network, such that the greater the within-network connectivity, the lower the cost (slowing) to gait speed during the DT. These associations are depicted in **Figure 1**.

#### Between-Network Associations

**Figure 2** illustrates the associations of inter-network averages with preferred (lower triangle) and DT (upper triangle) walking speeds. These associations were found to be rather consistent across the two conditions (see **Figure 2**), which is not surprising since the walking speeds themselves were strongly correlated (r = 0.92, p < 0.0001). In both conditions, faster walking was associated with greater DAN connectivity with the SOM (βadj > 0.26, p < 0.05) and the VAN (βadj > 0.27, p < 0.04). Faster walking while performing the serial subtraction task was further associated with greater communication between the VAN and executive control network (ECN; βadj = 0.33, p = 0.01). Interestingly, faster walking was also associated with lesser between-network limbic connectivity with motor and cognitive networks, including SOM, DAN, VAN, and ECN (βadj < −0.29, p < 0.05). Lower DTC was solely associated with increased VAN-ECN connectivity (βadj = −0.29, p = 0.04; not shown). No other significant between-network associations were found.

# DISCUSSION

In the current study, we uncovered associations between neural network connectivity and walking speed in a sample of

FIGURE 2 | Heat map reflecting between-network fMRI connectivity associations (standardized betas) with preferred walking (left of diagonal) and walking while dual-tasking (right of diagonal). Tiles are color-coded by strength (positive = blue, negative = red). Associations exceeding p = 0.05 threshold (uncorrected) are presented with asterisk (<sup>∗</sup> ).

community-dwelling older adults from the MOBILIZE Boston Study. Specifically, we found that stronger cognitive and motor network connectivity was associated with faster walking in older adults, while greater communication with the limbic network was associated with slower walking. These findings were generally evident with and without a simultaneous task.

When we investigated individual differences in brain connectivity and preferred (i.e., quiet) walking speed, we found that faster walking speed was associated with increased connectivity within the motor and dorsal attention networks and decreased connectivity within the visual network. These findings align well with literature, as these networks include many of the cortical regions observed in task-based fMRI activation studies of gait speed, including the pre- and postcentral gyrus, inferior frontal gyrus, superior parietal lobes, and occipital areas (Hamacher et al., 2015). Further, stronger between-network connectivity (i.e., between motor and dorsal as well as ventral attention networks) was associated with faster walking. This provides evidence that these networks do not work in isolation, but rely upon one another in the engagement of motor, sensory and visual functions, as well as a balance of top-down (i.e., dorsal) and bottom-up (i.e., ventral) goal-directed attention (Corbetta and Shulman, 2002; Kincade et al., 2005). While we did not predict negative associations with the visual network, other studies suggest that this network has altered connectivity with older age (Goh, 2011; Geerligs et al., 2015), which could impact motor control and gait. In our DT paradigm, where individuals have to shift attention between serial subtraction and walking, we observed ''further relationships'' with the ventral attention network, suggesting a greater necessity for re-orienting of attention (Corbetta and Shulman, 2002). Reorienting of attention is potentially necessary for successful dual task performance, as supported by recent studies where VAN-DAN communication and VAN-DMN suppression were associated with better distractor suppression during visual search (Kelly et al., 2008; Poole et al., 2016). It should also be noted that the VAN spatially overlaps with the vestibular network identified in Yuan et al. (2015), which is also associated with walking speed during resting state fMRI.

The current study also provides evidence that the executive control network is associated with faster walking, especially during the DT condition. Previously, abnormal executive function, which is essential for attentional processes and the ability to plan, organize, and multi-task, had been linked with poorer dual tasking (Ohsugi et al., 2013), slower gait speed (Cohen et al., 2016), and falls (Mirelman et al., 2012) in older adults. While we have previously shown that walking speed is associated with ECN brain activation and structural connectivity (Jor'dan et al., 2017; Poole et al., 2018), we now suggest that how this network interacts with other neighboring networks may be essential to cognitive functioning in walking. To our knowledge, we are the first to explore all of these network relationships in cognitively healthy older adults.

On the other hand, slower walking was associated with greater widespread connectivity of the limbic network with motor and cognitive networks. The cortical areas contained in this network are associated with memory, arousal, and emotion (Agosta et al., 2012; Rolls, 2019). Prior imaging studies have found the limbic network to be hyper-connected in individuals with lesser motor expertise (Milton et al., 2004) and increased cognitive impairment (Badhwar et al., 2017). Thus, it is possible that increased communication with this network may lead to interference with motor and/or cognitive processes. It may also indicate active caution and/or fear while walking (Pannekoek et al., 2013).

# Clinical Implications and Future Directions

This study elucidates the underlying brain networks associated with preferred and DT walking performance, which are strong predictors of falls in older adults (Verghese et al., 2002; Quach et al., 2011). These network interactions could inform interventions that non-invasively target these brain networks (i.e., non-invasive brain stimulation) or their functions (i.e., motor-cognitive training) for individuals with mobility impairments. Further investigation of these networks may provide information on the progress of dysfunction, risk of falls, or the efficacy of rehabilitation.

However, these findings do not confirm the specific neural resources of utmost importance to mobility, especially in cases of ''cognitive reserve'' when alternate brain regions are recruited in the presence of structural degeneration (Stern, 2002; Venkatraman et al., 2010). Therefore, future studies should investigate the relative contributions of brain structure and function, as well as consider other clinical predictors of slow gait (Rosso et al., 2013). Furthermore, it is important to determine the reliability and generalizability of these findings in other older adult populations, in order to determine the best avenue for early intervention.

# Limitations

The current study has limitations. First, although this MOBILIZE Boston Study sample is a relatively small ''representative'' older adult population living with several comorbidities, previously collected Mini-Mental State Examination scores (administered within approximately 2 years of gait and MRI) indicate a range of cognitive health. To minimize the contributions of overt cognitive impairment to declines in walking speed, we only included those with a past MMSE of 25 or greater and no self-reported diagnosis of dementia. However, these results should be interpreted with caution since more recent cognitive measures were not collected.

Another limitation is that accuracy during the serial subtraction task was not explicitly emphasized in instructions, neither was it measured to allow calculation of a cognitive DTC. However, previous studies suggest a tendency for participants to prioritize focus on the cognitive task at the expense of the other (i.e., walking; Krampe et al., 2011). Nevertheless, without performance measures, it is unclear the degree to which serial subtraction task difficulty and attention prioritization contributed to the observed declines in gait speed without a baseline or report of math anxiety. As a result, we generally attribute slower walking during serial subtraction to multitasking, such that those with greater cognitive connections have greater ability to maintain walking speed while performing an additional task.

Finally, with regard to the rs-fMRI analysis, although many approaches exist [e.g., seed-based connectivity, independent component analysis (ICA), principal component analysis (PCA)], here we utilize a brain atlas from Schaefer et al. (2018) to parse the brain into 100 regions. This atlas was derived from local gradient and global similarity approaches and is co-registered with a widely used cortical atlas of seven functionally-connected networks (Yeo et al., 2011). The seven-network atlas, which was derived from an independent sample of 1,000 healthy individuals, is relatively coarse, but has been well characterized in health and disease. By parsing the brain according to the Schafer atlas, we are able to retain the ability to reduce the dimensionality of rs-fMRI data, and estimate functional connectivity within and across each brain network, while increasing the resolution of the aforementioned seven networks (though averaged into 28 within and between-network ''pairs''). This approach also allowed us to emphasize network-level conclusions and compare our results to other studies of older adult mobility. However, future studies should consider using surface-based approaches and complementary structure analyses to investigate individual differences since volumetric approaches are not sensitive to the heterogeneity in brain anatomy, especially across older adults with varying degrees of brain atrophy (Long et al., 2012).

# CONCLUSION

This study reports associations between brain functional connectivity and walking outcomes in a sample of communitydwelling older adults. Rs-fMRI analyses revealed that walking at faster preferred speed is associated with stronger connectivity within and between sensorimotor and dorsal attention networks, networks associated with motor control and goal-directed attention. With an added serial subtraction dual task, we found that stronger connectivity between attention and executive networks is associated with faster walking speed. However, stronger connectivity between these networks and the limbic network is associated with slower walking during both tasks.

# REFERENCES


# 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 Hebrew SeniorLife and VA Boston Healthcare System institutional review boards. All subjects gave written informed consent in accordance with the Declaration of Helsinki.

# AUTHOR CONTRIBUTIONS

VP: data processing, analysis, and interpretation; wrote the manuscript. O-YL: analysis, interpretation. TW and II: project management; acquisition of data. LL and ME: study concept and design, study supervision, critical revisions of the manuscript for important intellectual content.

# FUNDING

This work was supported by the National Institute on Aging (Grant Nos. R01-AG041785, T32-AG023480); the United States Department of Veterans Affairs Clinical Sciences R&D (CSRD) Service (Merit Review Award I01CX001653); and a KL2/Catalyst Medical Research Investigator Training award (an appointed KL2 award) from Harvard Catalyst—The Harvard Clinical and Translational Science Center (National Center for Advancing Translational Sciences, National Institutes of Health Award KL2 TR002542 and UL 1TR002541). The content is solely the responsibility of the authors and does not necessarily represent the official views of Harvard Catalyst, Harvard University, its affiliated academic healthcare centers, or the National Institutes of Health.


**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 Poole, Lo, Wooten, Iloputaife, Lipsitz and Esterman. 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.

# Age-Related Adaptations of Lower Limb Intersegmental Coordination During Walking

Mathieu Gueugnon1,2, Paul J. Stapley <sup>3</sup> , Anais Gouteron2,4,5, Cécile Lecland<sup>6</sup> , Claire Morisset 1,2, Jean-Marie Casillas 1,2,4,5, Paul Ornetti 1,2,4,7 and Davy Laroche1,2,4 \*

1 INSERM, CIC 1432, Module Plurithematique, Plateforme d'Investigation Technologique, Dijon, France, <sup>2</sup> CHU Dijon-Bourgogne, Centre d'Investigation Clinique, Module Plurithématique, Plateforme d'Investigation Technologique, Dijon, France, <sup>3</sup> Neural Control of Movement Laboratory, Faculty of Science, Medicine and Health, School of Medicine, University of Wollongong, Wollongong, NSW, Australia, <sup>4</sup> INSERM, UMR 1093-CAPS, Université de Bourgogne Franche Comté, UFR des Sciences du sport, Dijon, France, <sup>5</sup> Department of Physical Medicine and Rehabilitation, Dijon-Bourgogne University Hospital, Dijon, France, <sup>6</sup> Damartex Group, Roubaix, France, <sup>7</sup> Department of Rheumatology, Dijon University Hospital, Dijon, France

#### Edited by:

Helena Blumen, Albert Einstein College of Medicine, United States

#### Reviewed by:

Simona Ferrante, Politecnico di Milano, Italy Chi-Wen Lung, Asia University, Taiwan

\*Correspondence: Davy Laroche davy.laroche@chu-dijon.fr

#### Specialty section:

This article was submitted to Biomechanics, a section of the journal Frontiers in Bioengineering and Biotechnology

Received: 13 February 2019 Accepted: 04 July 2019 Published: 17 July 2019

#### Citation:

Gueugnon M, Stapley PJ, Gouteron A, Lecland C, Morisset C, Casillas J-M, Ornetti P and Laroche D (2019) Age-Related Adaptations of Lower Limb Intersegmental Coordination During Walking. Front. Bioeng. Biotechnol. 7:173. doi: 10.3389/fbioe.2019.00173 Lower-limb intersegmental coordination is a complex component of human walking. Aging may result in impairments of motor control and coordination contributing to the decline in mobility inducing loss of autonomy. Investigating intersegmental coordination could therefore provide insights into age-related changes in neuromuscular control of gait. However, it is unknown whether the age-related declines in gait performance relates to intersegmental coordination. The aim of this study was to evaluate the impact of aging on the coordination of lower limb kinematics and kinetics during walking at a conformable speed. We then assessed the body kinematics and kinetics from gait analyses of 84 volunteers from 25 to 85 years old when walking was performed at their self-selected speeds. Principal Component Analysis (PCA) was used to assess lower-limb intersegmental coordination and to evaluate the planar covariation of the Shank-Thigh and Foot-Shank segments. Ankle and knee stiffness were also estimated. Age-related effects on planar covariation parameters was evaluated using multiple linear regressions (i.e., without a priori age group determination) adjusted to normalized self-selected gait velocity. Colinearity between parameters was assessed using a variation inflation factor (VIF) and those with a VIF < 5 were entered in the analysis. Normalized gait velocity significantly decreased with aging (r = −0.24; P = 0.028). Planar covariation of inter-segmental coordination was consistent across age (99.3 ± 0.24% of explained variance of PCA). Significant relationships were found between age and intersegmental foot-shank coordination, range of motion of the ankle, maximal power of the knee, and the ankle. Lower-limb coordination was modified with age, particularly the coordination between foot, and shank. Such modifications may influence the ankle motion and thus, ankle power. This observation may explain the decrease in the ankle plantar flexor strength mainly reported in the literature. We therefore hypothesize that this modification of coordination constitutes a neuromuscular adaptation of gait control accompanying a loss of ankle strength and amplitude by increasing the knee power in order to maintain gait efficiency. We propose that foot-shank coordination might represent a valid outcome measure to estimate the efficacy of rehabilitative strategies and to evaluate their efficiency in restoring lower-limb synergies during walking.

Keywords: gait analysis, aging, panar covariation, biomechanics, locomotor control

# INTRODUCTION

Human walking is a common task with efficient motor control. Synergic muscle activation for the control of limb movements requires the integration of inputs from the central nervous system and feedback from proprioceptive sensors in the muscles, tendons, and limbs. In healthy persons, the neural command ensures a rhythmic, stable gait with a highly consistent intersegmental coordination, and overall walking patterns. This coordination, corresponding to the process of mastering redundant degrees of freedom of the body into a controllable system, allows the efficiency of gait by maintaining dynamic equilibrium, and the lowest energetic cost during gait (Bernstein, 1967; Lacquaniti et al., 1999). The movement coordination during gait might therefore reflect neuro-muscular synergies. While an inability to modulate the intersegmental coordination may induce gait deviations, it might also provide insights into the organization and adaptation of gait patterns with pathology or aging (Winter et al., 1990).

Declining mobility and gait performance is one of the major functional hallmarks of aging (Boyer et al., 2017). Age-related differences in gait performance include a decrease in gait speed, a reduction in step length, and/or an increased cadence (Mcgibbon and Krebs, 2001; Lewis and Ferris, 2008). These changes are associated with impaired balance control, a reduction of muscle strength, and mass as well as an increase of the energy cost of walking (Sepic et al., 1986; Winter et al., 1990; Judge et al., 1996; Kerrigan et al., 2000, 2001; Pavol et al., 2002; Cofré et al., 2011; Frimenko et al., 2015). As a result, the coordination was impaired with aging and linked to a history of falls in the past year (Hutin et al., 2011; Chiu and Chou, 2012; Ghanavati et al., 2014; James et al., 2017; Hafer and Boyer, 2018 Hutin et al., 2011; Ghanavati et al., 2014. However, these studies used standard frequency-decomposition methods to evaluate the intersegmental coordination during walking (i.e., continuous relative phase and coding vector). While these methods are well documented, they do not provide a complete overview of the gait processing of the lower limb intersegmental coordination due to their analysis of singular parameters. Indeed, it is known that during human walking, the lower-limb coordination is controlled through a coupling of all the segments (thigh, shank, and foot) in order to simplify the spatiotemporal control of locomotion and equilibrium (Borghese et al., 1996; Lacquaniti et al., 2002). The elevation angles of these segments are consequently related. When lower-limb segment rotations (temporal changes in the elevation angles) are plotted one vs. each others, they covary along a plane and constitute a loop [i.e., covariation plane, (Ivanenko et al., 2008; Lacquaniti et al., 2012a)]. Principal component analysis (PCA) was used to analyse that plane and when applied produced three components. In normal walking, the first and the second components define the robustness of the planarity of the loop whereas the third component defines its orientation. In this context, the properties of the covariation plane provide insights about how the central nervous system controls the limbs during walking and therefore might reflect the adaptation of the neural and neuromuscular systems with aging. In particular, the work of Lacquaniti and others (for a details see Ivanenko et al., 2006; Lacquaniti et al., 2012b) postulated that planar covariation may provide a link between neuromuscular control and mechanics of gait by matching the control of lower limb muscle patterns to those of the body's center of mass (Bleyenheuft and Detrembleur, 2012). Consequently, this study aimed to evaluate the impact of aging on the coordination of lower limb kinematics and kinetics during walking at comfortable speed using the planar covariation of elevation angles. We hypothesized that the planar covariation of elevation angles should be modified throughout the lifespan in order to adapt the locomotor pattern to the constraints of aging. To this end, we assessed effects of walking speed and age on the pattern and variability of lower limb intersegmental coordination in a cohort of healthy subjects from 25 to 85 years old.

# MATERIALS AND METHODS

# Participants

Eighty-four volunteers (51 women and 33 men) from 28 to 85 years old were recruited from a previous asymptomatic cohort (clinical trial registration: NCT02042586) to participate in this prospective study. All showed no symptomatic musculoskeletal, neurological, or cardiovascular disease. Exclusion criteria were significant pain, ankle, hip or foot disorders, chronic back pain, Alzheimer's disease, Parkinson's disease, motor neuron disorders, non-stabilized diabetes mellitus, cardiac or respiratory insufficiency, and any inability to understand the procedures. The study protocol was approved by the local ethics committee (CPP Est I, Dijon, France). The study was conducted in compliance with the principles of Good Clinical Practice and the Declaration of Helsinki, and all patients gave their informed consent.

# Task and Procedure

Participants were asked to walk 10 times barefoot while following a straight-line path, 10 meters in length traced on the floor. After each walking trial, they were asked to return to the starting point. They were instructed to adopt a natural and comfortable gait speed, as if they walked "along the street." Lower body kinematics (i.e., movements of pelvis, hips, knees, and ankles sagittal, frontal, and transverse plane) during walking were measured using an 8 optoelectronic camera motion capture system (Vicon MX, Vicon <sup>R</sup> , Oxford, UK) sampling at 100 Hz. The marker set used, the Plug-in-Gait marker set (Davis et al.,

1991), was composed of 16 reflective markers positioned on specific anatomical landmarks on the lower limb (see Laroche et al., 2014 for placement on a representative participant).

# Data Analysis

Marker trajectories were recorded by the optoelectronic camera allowing to reconstruct embedded coordinate systems associated to each rigid body segment (pelvis, femur, tibia, and foot) defining then a complete 3-dimensional model of the lower limb. To access kinetics data (i.e., joint moment and power), ground reaction forces were also recorded with two force platforms (AMTI <sup>R</sup> , USA) sampled at 1,000 Hz (**Figure 1A**).

Marker trajectories were interpolated with Woltring polynomial and then filtered with a low pass zero phase shift Butterworth filter with a respective cut off frequency of 10 Hz. Similarly, ground reaction forces were filtered with a low pass zero phase shift Butterworth filter with a respective cut off frequency of 50 Hz (van den Bogert and de Koning, 1996). Displacements of the center of mass (CoM), joint kinematics and kinetics were calculated with the Nexus software (Vicon <sup>R</sup> , Oxford, UK) using inverse dynamic on the Plug-in-Gait model. The gait events were detected using a method proposed by Zeni et al. (2008) and expressed by gait cycle. Briefly, this method defines the heel-strike and the toe-off as the instant where the foot, respectively, begins to move backward and forward in the pelvis frame.

We chose the most representative variables of the gait kinematics, kinetics, and stiffness that could associated with neuromuscular adaptation during gait (Lacquaniti et al., 2012b; Herssens et al., 2018). We first computed amplitude of displacements of CoM in the vertical plane (AmpCoM). We then extracted the gait speed (v), step width, step length, and computed the Froude number Fr = v²/g.L with g as the acceleration due to gravity and L, the subject's leg length. This parameter allows normalizing the velocity across participants (Saibene and Minetti, 2003). We also computed variablility of the step length and step width (Herssens et al., 2018). From the joint kinematics, we computed the range of motion (ROMHip, ROMKnee, and ROMAnkle) during walking, defined as the sum of the peak flexion and extension, or peak dorsal and plantar flexion (**Figure 1B**). From the joint kinetics, we extracted the maximal positive power during gait for each joint (PHip, PKnee, and PAnkle) and computed the associated joint moments normalized by the subject's body weight (Mjoint) (**Figure 1C**). We then estimated the stiffness of the knee and ankle joints (KKnee and KAnkle) using the torsional spring model (Farley and Morgenroth, 1999; Kuitunen et al., 2002). The stiffness (Nm.kg−<sup>1</sup> .deg−<sup>1</sup> ) was calculated as a change in the joint moment divided by the change in joint angular displacement in the middle of the ground contact phase (Hobara et al., 2013; **Figure 1C**).

The spatio-temporal structure of the lower limb intersegmental coordination was evaluated using a principal component analysis. Three segments per lower limb were taken into account: the feet (defined as the virtual lines joining the marker located in the second metatarsal head and the marker


Fr, Froude number corresponding to a normalized self-selected gait speed; AmpCoM, amplitude of displacements of the center of mass in the vertical plane; ROMHip, range of motion of the hip in the sagittal plane; ROMKnee, range of motion of the knee in the sagittal plane; ROMAnkle, range of motion of the ankle in the sagittal plane; PHip, maximal positive hip power; PKnee, maximal positive knee power; PAnkle, maximal positive knee power; KKnee, knee stiffness; KAnkle, ankle stiffness; VarCovPlane, explained variance of the covariation plane; µ1, shank-thigh coordination; µ3, foot-shank coordination; SD, Standard Deviation.

located in the lateral malleolus), the shanks (defined as the virtual lines joining the marker located in the lateral malleolus and the marker located in the lateral femoral condyle), and the thighs (defined as the virtual lines joining the marker located in the lateral femoral condyle and the marker located in antero-superior iliac spine). Such analyses were computed randomly for one lower limb independently by means of the covariance matrix of the angular variation of foot, shank, and thigh segments as described previously (Borghese et al., 1996; Bianchi et al., 1998; Lacquaniti et al., 2002; Ornetti et al., 2011). The first two principal eigenvectors, accounting for almost 99% of data variance, correspond to the "covariation plane" (VarCovPlane). The temporal coupling between the elevation angles of the shank and the thigh segments (µ1) is illustrated with the first eigenvector and its projection on the thigh axis. The temporal coupling between the elevation angles of foot and shank segments were given by the third eigenvector (µ3) normal to the plane. All these parameters were obtained for each gait cycle allowing to obtain two values per subject (mean and standard deviation).

# Statistical Analysis

Data analysis was performed with Stata statistical software (version 15.1, Statacorp, College station TX, USA). We first applied univariate correlation between age and either normalized gait speed or planar covariation indices (µ1, µ3). We applied stepwise regression analysis to identify the most relevant variables associated with age. Entry criterion of the three stepwise procedures was set at 0.20 and stay criterion at 0.10. The procedure stopped when no more variables satisfied the previous criteria. In order to validate the model, the colinearity between variables and the residuals homogeneity were checked, respectively, by the calculation of the Variance Inflation Factor (VIF) and the read of residuals vs. predicted values graphic. A VIF value higher than 5 enabled us to admit colinearity between variables (Kutner et al., 2004), those variables were then removed from the model, if necessary. Data from the gait analysis were entered as follows into the multivariate stepwise linear regression model:


TABLE 2 | Multivariate linear regression model between kinematics and kinetics variable and age.


ROMAnkle, range of motion of the ankle in the sagittal plane; PKnee, maximal positive knee power; PAnkle, maximal positive knee power; µ3, foot-shank coordination; 95%CI, 95% confidence interval; VIF, variance inflation factor.


Statistical significance was defined as P < 0.05. The parameter estimates, 95% confident interval and partial R-square are given and compared to Cohen's suggestions (Cohen, 1992).

# RESULTS

The characteristics of participants are summarized in **Table 1**.

We performed univariate correlations between normalized self-selected gait speed (Fr) and age and planar covariation (VarCovPlane) and age. A negative significant weak correlation (r = −0.24; P = 0.028) was found between normalized gait speed and age of the participants. Furthermore, we performed a multiple stepwise linear regression analysis between age and parameters computed from gait analysis (see methods for details). The regression model provided a moderate explanation of the variance (F = 6.20; adjusted R² = 0.49; p < 0.001) and revealed no significant relationship between age and normalized gait speed (p = 0.11). However, significant relationships were found between age and range of motion of the ankle, maximal power of the knee, and the ankle (**Table 2**; **Figure 2**), percentage of planar covariation and the intersegmental foot-shank coordination (**Table 2**; **Figure 3**).

# DISCUSSION

The present study aimed to assess the impact of non-pathological aging on the coordination of lower limb kinematics and kinetics during walking at conformable speed using the planar covariation of elevation angles. We showed the adaptation of planar covariation of lower-limb segments throughout the lifespan and the related kinematics and kinetics during walking.

Our results are first consistent with previous studies that showed a significant effect of aging on gait performance (Boyer

FIGURE 2 | (A) Mean (solid lines) and standard deviation (dotted lines) waveforms of sagittal ankle joint excursions for people of 25–49 (light gray)/50–64 (medium gray)/65–85 (dark gray) years old. (B) Mean waveforms of knee power for people of 25–49 (light gray)/50–64 (medium gray)/65–85 (dark gray) years old. (C) Mean waveforms of sagittal ankle power for people of 25–49 (light gray)/50–64 (medium gray)/65–85 (dark gray) years old. Relationships between age and ankle range of motion (D), knee maximal power (E), ankle maximal power (F). Partial R² and p-Value are provided. We choose to represent 3 classes of age in order to highlight change due to age.

et al., 2017), especially walking speed. a significant relationship between aging and normalized gait speed was found which would attest to a decline in speed with age. Interestingly, this relationship was not evident in the multivariate model indicating that confounding variables may have been present. Indeed, while aging is associated with a reduction in gait speed, it has been previously detailed that it also produces a broad range of physiological and biomechanical changes on the walking apparatus, from the loss of muscular strength and mass, to a reduction in joint range of motion (Pavol et al., 2002; Delmonico et al., 2009; Billot et al., 2010; Cattagni et al., 2014). In our study, we corroborate and extend these results by showing that these changes occur specifically at the level of the knee/shank and ankle/foot during walking. Moreover, stiffness at both ankles and knees seems have no influence on joint motion in our results and are in line to those reported by others with no evolution of joint stiffness with age (Ochala et al., 2004; Collins et al., 2018). In the same vein, variability of step length, and step width previously reported as gait instability surrogates did not reach significance in our model. One possible explanation is that the confortable walking velocity might have optimized balance during gait. Further study implementing more complex balance constraints need to explore the contribution of these parameters in the aging process.

Outcomes extracted from lower-limb coordination, ankle motion, and plantar-flexors muscles seem to be a key target for both scientist and therapist. More precisely, ankle power, ankle sagittal kinematic, and shank-foot coordination seem to be reduced with aging whereas knee power seems to increase. Such modifications may reveal a potential adaptive mechanism occurring throughout the lifespan. Consequently, we believe that the planar covariation method provides basic insights into how the central nervous system controls limbs during walking by taking into account the global coordination of the thigh, shank, and foot segments. One can expect that lowerlimb coordination has been modified with aging in order to compensate for the weakness progressively shown with aging and especially after the 6th decade of life. Such modifications of lower-limb coordination have been previously reported when the locomotor apparatus is impaired (Laroche et al., 2007; Ornetti et al., 2011; Leurs et al., 2012). However, a previous study (Bleyenheuft and Detrembleur, 2012) failed to observe lowerlimb coordination difference with aging. It could be explained by the weak statistical power and the absence of the shankfoot coordination, that seems to be modulated with aging. Thus, the planar covariation method seems to highlight the adaptation of the decline in the neuromuscular system with aging (Lacquaniti et al., 2012b). It could be argued therefore, that lowerlimb coordination may act as a compensatory mechanism for physiological, and biomechanical changes in order to optimize the locomotor control and the dynamical balance (Ivanenko et al., 2006). Recently, Song and Geyer (2018) proposed a computer simulation to investigate the physiological causes of altered gait with aging, They found potential evidence that muscle-activation changes dominantly contribute to the reduced walking speed. In others words, the alteration of ankle power with aging could be one of the primary symptoms of the physiological decline due to aging. Further work should investigate muscular activation along lifespan in order to corroborate this hypothesis. A particular attention has to be done on prevention programs specifically designed to enhance the strength and coordination of lower-limb muscles and determine its potential effect of ankle power, lower-limb synergies, and gait speed.

This study does however, have several limitations. First, the power of the multiple regression was limited by the number of volunteers. However, the advantage inherent in this limitation is that only very strong relationships could be demonstrated. Despite the linear relationship between age and walking parameters, this study did not provide longitudinal data of volunteers. However, in the majority of studies, only groups are compared. We provide in this study data from young adults to aging people that may highlights changes during the whole lifetime. Second, the absence of the maximal strength of the volunteers to quantify the functional capacity and possibly the related gait performance should be noted.

In conclusion, this study showed age-related effects on gait performance. In particular, the modification of shank-foot

# REFERENCES


coordination could constitutes a neuromuscular adaptation of the changes (biomechanical, physiological, etc.) occurred with aging. Furthermore, our results might have implications for clinical research and practice. Indeed, these four specific parameters could be relevant outcomes to measure efficacy of rehabilitative strategies and to evaluate their efficiency for restoring lower-limb synergies during walking. Consequently, it may be interesting to focus gait rehabilitation on the improvement of ankle amplitude and power as well as foot-shank coordination with healthy and pathological elderly people.

# DATA AVAILABILITY

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

# ETHICS STATEMENT

The study protocol was approved by the local ethics committee (CPP Est I, Dijon, France). The study was conducted in compliance with the principles of Good Clinical Practice and the Declaration of Helsinki and all patients gave their informed consent.

# AUTHOR CONTRIBUTIONS

DL, PO, CL, and CM designed the experiment. DL and PO performed the experiments. MG, DL, AG, J-MC, PS, PO, and CL analyzed the data. MG, DL, and PS drafted the manuscript. MG, DL, PS, AG, J-MC, and PO critical revision of the article for important intellectual content. All authors give final approval of the version to be submitted.

# ACKNOWLEDGMENTS

The studies included in this paper were supported by the Dijon-Bourgogne University Hospital, Authors are grateful to Hospital research staff and to all participants.


**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 Gueugnon, Stapley, Gouteron, Lecland, Morisset, Casillas, Ornetti and Laroche. 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.

# Identify the Alteration of Balance Control and Risk of Falling in Stroke Survivors During Obstacle Crossing Based on Kinematic Analysis

Na Chen1†, Xiang Xiao1,2†, Huijing Hu3†, Ying Chen<sup>4</sup> , Rong Song<sup>4</sup> and Le Li <sup>1</sup> \*

*<sup>1</sup> Department of Rehabilitation Medicine, First Affiliated Hospital, Sun Yat-sen University, Guangzhou, China, <sup>2</sup> Department of Rehabilitation Medicine, Luo Hu Peoples' Hospital, Shenzhen, China, <sup>3</sup> Guangdong Work Injury Rehabilitation Center, Guangzhou, China, <sup>4</sup> Key Laboratory of Sensing Technology and Biomedical Instrument of Guang Dong Province School of Engineering, Sun Yat-sen University, Guangzhou, China*

#### Edited by:

*Helena Blumen, Albert Einstein College of Medicine, United States*

#### Reviewed by:

*Carmela Conte, Fondazione Don Carlo Gnocchi Onlus (IRCCS), Italy Jasmine Menant, Neuroscience Research Australia (NeuRA), Australia*

\*Correspondence:

*Le Li lile5@mail.sysu.edu.cn*

*†These authors have contributed equally to this work*

#### Specialty section:

*This article was submitted to Neurorehabilitation, a section of the journal Frontiers in Neurology*

Received: *29 December 2018* Accepted: *15 July 2019* Published: *30 July 2019*

#### Citation:

*Chen N, Xiao X, Hu H, Chen Y, Song R and Li L (2019) Identify the Alteration of Balance Control and Risk of Falling in Stroke Survivors During Obstacle Crossing Based on Kinematic Analysis. Front. Neurol. 10:813. doi: 10.3389/fneur.2019.00813* This study aims to compare the differences in the kinematic characteristics of crossing obstacles of different heights between stroke survivors and age-matched healthy controls and to identify the changes of balance control strategy and risk of falling. Twelve stroke survivors and twelve aged-matched healthy controls were recruited. A three-dimensional motion analysis system and two force plates were used to measure the kinematic and kinetic data during crossing obstacles with heights of 10, 20, and 30% leg length. The results showed that during leading and trailing limb clearance, (AP) center of mass (COM) velocities of the stroke group were smaller than those of the healthy controls for all heights. The decreased distances between COM and center of pressure (COP) in the AP direction during the both trailing and leading limb support period were also found between stroke survivors and healthy controls for all heights. The COM velocity and COM-COP distance significantly correlated with the lower limb muscle strength. In addition, stroke survivors showed greater lateral pelvic tilt, greater hip abduction, and larger peak velocity in the medio-lateral (ML) direction. There was a positive correlation between the COM-COP distance in the AP direction and the clinical scales. These results might identify that the stroke survivors used a conservative strategy to negotiate the obstacles and control balance due to a lack of muscle strength. However, the abnormal patterns during obstacle crossing might increase the risk of falling. The findings could be used to design specific rehabilitation training programs to enhance body stability, reduce energy cost, and improve motion efficiency.

Keywords: stroke, gait, balance control, obstacle crossing, kinematics

# INTRODUCTION

The impairments from stroke impact patients' activities in daily life. Most community-dwelling stroke survivors can walk safely on level surfaces, but they have difficulties in maintaining balance during complex motor tasks such as obstacle crossing (1). Compared with healthy controls, stroke survivors are more likely to fall during obstacle avoidance, either by contacting the obstacle or losing balance (2). The consequences of falling include hip fractures, soft tissue injuries, fear of

**240**

falling, hospitalization, increased immobility, and greater disability (3, 4). Moreover, Said and colleagues found stroke survivors who failed in the obstacle crossing task demonstrated higher falling risk compared with who passed the task (5). Therefore, identifying falling risk during obstacle crossing and preventing falls are important for stroke survivors (6).

Successful obstacle crossing requires sufficient toe-obstacle clearance provided by the swing limb and body stability provided by the stance limb. This calls for complicated and coordinated controls of both limbs during crossing (7). Researches have been conducted to investigate the motor control strategies among young and old healthy adults and stroke survivors, and find the different strategies could be caused by age-related physical degradation and the stroke-induced muscle weakness (6–9). However, the motor control changes for stroke survivors to make safely step across the obstacle with insufficient muscle strength are still not clear. And quantitative evidence of the balance control changes during such complex task of obstacle crossing for stroke survivors need further investigation. In the previous studies, the most-used strategy that stroke survivors took named circumduction was found during level walking (9), and also the reduced muscle response in stroke survivors compared with healthy controls during obstacle (10). In addition, Lu and colleagues investigated motor performance in highly functional post-stroke patients during obstacle crossing, and found that stroke survivors appeared to adopt a specific symmetric kinematic strategy with an increased pelvic posterior tilt and swing hip abduction (11). Studies included motor control strategies and balance assessment among stroke survivors during such complex task of obstacle crossing are needed.

Balance is often quantified using laboratory-measured variables such as the velocity of the center of mass (COM) and the distance between COM and the center of pressure (COP) (12–15). Also, the Berg Balance Scale (BBS) and the Fugl-Meyer Assessment (FMA) are two clinical measures of balance and motor impairment that are widely used in the field of stroke rehabilitation (16, 17). However, the clinical scales could provide simple assessments and have doubtful ability to demonstrate equivalent quality to laboratory-measured characteristics. Corriveau and coworkers found significant negative linear correlation between clinical scales (BBS and FMA) and COM-COP distance during quiet stance in stroke survivors. They found that postural stability measured by the COM-COP distance was related to the functional measures of balance (18). However, there is little information about the correlation between clinical scales and variables reflecting dynamic balance such as the COM-COP distance during obstacle crossing.

This study aimed to identify the falling risk of stroke survivors during obstacle crossing and to investigate the motor control strategies combined with balance modulation to better understand the way that stroke survivors negotiate obstacles of different heights. We hypothesized that stroke survivors might have different balance control and gait pattern during obstacle crossing which may lead to high risk of falling. The correlation of kinematic data with muscle strength and clinical scales could further provide information and evidence to quantify the balance performance during obstacle crossing for stroke survivors. The TABLE 1 | Basic characteristics of study subjects.


*a Indicates significant effect using an independent t-test. FMA, Fugl-meyer assessment.*

findings of current study may help to design training protocol and evaluate rehabilitation interventions for motor recovery in stroke survivors and facilitate the improvement of conducting daily task such as obstacle crossing.

# METHODS

# Participants

The subjects included 12 stroke survivors and 12 gender-, age-, and height-matched healthy subjects. The basic characteristics of subjects was displayed in **Table 1**. The inclusion criteria for the stroke patients were (1) stroke with unilateral hemiparesis lesions confirmed by magnetic resonance imaging or computed tomography; (2) at least 3 months having passed since the stroke; (3) capability of walking 10 meters without a gait aid or assistance and across an obstacle with a height of 30% leg length. The exclusion criteria were other neurologic diseases, such as Parkinson's disease, diabetic polyneuropathy, Alzheimer's disease, and other cognitive impairments. This study was approved by the Ethics Committee of the local hospital and was conducted in accordance to the Declaration of Helsinki. The consent obtained from the participants was both informed and written before the experiments.

# Apparatus

Thirty-five 15-mm infrared-reflective markers were taped to the skin overlying body landmarks according to the Vicon Plug-In Gait marker placement method. A 6-camera 3D motion analysis system (Vicon Motion Systems, Oxford, UK) recorded the marker positions at a sample frequency of 100 Hz. Two force plates (464 mm × 508 mm × 83 mm; AMTI, Watertown, MA, USA) with a sample frequency of 1 kHz were placed in the middle of the path with obstacles between them. The data from the three-dimensional motion system and the force plates were synchronized. The obstacle has adjustable height and consists of two upright stands with a light-weight crossbar, which was set to three height conditions (10, 20, and 30% of the leg length). A handheld muscle-testing dynamometer (microFET3, Hogan Health, USA with the precision of 0.4 N and range from 13 to 1,330 N) was used to measure the peak isometric force (19).

# Procedure

## Anthropometric Measure

Before the gait analysis, the basic characteristics were first measured. Leg length was measured with a tape measure from the anterior superior iliac spine to the lateral malleolus and was used to calculate the obstacle height for each individual.

# Muscle Strength and Clinical Scales

The peak isometric forces of the knee extensors and flexors and the ankle dorisflexors and plantarflexors were also measured. Muscle strength was measured as the peak isometric force in 4 muscle groups including rectus femoris (RF), biceps femoris (BF), tibialis anterior (TA), and medial gastrocnemius (MG) of the affected lower limb of the stroke survivors and the dominant lower limb of the healthy controls (20). Details of the procedures of testing muscle strength could be found in our previous paper (21). An experienced physiotherapist who was blinded to the gait results evaluated the lower extremity FMA and BBS to assess the lower limb motor function and balance function of the stroke group.

# Kinematic Data

Subjects were then instructed to walk at a self-selected speed with bare feet along an 8-meter walkway, where an obstacle was placed midway, perpendicular to the walking direction, and parallel to the ground. Each trial began at a similar starting point of the walkway with a marker on the floor and the subjects were suggested to use it as a reference of similar walking distance. The stroke survivors were instructed to use their affected leg as the leading limb of the obstacle crossing, and the healthy controls were not restricted of which leg be used. First, the level unobstructed walking was performed, then the different heights were crossed in random order, and three successful trials were recorded for each height condition. We ignored the trials in which the subjects touched the obstacle. Trials were excluded from motion analysis if the participant required the therapist's assistance to maintain balance or tripped over the obstacle. Subjects were reminded to perform the task within their limits of safety and stop if they felt at risk. A therapist accompanied the subjects and walked alongside them to provide assistance if required.

# Data Processing

Vicon Nexus (Version 1.7.1) was used for data processing. The kinematic data were obtained during the crossing stride, which was defined as the period beginning with the trailing limb's heel-contact just before crossing the obstacle to the next heelcontact just after crossing the obstacle (22). The crossing stride can be further divided into five sub-phases: the pre-obstacle double support phase, single support phase of the trailing limb support period (TLP), obstacle-crossing double support phase, single support phase of the leading limb support period (LLP), and post-obstacle double support phase.

The angles of pelvis and lower limb joints of both the stance and swing limbs were calculated when the toe marker was above the obstacle (8). To calculate the end point data (the distance between the toe marker and the obstacle), we defined three distances between the lower limb and the obstacle (**Figure 1**): TOD is the horizontal distance between the trailing toe and the obstacle before crossing the obstacle; HOD is the horizontal distance between the leading heel and the obstacle after crossing the obstacle; TOC is the vertical distance between the toe of leading toe and obstacle when the toe is over the obstacle. COM and COP locations were calculated using Vicon Workstation software. In details, COM is based on Plug-in-gait model of VICON system with the weighted average value of head, sternum, humerus, radius, hand, pelvis, femur, tibia and foot. COP is based on data from force plates. Instantaneous anterior–posterior (AP) COM velocity was examined at two critical phases: leading limb clearance (LC) and trailing limb clearance (TC) (**Figure 1**). Medio-lateral (ML) COM velocity was quantified by examining the peak ML velocity during the double support periods: the preobstacle (peak 1), during-obstacle (peak 2), and post-obstacle crossing phases (peak 3). The distance between COM and COP was calculated as the root mean square (RMS) during TLP and LLP (18). When examining the correlations between the balance variables and the measured muscle strength, variables representative of obstacle crossing were first averaged within each of the three heights.

# Statistical Analysis

All statistical analyses were performed using IBM SPSS Statistics version 20.0. All the calculated variables for both groups were firstly subjected to a Kolmogorov–Smirnov test. The pelvis and joint angles did not show a normal distribution, and they were represented by the median value with interquartile range (IQR) (1, 5). We then applied the Mann-Whitney U test to these variables and Kruskal-Wallis one-way ANOVA to find the difference of heights within each group. Other variables showing a normal distribution were tested using two-way ANOVA with height as the within-group factor and group as the betweengroup factors. A post hoc test with Bonferroni correction was used to examine group differences among different heights. Pearson product-moment correlations were used to examine the relationship between the balance variables and the measured muscle strength, also between the balance variables and clinical scales. The significance level was set at 0.05.

# RESULTS

Typical trials of the kinematic behavior of the lower limbs and pelvis of the trailing and the leading limb in a patient after stroke and a healthy subject during the crossing stride sub-phases has been shown in **Figure 2**.

# Kinematic Data of COM-COP

During the leading and trailing limb clearance, the anteriorposterior center of mass (AP COM) velocity of the stroke group

was smaller than that of the healthy controls for all obstacle heights (p < 0.05, **Figures 3A,B**), and it decreased with the increasing obstacle height at leading limb clearance (p < 0.05, **Figure 3A**). The AP COM-COP distance of the stroke group during trail limb support period (TLP) showed a significant decrease compared with the healthy controls for 20% height (p < 0.05, **Figure 3C**). During leading limb support period (LLP), the AP COM-COP distance of the stroke group was significantly smaller than that of the healthy controls for all obstacle heights (p < 0.05, **Figure 3D**).

The peak medio-lateral (ML) COM velocity of the stroke group during the double support phases in both the pre-obstacle (peak 1) and post-obstacle (peak 3) phases were significantly greater than for the healthy controls at all heights (p < 0.05, **Figures 4A,C**). However, there were no significant differences between groups in the peak ML COM velocity during the double support phase (peak 2, p > 0.05, **Figure 4B**). During the TLP, the increasing obstacle heights resulted in decreases in the ML COM-COP distance in both groups (p < 0.05, **Figure 4D**). However, there were no significant differences between groups in the COM-COP distance in the ML direction during TLP and LLP (**Figures 4D,E**).

# Joint Angle and End Point Data During the Crossing

The crossing pelvis and joint angles data are shown in **Table 2**. Greater lateral pelvic tilt was found in the stroke group (p < 0.05) for all heights compared with the healthy controls. No significant difference was found in the anterior or posterior pelvic tilt or the pelvic rotation. In the leading swing limb, greater hip abduction (p < 0.05) and smaller knee extension (p < 0.05) were found in stroke survivors compared with healthy controls for all heights. In the trailing stance limb, greater hip abduction and knee extension were also observed in the stroke group compared with healthy controls for all heights (p < 0.05).

**Table 3** presents the end point data. Comparisons between the stroke and healthy elderly groups revealed that the stroke group had less HOD (10%: p = 0.024; 20%: p = 0.01; 30%: p = 0.045). No significant differences were found in the leading TOC and the trailing TOD (p > 0.05).

# Correlations Between Kinematic Data, Muscle Strength and Clinical Scales

**Table 4** shows the correlation between COM velocity, COM-COP distance and lower limb muscle strength for stroke survivors. The ankle dorsiflexor strength correlated significantly positively with COM velocity in the AP direction when during LC (r = 0.623, p < 0.05) and TC (r = 0.690, p < 0.05). The ankle plantarflexors strength correlated significantly positively with COM velocity in the AP direction when during LC (r = 0.623, p < 0.05). While in the ML direction, there were significant negative correlations between knee extensors strength (r = −0.608, p < 0.05) and COM velocity during peak 2, between ankle dorsiflexor (r = −0.787, p < 0.01) and plantarflexors strength (r = −0.578,

p < 0.05) and COM velocity during peak 3. As to COM-COP distance, the ankle dorsiflexor strength correlated significantly positively with it in the AP direction during TLP (r = 0.631, p < 0.05) and LLP (r = 0.694, p < 0.05). While the ankle plantarflexor strength correlated significantly positively with it in the AP direction during LLP (r = 0.674, p < 0.05).

**Figure 5** shows the significant correlations between the clinical scales and AP COM-COP distance for different periods when crossing the 10% leg length obstacle. The scores of BBS and lower extremity FMA demonstrated moderate positive correlations with AP COM-COP distance (p < 0.05). We did not observe any significant correlation between clinical scales and AP COM-COMP at other heights (i.e., 20 and 30%).

# DISCUSSION

We compared kinematic data from stroke survivors and agematched healthy control subjects when negotiating obstacles of different heights in order to understand the mechanisms of motor control changes during obstacle-crossing after stroke. Our results demonstrated that the stroke survivors might change balance control by using a conservative strategy during the obstacle crossing to ensure safe crossing with a lack of muscle strength and to prevent falls.

# COM-COP Distance and COM Velocity

The results showed that the AP COM velocity of the stroke group was significantly slower than that of healthy controls in both the leading limb clearance and trailing limb clearance (p < 0.05, **Figures 3A,B**). With decreased AP COM velocity, the stroke survivors could better control the anterior movement of the COM, which potentially increased the stability and reduced the risk of losing balance. This finding was similar to those of a previous study (15). In the AP direction, the COM-COP distance was significantly smaller in the stroke survivors than healthy controls only at 20% leg length obstacle during TLP, while it was significantly smaller at all heights during LLP (p < 0.05, **Figures 3C,D**). Maintaining the COM closer to the COP could result in smaller moment arms for the body weight of the stance limb and require less muscular effort to maintain balance (12). Said and coworkers demonstrated no differences between groups in AP COM-COP distance during TLP, but they did not investigate the distance after the clearance (15). Interestingly, in current study when stroke survivors were supported by the affected limb after clearance during LLP, it showed that they had smaller AP COM-COP distances. Our results in part supported Said's findings and further demonstrated poor balance ability for stroke survivors after they crossed the obstacle. The reductions in the COM velocity and COM-COP distance were considered as a conservative strategy used by the stroke survivors to deal with the mechanical challenge during obstacle crossing and to increase stability.

Quick weight shifting to the trailing limb and the lateral pelvic tilt to raise the toe resulted in high instantaneous velocity in the ML direction. The higher ML velocities indicated difficulty in maintaining dynamic stability in the frontal plane and could also

difference between heights. The error bar represents 1 SD.

3. (D) COM-COP distance during TLP. (E) COM-COP distance during LLP. \*Reflects the significant difference between groups, #reflects the significant difference between heights. The error bar represents 1 SD.

TABLE 2 | Median crossing pelvis and joint angles of the limbs when the leading toe was above the obstacle (Mean (SD), *n* = 12).


*a Indicates significant effect between groups.*

*b Indicates significant effect between heights.*

*Pelvic Rotation (anterior* +*, posterior* −*); Abduction (*+*)/Adduction (*−*); Flexion (*+*)/Extension (*−*).*

TABLE 3 | Mean distance (standard deviation) between the limb and the obstacle.


*a Indicates significant effect between groups.*

*TOD, the horizontal distance between the trailing toe and the obstacle before crossing the obstacle; HOD, the horizontal distance between the leading heel and the obstacle after crossing the obstacle; TOC, the vertical distance between the toe of leading toe and obstacle when the toe is over the obstacle.*

reflect difficulty in decelerating COM, which is governed by the relative loading and unloading of the two limbs during double support (23). As a result, the ML COM-COP distance following stroke was not decreased as in the AP direction, and there was no significant difference from the healthy controls (**Figures 3D,E**). This indicates poor balance maintenance in the ML direction and an increased possibility of falling to the side.

# Joint Angle and the Distance Between Lower Limb and Obstacle

To further examine the control strategy of the stroke survivors, we looked for more details about the kinematics using the pelvic and joint angles. Compared with the healthy controls, stroke survivors showed significantly larger lateral pelvic tilt angles in the ML direction and larger hip abduction angles with both trailing and leading limbs (**Table 2**). This implies that stroke survivors firstly shifted their weight to the unaffected side, raised the pelvis of the affected side, and abducted the hip to elevate the swing toe. According to neurological development theory, proximal control using the pelvis is more efficient than distal control using the hip or knee (24). Using proximal control, the stroke survivors elevated the toe to maintain a safe clearance between the toe and the obstacle and compensated for the decrease of the knee extension. Lu et al. found that stroke survivors with high motor function adopted an increased posterior pelvic tilt strategy during obstacle crossing (10). In our study, the decrease in the AP COM velocity provided enough time for the lateral pelvic tilt strategy where the stroke survivors elevated the toe and cleared the obstacle in a circumduction, and the decreased AP COM-COP distance improved the stability after the clearance. However, as a result of the lateral pelvic tilt strategy, the stroke survivors showed greater peak ML velocity toward the leading and trailing limbs than the healthy controls during peak 1 and peak 3 (**Figures 2A,C**), which indicated instability in the ML direction during both push-off and landing phase. Moreover, the HOD was significantly smaller in stroke survivors compared with healthy controls (**Table 3**), and which might place stroke survivors at risk of actual contact or trip of the obstacle (25). These findings implied a high falling risk for the stroke survivors during obstacle crossing.

# Clinical Correlations

We examined the correlations between the muscle strength of the lower limb muscles and the balance variables to further investigate the cause of the poor balance ability among stroke survivors. The significant correlations (**Table 4**) might provide evidence that deficit in muscle strength could be a cause of the altered strategies among stroke survivors during obstacle crossing. The COM velocity was reduced which may due to the deficit in muscle strength, and poor balance ability resulted in larger COM velocity in the ML direction and increased the falling risk. Stroke survivors with weaker muscle strength had to place the COM closer to the COP to maintain stability. The different locomotor performance caused by muscle strength has been demonstrated (24), which is similar to our findings and supports that the deficit in muscle strength could be related to the changing of balance control. In addition, the abnormal energy cost among stroke survivors might be another reason during obstacle crossing (26).

One interesting phenomenon in our results was that there were significant positive correlations between AP COM-COP distance and clinical scales (both BBS and lower extremity FMA) for the 10% leg length obstacle height (**Figure 5**). This provided more information about the balance mechanism adopted after stroke. Our findings showed that stroke survivors with higher clinical scores allowed for a relatively greater AP COM-COP distance to ensure safe crossing compared with those with lower clinical scores when crossing relatively low obstacles. Similarly, a significant negative correlation was reported between the BBS and the COM-COP distance following stroke during quiet stance (18). However, there was no significant correlation between the AP COM-COP distance and the clinical scales when crossing higher obstacles crossing (20% and 30% leg height) which might attribute to challenging task caused high variations among patients. Performance was more disturbed for the higher obstacle during obstacle crossing (27). Stroke survivors could not modulate themselves well when facing higher heights and adopted more abnormal patterns to step across the obstacle. Therefore, the measure of AP COM-COP distance could provide a reliable method for assessing dynamic balance for stroke survivors when crossing relatively low obstacles and to clinically assess balance. Similar tasks were performed in patients with traumatic brain injury, and also through COM-COP distance to demonstrate the patients had difficulty maintaining dynamic stability during obstacle crossing (28). Therefore, the obstacle crossing task challenges stroke survivor's ability to maintain balance, and makes them adopt a quite conservative strategy to safely


TABLE 4 | Correlations between COM velocity, COM-COP distance and lower limb muscle strength.

\**indicates p* < *0.05.*

\*\**indicates p* < *0.01.*

*AP, anterior–posterior; ML, medio-lateral; LC, leading limb clearance; TC, trailing limb clearance; TLP, trailing limb support period; LLP, leading limb support period.*

step across the obstacle. This study can help to understand the control strategy applied by stroke survivors during this complex task and provide understanding to design proper rehabilitation intervention and training theme to decrease the falling risk.

# Limitations

This study has several limitations which need cautions for the interpretation and generalizability of the data. The healthy controls were asked to cross the obstacle with self-selected speed. Previous studies demonstrated that gait changes following stroke were related to speed, and few differences between groups were found when healthy controls used matched speed (25). Moreover, the reductions in the speed of healthy controls might also potentially increase the risk of obstacle contact. Taking these into account, we finally investigated the difference between the two groups at their self-selected speed. In this study, the recruited stroke survivors have moderate to good motor functions and might demonstrated less significant difference compared with healthy controls. We did not record the successful rate of each subject on obstacle crossing but it may be an interesting topic warrant further investigation in the future. In addition, for the safety reason, we did not require the stroke survivor to use the affected side as supporting leg during cross and most of them selected unaffected side to support and all the data been analyzed was from this pattern only. But the healthy subjects were not restricted and they could cross the obstacle using left or right leg and we just found it interesting that most of the heathy subjects using right leg to cross and only those data were analyzed and compared. These might be the reasons that there is no group∗height interaction effect. In future work, a larger sample size of different kind of stroke survivors as well as an investigation of the mechanism of the neuromuscular activation following stroke [e.g., EMG data (21)] could be added to better examine the biomechanical mechanisms during obstacle crossing.

# CONCLUSION

The current study investigated the motion patterns in stroke survivors during obstacle crossing compared with healthy controls. The balance of crossing is compromised following stroke and stroke survivors might use a conservative strategy to negotiate the obstacles to prevent tripping which might due to a lack of muscle strength. In addition, there were some abnormal patterns during the crossing, which might increase the risk of falls and instability of balance. The positive correlation between the COM-COP distance and the clinical scales indicates potential as a suitable method for assessing the ability to maintain

# REFERENCES


balance during obstacle crossing. Muscle strength training is recommended for rehabilitation to regain balance ability and to correct the abnormal gait patterns for stroke survivors.

# ETHICS STATEMENT

This study was approved by the Ethics Committee of the First Affiliated Hospital, Sun Yat-sen University, and all subjects provided informed consent before the experiments.

# AUTHOR CONTRIBUTIONS

NC, XX, HH, and LL conceived and designed the study. NC, XX, HH, and YC performed the experiments. NC, XX, HH, and YC wrote the paper. NC helped to response the comments from the reviewers and revise the manuscript. RS and LL made a contribution to experiments. RS and LL reviewed and edited the manuscript. All authors had read and approved the manuscript.

# FUNDING

This work was supported by the National Natural Science Foundation of China (Nos. 31771016, 81702227), and the Guangdong Science and Technology Department (Nos. 2017B020210011, 2017B010110015). This project was also partly supported by Royal Society International Exchanges grant (No. 170168), UK.

# ACKNOWLEDGMENTS

The authors would like to thank all the participants of this study.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fneur. 2019.00813/full#supplementary-material


functional walkers poststroke. Neurorehabil Neural Repair. (2013) 27:230–9. doi: 10.1177/1545968312462070


**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 Chen, Xiao, Hu, Chen, Song and Li. 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.

# Playing Exergames Facilitates Central Drive to the Ankle Dorsiflexors During Gait in Older Adults; a Quasi-Experimental Investigation

#### Eling D. de Bruin1,2 \* † , Nadine Patt<sup>1</sup> , Lisa Ringli<sup>3</sup> and Federico Gennaro<sup>1</sup>†

1 Institute of Human Movement Sciences and Sport, Department of Health Sciences and Technology, ETH Zurich, Zurich, Switzerland, <sup>2</sup> Division of Physiotherapy, Department of Neurobiology, Care Sciences and Society, Karolinska Institutet, Stockholm, Sweden, <sup>3</sup> SRH Hochschule für Gesundheit, Gera, Germany

Purpose: Gait training might be of particular importance to reduce fall risk in older adults. In the present study we explore the hypothesis that video game-based training will increase tibialis anterior (TA) muscle EMG-EMG coherence and relates to functional measures of lower limb control.

#### Edited by:

Helena Blumen, Albert Einstein College of Medicine, United States

#### Reviewed by:

Luwen Wang, Case Western Reserve University, United States Bettina Wollesen, Universität Hamburg, Germany

> \*Correspondence: Eling D. de Bruin eling.debruin@hest.ethz.ch

#### †ORCID:

Eling D. de Bruin orcid.org/0000-0002-6542-7385 Federico Gennaro orcid.org/0000-0003-2203-2858

Received: 29 April 2019 Accepted: 05 September 2019 Published: 20 September 2019

#### Citation:

de Bruin ED, Patt N, Ringli L and Gennaro F (2019) Playing Exergames Facilitates Central Drive to the Ankle Dorsiflexors During Gait in Older Adults; a Quasi-Experimental Investigation. Front. Aging Neurosci. 11:263. doi: 10.3389/fnagi.2019.00263 Methods: We focus on video game-based training performed in standing position, where the subjects have to lift their toes to place their feet on different target zones in order to successfully play the game. This type of training is hypothesized leading to progressive changes in the central motor drive to TA motor neurons and, consequently, improved control of ankle dorsiflexion during gait.

Results: Twenty older adults, 79 ± 8 years old, 13 females/7 males, participated. Results showed a significant difference against 0 in the experimental 1POST condition in dual-task walking and beta Frequency Of Interest (p = 0.002). Walking under dual task condition showed significant change over time in minimal Toe Clearance for both the left [χ 2 (2) = 7.46, p = 0.024, n = 20] and right [χ 2 (2) = 8.87, p = 0.012, n = 20] leg. No change in lower extremity function was detectable.

Conclusion: Overall we conclude that the initiation of an exergame-based training in upright standing position improves neural drive to the lower extremities in older adults, effects on minimal Toe Clearance and seems an acceptable form of physical exercise for this group.

Keywords: older adults, exergame, central drive, tibialis anterior, foot clearance

# INTRODUCTION

Increasing the number of years of good health while maintaining independence and quality of life as long as possible is a primary public health goal. Avoidance of disease and disability, maintaining high physical and cognitive function, and sustained engagement in social and productive activities are components of healthy aging that together define successful aging (Rowe and Kahn, 1997).

A large component of successful aging interventions aims to maximize physical performance levels. Being able to fully participate in daily life activities may be affected when the capability to easily perform common physical functions decreases (Rowe and Kahn, 1997). Consequently, in older adults health status can be regarded an important indicator of quality of life (Johnson and Wolinsky, 1993; Spirduso and Cronin, 2001). The way how middle-aged and older adults perceive their health seems especially related to components of health-related fitness and functional performance, or to chronic conditions and diseases that influence these fitness components (Johnson and Wolinsky, 1993; Malmberg et al., 2002; Malmberg et al., 2005).

Older adults that are physically active or who regularly exercise help in preventing the development and progression of chronic degenerative diseases (Physical Activity Fundamental To Preventing Disease, 2002; Chodzko-Zajko et al., 2009; Stuart et al., 2009). Being physically active or adherent to regular exercise are means to consistently improve age-related muscle weakness, physical function, cognitive performance, and mood in older adults (Landi et al., 2010). Where a sedentary lifestyle in older adults will increase the risk of unintentional falls, it is observable that falls risk is reduced by being physically active (Thibaud et al., 2012). Approximately 30% of older people experience falls on a yearly basis (Berg et al., 1997; Hausdorff et al., 2001; Gill et al., 2005) and falling can be seen as a common problem in the growing elderly population (de Bruin et al., 2012).

Individuals with gait impairments have an increased risk for repeated falls (Tinetti et al., 1995) and, accordingly, walking ability and the risk of falling are linked with each other. The three most frequent motor control related direct causes of falling are tripping, slipping and loss of balance (Lord et al., 1993; Sherrington et al., 2004). Furthermore, forward walking activity is associated with a high proportion of falls (Robinovitch et al., 2013).

Greater variability in Minimum Foot Clearance (MFC) is a contributing factor to a trips risk increase, and concomitant associated heightened falls occurrence, in those cases where older are compared with younger adults and older fallers to older non-fallers. MFC is "the minimum vertical distance between the lowest point of the foot of the swing leg and the walking surface during the swing phase of the gait cycle (Barrett et al., 2010)." Toe clearance control with the central nervous system in critical situations is impaired in older adults at high risk of tripping (Hamacher and Schega, 2014), an observation in line with research indicating that corticospinal transmission to skeletal muscle may be impaired with advancing age (Manini et al., 2013).

Motor-cognitive training with exergames may help in effectively supporting physical, psychological, and cognitive rehabilitative outcomes in older adult populations (Zeng et al., 2017; van Santen et al., 2018). Patients with mobility problems, for example, have shown transfer of training effects obtained in a virtual environment to real-life (Erren-Wolters et al., 2007). Two systematic reviews have shown that especially cognitivemotor stepping interventions with video-games positively effect on gait of older adults (Schoene et al., 2014; Okubo et al., 2017) and this approach, furthermore, is considered task-specific for neuroplasticity improvements (Netz, 2019). Previous studies from our group pointed to the potential of video game-based training and highlighted that specifically designed game-based training improves walking (Pichierri et al., 2012a,b; Fraser et al., 2014) and effects on the brain (Eggenberger et al., 2016; Schattin et al., 2016; Stanmore et al., 2017). The game play also requires ankle dorsiflexion to play the game, a MFC-associated factor (Sato, 2015). Here we explore the hypothesis that video game-based training effects on corticospinal transmission to the tibialis anterior (TA) muscle assessed by means of EMG-EMG coherence and lower limb control. We focus on video game-based training performed in standing position, where the subjects have to lift their toes to place their feet on different target zones in order to successfully play the game. We hypothesize that this type of training leads to changes in the central drive to TA and, thus, to improved ankle dorsiflexion control during gait.

# MATERIALS AND METHODS

In this "pretest – posttest" quasi experimental single group design the older adults acted as their own controls to control for inter-subject variability (Willerslev-Olsen et al., 2015). Comparing against a not training group of older adults would complicate the comparison of data given the expected diversity of functional abilities in older adults. Furthermore, this design allowed determining whether intervention effects can be explained by simple test-retest variability and whether significant changes would be observed in the measured variables during a 6-week control period comparable to that of the 6-week intervention period.

The study was designed for autonomous and independent living older adults. For study inclusion, the older adults had to be 65 years or above, had to be in good physical health by self-report (assessed by means of a Health Questionnaire), had to be able to walk at least 500 m independently (with or without walking aids), and they had to be not experienced in exercising with virtual reality-based games before this study. Moreover, they were considered eligible if they had not been diagnosed for cognitive impairments and they had a Montreal Cognitive Assessment (MoCA) score of 26 or more points. Participants were not eligible when exhibiting acute or unstable chronic diseases, rapidly progressing or terminal illnesses or were suffering from severe health problems (e.g., recent cardiac infarction, uncontrolled diabetes or hypertension). The ethics committee of the ETH Zurich, Switzerland (EK 2017-N-22) approved the study protocol. Before any measurements were performed, an informed consent according to the Declaration of Helsinki was administered and signed by each eligible participant.

# Protocol and Training Intervention

Twenty older adults (average age 79 years old, 13 females, and 7 males) from the Alterszentrum Kehl (Baden, Switzerland) were recruited and participated voluntarily during a 12 weeks period. Gait analysis, EMG recordings of TA activity, physical functioning, and cognition were assessed during three test sessions separated by approximately 6 weeks intervals. All

test sessions took place in the Alterszentrum Kehl (Baden, Switzerland<sup>1</sup> ). The first test session occurred approximately 6 weeks before training commenced, and the second testing was organized 1 week before training initiation. The final test session took place following the last training day. All three testing sessions included the same measurements performed in the same order on every occasion and at the same time of the day. Test sessions began with concurrent gait analysis and EMG recordings during over ground walking, followed by lower extremity function and cognitive testing.

The participants performed a total of 18 training sessions lasting 20 min (with a break of 10 min in between). Each training session was performed on a Senso exergame system (dividat, Schindellegi, Switzerland; **Figure 1**). The Exergame had to be played using body movements to trigger sensors positioned on a base plate that was connected to a TV screen. Through these body movements participants apply forces and steps of which the dynamics are recorded. Real-time visual and auditory feedback on Exergame performance was provided

#### <sup>1</sup>https://www.daskehl.ch/

through electronic sensors in the dance pad detecting position and timing information.

The equipment was set up in a quiet room in the Alterszentrum Kehl (Baden, Switzerland). The participant stood in front of a computer screen positioned at eye-level during training. A handrail was placed in front of the participant to provide safety support, if needed. During the first session exercise tasks were explained by a study coordinator to the participant. During subsequent sessions, participants conducted training using screen-based feedback only. To guarantee safety the study coordinator remained with the participant during training.

The used exergames are designed to train different cognitive domains; e.g., divided attention, working memory, inhibition, attention shifting, spatial orientation and postural control (**Table 1**). The training intervention program was specifically designed for healthy seniors and the participants were gradually accustomed to the games. During the individual training period, each participant was constantly monitored by the instructors and received one-on-one supervision. The training protocol contained a variety of exercises in which the difficulty progressively adapted to an

#### TABLE 1 | Description of the video games.

fnagi-11-00263 September 19, 2019 Time: 14:58 # 4


individual skill level. Attendance to the training sessions was monitored by the instructors. Participants were provided with individual diaries at each training session, to record the date of training, which specific exergames were used, and both intensity and duration of each exergame performed in that session.

# EMG Data Acquisition and Pre-processing

Surface EMG signals were recorded at a sampling frequency of 1500 Hz (Noraxon DTS TeleMyo, Scottsdale, AZ, United States). For this two pairs of bipolar Ag-AgCl electrodes (Ambu Blue Sensor N, Ambu A/S, Ballerup, Denmark) over the left and right TA muscles (**Figure 2**) were used. Each bipolar configuration of the pair was placed either proximally or distally with respect to the muscle belly according to previously described anatomical landmarks (van Asseldonk et al., 2014). The inter-electrodes distance (electrodes' center-to-center) was set to 2 cm, whereas the two bipolar configurations, within each pair, were separated by ∼10 cm (range: ∼8.50 cm to ∼13.50 cm) in order to reduce the risk of cross-talk as well as the recording of muscle activity from overlapping motor unit areas (Hansen et al., 2005). To ensure the replication of the two bipolar placements across all the assessments, placements (for each TA muscle) were recorded by keeping trace of distance metrics from anatomical landmarks at each subject level. The skin was prepared (i.e., cleaned and, when necessary, shaved) before placing the EMG electrodes using a specific paste (H+H Medizinprodukte GbR, Münster, Germany). Moreover, in order to detect the onset of the heel strikes, two footswitches were placed approximately on the midpoint of the calcaneus in both feet. Continuous EMG data was first demeaned and de-trended by removing the zero- and firstorder polynomial, respectively, and then high-pass filtered at 10 Hz (zero-phase 4th-order Butterworth filter). Subsequently, powerline noise and its harmonics were filtered out using a notch filter based on Discrete Fourier Transformation (DFT) followed by a low-pass filter at 500 Hz (zero-phase 4th-order Butterworth filter) and rectification of the Hilbert transform of the filtered data. The latter non-linear transformation produces an output similar to performing solely rectification without previous transforming by Hilbert (Myers et al., 2003; Boonstra and Breakspear, 2012). This approach represents a widely used strategy in the preprocessing steps before computing either cortico-muscular coherence (i.e., by means of EEG and EMG) or intramuscular coherence (Schoffelen et al., 2011; Boonstra et al., 2015). Considering the debate on rectification as a proper preprocessing step before coherence analysis (Halliday and Farmer, 2010; Neto and Christou, 2010), the adopted filtering strategy with included rectification has shown to produce more reliable EMG-EMG coherence results during walking in comparison to different preprocessing strategies (van Asseldonk et al., 2014). The left and right heel-strike events were merged and used to epoch the EMG data relative to the gait cycle, similar as in previous analyses (Petersen et al., 2012; van Asseldonk et al., 2014; Willerslev-Olsen et al., 2015; Kitatani et al., 2016). The duration of the resulting epochs started one sample before the onset of the heel-strike (to avoid excessive contamination with artifacts of the EMG data caused by the collision of the foot with the ground)

FIGURE 2 | Placement over both tibialis anterior muscles of a pair of bipolar EMG electrodes using an inter-electrodes distance of ∼2 cm in each bipolar set and a distance of ∼10 cm between the two bipolar set of EMG electrodes to reduce the risk of cross-talk. Footswitches placed over the midpoint of each heel were placed in order to record heel-strike events during the overground gait trials.

to the preceding 400 ms. The preprocessed EMG data segments were subsequently down-sampled to 500 Hz.

# Spectral Analysis of EMG Data

A multi taper frequency transform was used to achieve cross- and power spectra from the preprocessed data segments, by tapering each of these with a variable set of discrete prolate spheroidal (Slepian) sequences. The used epochs of data of 400 ms duration, yielded to a frequency resolution of 2.5 Hz. A broad powerand cross-spectra was calculated (0–100 Hz). Within these, three frequency bands were of interest (FOI) for further analysis: beta (15–30 Hz), low-gamma (32.5–47.5 Hz), and high-gamma (50–65 Hz). The focus on the first two FOI (beta and lowgamma) is based on mounting evidence of a high association of the coherence in these frequency bands (gathered range: ∼15– 50 Hz) with the quality of the neural drive to the muscle during walking and, therefore, with the integrity of the pyramidal system. Furthermore, they also show a high potential of being modifiable by several types of training intervention; e.g., gait training (Norton and Gorassini, 2006; Barthelemy et al., 2010; Petersen et al., 2012; Willerslev-Olsen et al., 2015; Kitatani et al., 2016). The focus on the third FOI (high-gamma) is based on its high association with corticospinal interaction effectiveness, as shown in experimental paradigms involving both motor and cognitive resources; e.g., reaction time readiness (Schoffelen et al., 2005), where the role of coherence in high-gamma during locomotion (e.g., gait) has been investigated in first promising research work (Clark et al., 2013).

Two sets of tapers were adopted in two runs, to obtain an optimal spectral concentration and sensitivity relative to the three FOI. For the FOI up to 30 Hz (i.e., beta) three tapers were used with a resulting spectral smoothing of ±5 Hz around each frequency bin, whereas for the FOI higher than 30 Hz (i.e., lowand high-gamma) nine tapers were used with a resulting spectral smoothing of ±12.5 Hz around each frequency bin. Usually, the beta frequency bandwidth is represented by ∼10 Hz and the gamma frequency bandwidth by ∼25 Hz (Schoffelen et al., 2011), thus supporting the utilization of this spectral smoothing strategy. The following equation was used to calculate power- and cross-spectra:

$$\left(S\_{\rm xy}(f)\right) = \left|F\_{\rm x}(f)\right| \times \left|F\_{\rm y}(f)\right|^{\*}\tag{1}$$

where Fx(f) [or Fx(f)] denotes the Fourier transform of the signal x (or y) relative to the frequency f and <sup>∗</sup> denotes the complex conjugate. In this analysis signal x and signal y represent proximal and distal EMG data segments, respectively. When x 6= y, Sxy(f) denotes the cross-spectra between signal x and signal y, relative to the frequency f. When x = y, Sxy(f) is reduced to Sxx(f) [or Syy(f)], which consists of the (auto) power spectra of the signal x (or y), relative to the frequency f. Single segments of power- and cross-spectra yielded after averaging across tapers were used in order to calculate the coherence estimate between proximal and distal EMG data, with the following equation:

$$\text{Coh}\_{\text{xy}} = \frac{\left| \langle \text{S}\_{\text{xy}} \rangle \right|}{\sqrt{\langle \text{S}\_{\text{xx}} \rangle \times \langle \text{S}\_{\text{yy}} \rangle}} \tag{2}$$

where h·i denotes the obtained power- or cross-spectra after averaging across data segments. Coherence is a spectral measure representing the linear correlation between signal x and signal y, where the estimate ranges between 0 and 1, with 0 representing no linear association and 1 perfect relation at a specific frequency f.

In previous studies, 70–100 heel-strikes (i.e., data segments) were used to calculate the coherence estimates (Norton and Gorassini, 2006; van Asseldonk et al., 2014). However, it has been shown that a rather small number of trials (i.e., 25 or 50) may result in a larger variability of gait-related coherence estimates, which tends to decrease as the number of segments increase (van Asseldonk et al., 2014). Notably, a large number

of trials (i.e., 200) has been shown to relate with a still rather large variability (i.e., 50%) to gait-related coherence estimates (van Asseldonk et al., 2014). This large variability may be problematic when comparing coherence estimates obtained from an unequal number of segments within several testing sessions (Maris et al., 2007). In this study, all the data segments from the obtained heel-strike events in all the measurement time points were used for further analysis (mean ± SD across the three measurement time points in normal and dual-task walking: 224 ± 96, 213 ± 102, 246 ± 89, 244 ± 106, 212 ± 102, 252 ± 80, respectively).

# Gait Analysis

Toe clearance (cm), was measured with the Physilog (Gait up Sàrl, Lausanne, Switzerland) wearable movement sensors (50 × 37 × 9.2 mm, 19 g). Data transfer to a computer for further analysis was allowed through a micro-USB port. Elastic straps fixed the sensors at the right and left forefoot of the participants to allow flat over ground gait analysis. Physilog provides a valid quantitative assessment of gait kinematic parameters (Aminian et al., 1999; Dubost et al., 2006; de Bruin et al., 2007).

A figure-8 walking path was settled up, by placing two structures at a distance of approximately 7 m, and a gait testing protocol with at least 50 gait cycles (five repetitions of the path; **Figure 3**) had to be accomplished by the participants. Such a number of gait cycles has been shown to be sufficient to reliably estimate gait kinematics parameters (König et al., 2014). Participants performed five repetitions of the figure-8 gait path during both single- and dual-task condition as well as both under preferred and fast walking speed. In the dualtask walking condition, participants counted backward in steps of seven after receiving a random given number between 200 and 250. The participants were told to count loud; otherwise, the trial was recorded as failure. The dual task-condition quantifies the automaticity of movement and multi-tasking capabilities (Abernethy, 1988; Wright and Kemp, 1992; Wulf et al., 2001) and assessed the control of MTC under dual task conditions as a marker of motor control (Hamacher et al., 2016). For each participant, we calculated the variability of toe clearance (Mills et al., 2008). A measure of variability (e.g., standard deviation) of MTC height in preferred walking is indicative for diminished gait control in older adults (Begg et al., 2007; Mills et al., 2008).

# Lower Extremity Functioning

The Short Physical Performance Battery (SPPB) was used to assess lower extremity functioning. The test battery contains (i) a balance test, (ii) a 4-meter gait test, and (iii) a 5-chair-rise test. Timed results from each test are categorized into variables ranging from 0 (unable to perform) to 4 (best performers) according to well-established cut-off values (Guralnik et al., 2000). The sum of the results [tests (i)–(iii); theoretically ranging from 0 to 12] is used for the analyses where 12 indicates the highest degree of functioning.

At the pre-clinical stage this test is a predictor of subsequent disability (Vasunilashorn et al., 2009) and is applicable in routine clinical settings for monitoring of functioning (Perera et al., 2005). Test administration criteria have been published at "www.grc.nia.nih.gov/branches/ledb/sppb/index.htm." The participants were tested within a single session lasting around 10 min.

# Montreal Cognitive Assessment

The MoCA "paper-and-pencil-test" screens cognitive domains such as memory, language, executive functions, visuospatial skills, attention, concentration, and orientation (Nasreddine et al., 2005; Julayanont et al., 2013). The maximal score possible is 30 points and reflects a quantitative estimate of the overall cognitive abilities (Koski et al., 2011). A cut-off score below 26 points is often taken as indicator of possible mild cognitive impairment and dementia. The instrument is sensitive to change (Tiffin-Richards et al., 2014) and available in German.

# Statistical Analysis; EMG-EMG Coherence

Statistical analysis of the EMG-EMG coherence estimates was performed as follows. The Z-coherence spectra z(f) was computed at subject level to account for the variability before performing statistical testing according previously reported procedures (Maris et al., 2007; Schoffelen et al., 2011) and calculating the estimated change between measurement time points, using:

$$Z(f) = \frac{(Tanh^{-1}(Coh\_1) - 1/(2tap\_1 - 2)) - (Tanh^{-1}(Coh\_2) - 1/(2tap\_2 - 2))}{\sqrt{(1/(2tap\_1 - 2) + 1/(2tap\_2 - 2))}} \tag{3}$$

where Tanh−<sup>1</sup> is the inverse hyperbolic tangent, Coh<sup>1</sup> denotes the coherence estimate in measurement time point n and tap<sup>n</sup> denotes the total number of tapers used for the spectral estimation of the frequency in measurement time point n. The total number of tapers (tapn) was obtained by multiplying the tapers used (in this analysis either 3 or 9, as described above) by the number of data segments for each subject and measurement time point. The individual 1Z-coherence spectra obtained using equation (3) were performed by subtracting, for each frequency band, the 1st measurement time point from the 2nd (1PRE) and the 2nd measurement time points from the 3rd (1POST). For the study design here presented, the 1PRE was considered the "control" condition while 1POST represented the "experimental" condition. This is assuming that no significant difference is expected from pre-measurement time points one to two, and a significant difference is expected between the 2nd and the 3rd measurement time points.

Statistical inference was based on pooled individual 1PRE and 1POST separately for each using a non-parametric permutation test approach as described elsewhere (Maris et al., 2007; Schoffelen et al., 2011). Briefly, pooled 1PRE and 1POST for both normal and dual-task walking condition and for each frequency band (where the average across frequencies, within frequency band, was used) were tested for significant differences from 0 by approximating the p-values by 100000 Monte Carlo permutations. The level of significance was set

FIGURE 3 | Pooled data for both normal and dual-task walking (in grayish and reddish color palette, respectively), across the three measurement time points (1st pre-intervention, 2nd pre-intervention and post-intervention, denoted by the numeric subscripts: 1, 2 and 3, respectively) at different preprocessing stages: Proximal and distal emg sensors over the tibialis anterior muscle (TA) were first filtered (A1–A3) and then rectified (B1–B3). The filtered and rectified emg data were then epoched from the heel strike (excluded) to the preceding 400 ms before performing spectral analysis of frequency (C1–C3) and intramuscular coherence (D1–D3) between proximal and distal TA (TA-TA). Then, Z-transformed coherence of the 1st pre-intervention measurement (D1) was subtracted from the Z-transformed coherence of the 2nd pre-intervention measurement (D2) as well as (D2) from the post-intervention measurement (D3) (1PRE and 1POST, respectively). In the Dual-Task condition (red color) of 1POST, statistically significant difference of 1z-coherence resulted in the averaged beta frequency band of the spectrum (i.e., 15–30 Hz).

to α = 0.05 and a two-tailed test was adopted and corrected by multiplying the p-values with a factor of two prior to α thresholding.

# Statistical Analysis; Secondary Outcomes

Normality was tested for the remaining outcomes data before analysis with the Shapiro–Wilk test. Then the Friedman test was used to test for differences between time points because of non-normal data distributions. Post hoc analysis with Wilcoxon signed-rank tests and Dunn-Bonferroni correction was conducted when the Friedman test revealed significant values. All values are reported as means ± 95% confidence intervals (CI) and all statistical analyses were performed with SPSS.

Pearson's correlation was used for Effect Size determination with r = 0.1 meaning "small", r = 0.3 "medium," and r = 0.5 meaning a "large" effect (Cohen, 1988).

# RESULTS

Demographics and clinical characteristics of included participants (N = 20) is summarized in **Table 2**.

# EMG-EMG Coherence

All the control 1PRE conditions revealed non-significant differences in both normal (p = 0.815, p = 0.428, and p = 0.963, for beta, low and high-gamma FOI, respectively) and dual-task walking (p = 0.503, p = 0.056, and p = 0.066, for beta, low and high-gamma FOI, respectively).

Regarding the coherence estimates during normal and dualtask over ground walking, the non-parametric permutation test showed a significant difference against 0 in the experimental 1POST condition in dual-task walking and beta FOI (p = 0.002), while no significant difference was observed in the low and high gamma FOI (p = 0.307 and p = 0.372, respectively). No significant difference was observed in the normal walking condition in both beta, low and high gamma FOIs (p = 0.168, p = 0. 257, and p = 0.715, respectively).

# Gait Analysis and MTC

No significant change over time for MTC was observed for walking under single task condition.

Walking under dual task condition showed significant change over time in MTC for both the left [χ 2 (2) = 7.46, p = 0.024, n = 20] and right [χ 2 (2) = 8.87, p = 0.012, n = 20] leg.

Post hoc testing (Dunn-Bonferroni) revealed the difference (left leg) to be between time points 1 and 3: p1−<sup>3</sup> = 0.027; p1−<sup>2</sup> ≥ 0.9; and p2−<sup>3</sup> = 0.173. Effect sizes were large and mediumlarge (time point 1 – 3 r1−<sup>3</sup> = 0.58 and time point 2 – 3 r2−<sup>3</sup> = 0.42). A small-medium Effect resulted for time point 1 – 2 (r1−<sup>2</sup> = 0.16).

Post hoc testing (Dunn-Bonferroni) revealed the difference (right leg) to be between time points 1 and 2: p1−<sup>2</sup> = 0.022; p1−<sup>3</sup> ≥ 0.9, p2−<sup>3</sup> = 0.066). Effect sizes were large (time point 1 – 2 r1−<sup>2</sup> = 0.6 and time point 2 – 3 r2−<sup>3</sup> = 0.51) and small (time point 1 – 3 r1−<sup>3</sup> = 0.08).

# Secondary Outcomes

The inferential statistical testing for the secondary outcomes, revealed no significant changes in lower extremity functioning: χ 2 (2) = 3.68, p = 0.159, n = 20, whereas a significant change in MoCA scores over time was observed with χ 2 (2) = 11.76, p = 0.003, n = 20). Pairwise comparisons revealed this difference to be between time points 1 and 3 (p1−<sup>3</sup> = 0.008; p1−<sup>2</sup> = 0.081; and p2−<sup>3</sup> ≥ 0.9) (**Table 3**).

The amount of points achieved in the video games in the final training week was significantly higher compared to the 1st week of training for all video games played ("Simple": z = −3.659, p < 0.001; "Targets": z = −3.920, p < 0.001; "Divided": z = −3.920, p < 0.001; "Simon": z = −3.884, p < 0.001; "Flexi A+B": z = −3.921, p < 0.001; "Snake": z = −3.921, p < 0.001; "Tetris": z = −3.809, p < 0.001) (**Table 4**).

# DISCUSSION

This study aimed to explore whether video game-based training effects on the central drive to the ankle dorsiflexors during over ground walking and on MTC as measure of lower limb control. We hypothesized that this type of training would lead to changes in the corticospinal transmission to TA muscles and, thus, to improved motor control of ankle dorsiflexion while walking. The main finding was an observed change in the experimental 1POST condition in dual-task walking and the beta FOI which, thus, indicates the intervention effected on neural drive through enhanced quality of the neural drive (Norton and Gorassini, 2006; Barthelemy et al., 2010). This finding closely resembles investigations where intramuscular coherence in the beta frequency has shown to be related to neural drive to the muscle and, thus, the control of gait (Petersen et al., 2012; van Asseldonk et al., 2014; Kitatani et al., 2016). Furthermore, this finding gets supported by the observed change of the mean value of MTC that is accompanied by decreasing standard deviation values. Evidence of a possible linkage between lower extremity activation through training and changes in the neural drive also

de Bruin et al. Facilitate Central Drive to the Ankle Dorsiflexors

TABLE 2 | Demographics and clinical characteristics of participants (N = 20).


SD = standard deviation; n = number of participants.

TABLE 3 | Resulting MTC, SPPB, and MoCA values at the different measurement time points.


TP = time point; ST = single task; DT = dual task; min = minimum toe clearance; SD = standard deviation.

stem from a study from Dalgas et al. (2013). These authors showed that lower extremity resistance training in multiple sclerosis improves the neural drive to lower limb muscles. Similar results for resistance training for non-impaired individuals were reported (Aagaard et al., 2002). However, changes in the neural TABLE 4 | Points achieved in the video games in the 1st week vs. the last week.


The points from seven games were used for the analysis. It was not possible to extract the points from the game "Seasons." n = 20 participants completed all 18 training sessions and were used for the analysis in this table.

drive are also seen with other forms of initiating chronic physical activity (Enoka, 1997). This study is the first to show enhanced neural drive due to an exergame intervention. This finding is relevant for our society that is currently challenged to find an answer for supporting public health policies aimed at helping senior citizens achieving the goals of primary and secondary (i.e., reducing readmission rates) prevention to remain independence in functioning (McCaskey et al., 2018). Physical activity and exercise for older adults may help in this context to foster physical and cognitive functioning at the highest possible levels (DiPietro, 2001; American College of Sports Medicine, 2004; Elsawy and Higgins, 2010; de Souto Barreto et al., 2016; Karssemeijer et al., 2017) and considering that motivation to continue and adhere to conventional exercise is often difficult (Phillips et al., 2004). More research that aims at establishing the best ways to encourage older adults to be more physically active in the long term is needed (Bennett and Winters-Stone, 2011). The incorporation of progressively intense but short exercise as part of a tailored and combinatory program; e.g., with exergames, may be beneficial (Hwang et al., 2016). This type of training is also feasible for co-morbid geriatric patients Guadalupe-Grau et al. (2017).

The exergame used in this study is a motor-cognitive exercise from which can be hypothesized that it improves the synapse communication in brain networks responsible for movement coordination and execution (Eggenberger et al., 2016; Schattin et al., 2016; McCaskey et al., 2018). This, in turn, could positively influence the communication from the motor area of the brain to the muscles. Previous studies pointed to the potential of video game-based training and highlighted that video games that are specifically designed for training purposes improve walking (Pichierri et al., 2012a,b; Fraser et al., 2014) and effect on the brain (Eggenberger et al., 2016; Schattin et al., 2016). Emerging evidence indicates that part of the age-associated loss in muscle strength is due to an impaired communication between the brain and the effector organs (Clark and Manini, 2008; Manini et al., 2013); e.g., the muscles of the lower extremities. Furthermore, such deficits in the neural drive can lead to much of the muscle weakness observed in older adult populations (Clark and Taylor, 2011). Improved neural drive to the lower extremity muscles might explain why strength gains in lower extremity muscles are observable following exergame interventions (Jorgensen et al., 2013; Sato et al., 2015).

Older adults exhibit large muscle activation deficits (Clark and Fielding, 2012) which may explain the divergence between the rates in loss of strength and muscle mass, and may account for up to one third of loss in force production. Furthermore, traditional rehabilitation programs that use conventional strength training in patients with impaired muscle activation may not be optimal to reverse the loss in muscle strength (Stevens et al., 2003). Part of the muscle strength loss seen in older adults may be due to qualitative and quantitative changes in the motor cortex that negatively impacts on voluntary activation of the muscles (Clark and Taylor, 2011). Voluntary activation can been defined as the "level of voluntary drive during an effort" (Gandevia, 2001; Taylor, 2009). The descending drive from the motor cortex is considered the major determinant of the timing and strength of voluntary contractions (Clark and Taylor, 2011). Motor unit firing rates for the ankle dorsiflexor, for example, show slower rates for older than younger men (Roos et al., 1997; Todd et al., 2003). It can be hypothesized that training programs that focus on aspects of voluntary muscle activation in addition to more conventional types of resistance training, such as in the exergame we used, may result in greater strength gains in individuals that show larger voluntary activation deficits. Future longitudinal studies that use such combinations are warranted.

Our result seems at variance with the lack of an effect on the lower extremity function measured by SPPB. However, when we observe the values for all measurement time points for this parameter, it becomes clear that we may have possibly suffered a ceiling effect for this measurement in our sample. The values for this measure indicate we had a high level of lower extremity functioning in our selected senior sample.

Although we had a significant change in the MoCA outcome measure, this finding should also be interpreted with prudence in relation to its clinical relevance. First, we had high values for this measure indicating high levels of cognitive functioning in our sample and, second, we know from previous research that we can only be confident that an observed change would not be due to measurement error when we see individuals changing their score with 4 or more points (Feeney et al., 2016) and this should be two points or above for group assessments (Krishnan et al., 2017). Based on these values it becomes clear that there was not enough room for meaningful improvement for this outcome in our sample.

### Limitations of the Study

A limitation of our study relates to the testing threat known to possibly occur in pre-post research design. It might have been that testing our subjects with for example the MoCA at pretest made some of the participants more aware of cognitive skills and, hence, "primed" these individuals for the test so that when they repeated the measurement they were ready for it in a way that they wouldn't have been without the pretest. Mortality Threat, used metaphorically here, might have been another limitation. This means that people are dropping out of the study. It is difficult to estimate whether the observed loss of three individuals (±10%) between pretest and posttest was non-trivial notwithstanding that the reasons reported for dropping out were not related to the intervention. However, compared to rates that might be expected for community dwelling older trainees after 12 months (Nyman and Victor, 2012) the dropout rate seems acceptable and in line with what could be expected.

The used research design, although having strengths, also has some limitations attached to it. To deal with the single group threats to internal validity this study should be replicated using a more stringent research design in which a control group is considered; e.g., a randomized control design that also considers blinding of participants and assessors where possible. In this scenario, we would have two groups: one receiving the exergames and the other one doesn't with the aim of ruling out the single-group threats to internal validity.

Overall we can conclude that the initiation of an exergamebased training in upright standing position indicates to improve neural drive to the lower extremities in older adults and seems an acceptable form of physical exercise for this group. Further studies with more stringent designs are needed to refute or confirm this finding.

# DATA AVAILABILITY STATEMENT

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

# ETHICS STATEMENT

The ethics committee of the ETH Zurich, Switzerland (EK 2017- N-22) approved the study protocol. Before any measurements were performed, an informed consent according to the Declaration of Helsinki was administered and signed by each eligible participant.

# AUTHOR CONTRIBUTIONS

EB and FG developed the research question. The concept and design were established by EB while FG acted as methodological council. NP and LR conducted the data acquisition, analysis, and interpretation of the results (secondary outcomes) with editing and improvement by EB and FG. FG performed EMG data analysis and interpretation of the EMG results which was edited and improved by EB. EB produced a first version of the manuscript. FG substantially revised the manuscript to bring it to its current version. All authors have read and approved the final manuscript.

# REFERENCES

fnagi-11-00263 September 19, 2019 Time: 14:58 # 11


physical activity and exercise for older adults living in long-term care facilities: a taskforce report. J. Am. Med. Dir. Assoc. 17, 381–392. doi: 10.1016/j.jamda. 2016.01.021



screening tool for mild cognitive impairment. J. Am. Geriatr. Soc. 53, 695–699. doi: 10.1111/j.1532-5415.2005.53221.x


Rowe, J. W., and Kahn, R. L. (1997). Successful aging. Gerontologist 37, 433–440.



in tibialis anterior muscle. PLoS One 9:e88428. doi: 10.1371/journal.pone.008 8428


**Conflict of Interest:** EB was a co-founder of dividat, the spin-off company that developed the video step platform used for the training of the seniors and is associated to the company as an external advisor. No revenue was paid (or promised to be paid) directly to EB or his institution over the 36 months prior to submission of the work.

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 de Bruin, Patt, Ringli and Gennaro. 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 Motion of Body Center of Mass During Walking: A Review Oriented to Clinical Applications

Luigi Tesio1,2 \* and Viviana Rota<sup>2</sup>

<sup>1</sup> Department of Biomedical Sciences for Health, Università degli Studi, Milan, Italy, <sup>2</sup> Department of Neurorehabilitation Sciences, Istituto Auxologico Italiano, IRCCS, Milan, Italy

Human walking is usually conceived as the cyclic rotation of the limbs. The goal of lower-limb movements, however, is the forward translation of the body system, which can be mechanically represented by its center of mass (CoM). Lower limbs act as struts of an inverted pendulum, allowing minimization of muscle work, from infancy to old age. The plantar flexors of the trailing limbs have been identified as the main engines of CoM propulsion. Motion of the CoM can be investigated through refined techniques, but research has been focused on the fields of human and animal physiology rather than clinical medicine. Alterations in CoM motion could reveal motor impairments that are not detectable by clinical observation. The study of the three-dimensional trajectory of the CoM motion represents a clinical frontier. After adjusting for displacement due to the average forward speed, the trajectory assumes a figure-eight shape (dubbed the "bow-tie") with a perimeter about 18 cm long. Its lateral size decreases with walking velocity, thus ensuring dynamic stability. Lateral redirection appears as a critical phase of the step, requiring precise muscle sequencing. The shape and size of the "bow-tie" as functions of dynamically equivalent velocities do not change from child to adulthood, despite anatomical growth. The trajectory of the CoM thus appears to be a promising summary index of both balance and the neural maturation of walking. In asymmetric gaits, the affected lower limb avoids muscle work by pivoting almost passively, but extra work is required from the unaffected side during the next step, in order to keep the body system in motion. Generally, the average work to transport the CoM across a stride remains normal. In more demanding conditions, such as walking faster or uphill, the affected limb can actually provide more work; however, the unaffected limb also provides more work and asymmetry between the steps persists. This learned or acquired asymmetry is a formerly unsuspected challenge to rehabilitation attempts to restore symmetry. Techniques of selective loading of the affected side, which include constraining the motion of the unaffected limb or forcing the use of the affected limb on split-belt treadmills which impose a different velocity and power to either limb, are now under scrutiny.

### Edited by:

Eric Yiou, Université Paris-Sud, France

#### Reviewed by:

Manh-Cuong Do, Université Paris-Sud, France Romain Tisserand, University of British Columbia, Canada

> \*Correspondence: Luigi Tesio luigi.tesio@unimi.it; l.tesio@auxologico.it

#### Specialty section:

This article was submitted to Neurorehabilitation, a section of the journal Frontiers in Neurology

Received: 27 March 2019 Accepted: 02 September 2019 Published: 20 September 2019

#### Citation:

Tesio L and Rota V (2019) The Motion of Body Center of Mass During Walking: A Review Oriented to Clinical Applications. Front. Neurol. 10:999. doi: 10.3389/fneur.2019.00999

Keywords: walking, body center of mass, pathological gaits, system approach, gait rehabilitation

# WALKING: MOVING THE BODY SEGMENTS IN ORDER TO TRANSLATE THE BODY SYSTEM

A huge amount of research, a true odyssey (1), has been dedicated to the physiology of human walking. Although increased physiological knowledge has deepened our understanding, clinical science is still waiting for a consensus on the best strategy for diagnosis and treatment of many walking impairments. Human walking includes motion of virtually all body segments, meaning that alteration of the motion of any one segment induces adaptive movements across the whole body. In some cases, local malfunctioning is responsible for the impairment, while in others, such as individuals presenting with balance deficits or paresis due to diffused lesions of the central nervous system, it may be difficult to interpret the numerous concurrent alterations and identify critical targets for clinical observation. These alterations may be causes or effects of the underlying impairment and might, therefore, represent either additional impairments or useful adaptive mechanisms. This review is focused on the study of walking seen as the translation of the body system as a whole, represented by its center of mass (CoM) and aims at demonstrating that abstracting from segmental motions may help clinicians interpreting and possibly, treating the segmental impairments. Wide fields of research relating to CoM motion, such as studies on running or on the various forms of gait of legged animals, such as walking, running, trotting, canter, galloping, hopping, and so on, were treated tangentially to the main topic or disregarded. Studies of metabolic energy expenditure during walking, which can be considered as a form of system approach, were only treated marginally when the results were closely related to the CoM motion. Only studies of steadystate human walking on level ground in a straight direction, both in health and disease, were considered. Studies of the physiology of gait initiation (2) and termination (3) were ignored, along with a number of others including those of walking up- and downhill (4), along curved trajectories (5), across expected and unexpected obstacles (6), and in hypo- and hyper-gravity (7) among many others. The general principles of the physiology of walking in Man and, in general, in legged animals, are extensively covered in excellent, comprehensive textbooks (8–10).

# The Body System as a Whole: The Concept of Center of Mass

The body system as a whole may be represented, from a mechanical standpoint, by its CoM. The CoM of a distribution of mass is the unique point in space whose linear acceleration is determined only by the total external force acting on the system, without effects due to internal forces (11). When applied to the CoM, such force causes a linear acceleration without angular acceleration. One may also describe the CoM as the unique point which invariably lies in planes dividing the body into two parts, sharing the same moment of inertia (see **Note S1** for a mathematical description). The CoM usually moves within the body when body segments are displaced with respect to each other. This "center" may even lie outside the body mass. In a donut, the CoM lies in the hole. By keeping the CoM outside the body and below the bar by arching his body, Dick Fosbury was able to unexpectedly win the high-jump gold medal in the 1968 Olympic games in Mexico (12). These familiar examples show that, albeit virtual, the CoM is far from being imaginary. Some help in grasping the CoM concept in the study of walking may come from imagining the displacement of a point on a screen tracking a GPS transmitter embedded in the belt of a walking subject (although, the real CoM is free to move outside the belt). Whichever examples and metaphors are suggested, it remains true that thinking of the CoM, a virtual point, requires abstraction from morphology and therefore, from anatomy, which is the reason why it is often neglected in clinical studies.

# Highlighting the Two Opposing Viewpoints

Two distinct viewpoints still characterize the study of human legged animals' walking (13).

# Walking as a Sequence of Joint Rotations

The earlier, yet still dominant, viewpoint considers walking to be a series of cyclic rotations of the limbs and trunk, referred to as the "segmental" approach or viewpoint from here on. To illustrate this concept, one can imagine a man walking on a treadmill; the cyclic nature of the joint rotations is apparent. The beginning of the cycle is a matter of convention; usually, but not exclusively, this is considered as the instant of heel strike. The series of events that takes place between subsequent heel strikes of opposite sides of the body is called a step. Two subsequent steps make a stride.

### Walking as the Translation of the Body System as a Whole

The less widely accepted viewpoint highlights that walking has the primary goal of achieving forward translation of the body system as a whole (locomotion), although this results from the interplay between gravity and the coordinated and cyclic contraction of numerous muscles. In this view, the body system as a whole can be represented by its CoM. This will be referred to as the "system" viewpoint or approach from here on.

# The CoM Motion: The Cinderella of Gait Analysis

The umbrella term "gait analysis" describes the synchronous recording of many segmental variables, which has become accessible to clinical settings in recent decades, due to the rapid progression of technology [for kinematics, see (14)]. However, the CoM motion is not yet considered in routine gait analysis. This can be a reason for its limited success in clinical practice, as it will be discussed later in this Review.

# MODELING THE WALKING PHENOMENON

Anthropometric studies have demonstrated that the CoM of quietly standing humans lies approximately a few centimeters in front of the lumbosacral joint, about 57% of the body height from the feet plants, in both children and adults (15). Sophisticated modeling has allowed the estimation of its location in various postures and during various motor tasks; again, in children (15) and in adults (16, 17). Tracking the CoM during walking, however, is much more difficult.

# The CoM Motion as the Oscillation of an Inverted Pendulum: An Old Intuition

How to measure the motion parameters of the invisible CoM during walking has been a challenge for generations of researchers. A first approximation was reducing the body motion to the visible motion of the sacrum. This approximation is rather rough. In fact, the movement of the CoM is affected by the movement of individual body segments, while the sacral markers are not (18). However, this simplification had the advantage of reinforcing the very old intuition that the CoM moves up and down and accelerates and decelerates forward during each step, like an inverted pendulum, of which a mechanic metronome is perhaps the most familiar example.

A famous clinical article has suggested that smoothing the CoM oscillations is the aim of some of the segmental motions of the lower limbs, such as pelvic tilt and knee flexion, named the six "major determinants" of gait (19) (**Figure 1**).

The authors suggested that these six determinants deserved special attention in clinical observation. Unfortunately, measuring energy changes and displacements of the human CoM proved to be considerably more challenging than intuitively capturing the pendulum analogy.

# Modern Methods for Analyzing the CoM Motion

Nowadays, numerous and valid methods exist to observe and measure the CoM motion. In this review, articles based on the "sacral marker" simplification will be overlooked. Articles will be considered, based on the analysis of ground reaction forces (the so-called "double integration" or Newtonian method) or on kinematic analysis of body segments (usually, through optoelectronic "capturing" of retroreflective skin markers) as per anthropometric modeling. Both methods provide reliable and very similar results with respect to the CoM motion. The corresponding technicalities and criticisms are summarized in the following paragraphs and in **Note S2**. The overground walking mechanism can be reproduced and measured more easily on a treadmill. The treadmill imposes a known and constant average velocity, thus enabling the investigator to record several reproducible steps in a matter of seconds. Experimental sessions can be very short, which is advantageous for the examination of children or impaired adults. The treadmill also imposes reproducibility of velocity across time points. Furthermore, these devices can be mounted on force sensors, thus providing a virtually unlimited distance with embedded force sensors in a limited space. Although the two walking modalities, in theory, are mechanically equivalent, they present with some behavioral differences. The treadmill approach has both limitations and advantages which are discussed in a dedicated paragraph and in **Note S3**. Nevertheless, given their substantial similarity, results based both of ground and treadmill walking will be presented here. Some consideration will be given here to walking on treadmills made of two independent belts running at different velocities (split- or dual-belt treadmills). These devices are of interest because they induce an artificial form of limping, and might offer possible therapeutic approaches to help unilaterally impaired subjects recover symmetric motion of the CoM (see further paragraphs).

# The Three-Dimensional Path of the Center of Mass: Seeing the Invisible

The 3D trajectory of the CoM is among the most recent outcomes of an old line of research. The only abstraction required to the clinician is subtracting the average forward velocity of the body: stated otherwise, one can imagine observing a subject walking on a treadmill, with zero average velocity with respect to the ground. The subject's trunk will show rhythmic back-forth, and left-right oscillations, with no average forward displacement. This approach, adopted here, will not require delving into the underlying principles of physics and engineering. Also, it might better fit the clinical assessment, still based mostly on visual observation of gait impairments.

# From 2D to 3D Analysis. Considering the Lateral Motion

The pendulum-like CoM motion was usually analyzed in the sagittal plane although, in principle, the same technology could well be applied to the frontal and horizontal planes. On the other hand, most of the work to move the body system and its segments, and its largest displacements, can be observed in the sagittal plane (20) so that most of the conclusions on work production do not change remarkably when the other planes are also considered. The lateral displacement of the CoM, however, may be of crucial importance from a clinical standpoint, because lateral stability is challenged in many pathologic conditions that are caused by neurologic or orthopedic impairments. At each step, the CoM oscillates laterally toward the supporting leg, then swings toward the opposite leg during the next step, producing an inverted pendulum-like mechanism in the frontal plane, too. The composite motion of the sagittal and frontal pendulum during a stride follows a curved path around the line of progression. If advancement due to the average forward velocity is subtracted (instantaneous velocity still undergoes periodic changes, of course) the CoM path assumes a closed figure-eight shape, upwardly concave in the frontal plane, with an overall length of about 18 cm (**Figure 3**).

A systematic study of the 3D path of the CoM in healthy adults at various walking velocities, based on the double-integration method (see above, and **Note S2**), was first published in 2010 (21). This elegant representation of CoM motion was dubbed the "bow-tie," and planar projections of this figure were first demonstrated in two studies: one related to the 3D motion of the CoM in healthy adults, patients with unilateral hip-joint replacement and patients with post-stroke hemiparesis (22); the other analyzed the CoM path in the frontal plane only, in children with various forms of cerebral palsy and in healthy controls (23). The bow-tie shape has since been confirmed and mathematically modeled in an independent study (24).

# Linking the CoM Path to Walking Velocity

Most velocity-related changes in the total length of the curved bow-tie path occur due to the shortening of its lateral oscillations (**Figure 2**).

Furthermore, the path length shows a U-shaped relationship to the average forward velocity with a minimum of around 1.3 m s−<sup>1</sup> , which is very close to the optimum speed with regards to the minimization of metabolic energy expenditure and maximization of passive exchange between kinetic and gravitational potential energy of the CoM (21).

Notably, the upward concavity of the bow-tie increases with increasing velocity, so that lateral oscillations of the CoM, unlike the total path length, are monotonically restricted from about 10 cm at 0.3 m s−<sup>1</sup> to about 5 cm at 1.4 m s−<sup>1</sup> . This might explain why ataxic subjects are, paradoxically, more stable at faster walking speeds. Although this phenomenon is well-known by clinicians, it is documented only anecdotally (25, 26). This has been interpreted as a favorable shift toward a more automatic spinal control of gait bypassing the impaired vestibular and/or cerebellar inputs (26). However, as **Figure 2** suggests, this may be a purely mechanical phenomenon.

# Dynamics of the Path of the Center of Mass: The Three-Dimensional Curvature

The dynamics of the CoM path, that is, its trajectory, provide more information than the morphology. The instantaneous curvature, which is the inverse of the radius, and the tangential speed were measured along the bow-tie, in parallel with the instantaneous values of the recovery of mechanical energy (Rinst) as a function of the distance (arc-length) covered during one stride (27). As **Figure 3** shows, one phase of external power production falls within the push-off period around the double stance (see below for further description); while the other phase is approximately coincident with the inversion of the lateral oscillation in the middle of the single stance (**Figure 3**).

The time course of Rinst during one stride is presented in **Figure 4** (second tracing from top). This course appears jerky: Rinst remains above 70% for much of its path and peaking close to 100%, a value which indicates a fully ballistic pendulumlike motion. There are two short phases of the step, however, when Rinst drops suddenly to zero and the curvature (bottom curve) shows evident peaks. The larger peak occurs during the single-stance period (labeled b<sup>c</sup> in **Figure 4**), around the lateral redirection of the CoM. Incidentally, these curvature peaks may differ between the left and right step, as in the present case. Smaller curvature peaks occur during double stance (a<sup>c</sup> in **Figure 4**). Interestingly, while a<sup>c</sup> is coincident with a peak of external power (uppermost curve) as would be expected), the larger b<sup>c</sup> curvature peak is synchronous with a narrow peak of negative external power (i.e., power absorbed by muscles that are active during elongation). In other words, the lateral oscillation of the CoM during the single-stance period requires active braking, followed by active propulsion (no passive energy exchange, Rinst = 0). These events allow the CoM to be actively redirected toward the opposite side. It is known that lateralexternally oriented ground reaction forces occur throughout the entire double-stance and early single-stance periods but not during late stance (28). The internal redirection during single stance is thus a critical phase of the step, in which the passiveballistic motion must be suddenly and briefly superseded via active neural control imposing a small and swift absorption of external power in order to brake the CoM, just before the active redirection toward the opposite side. If this mechanism fails, a lateral fall might ensue.

FIGURE 3 | The "bow-tie" path of the body center of mass (CoM) averaged over a series of strides. In the upper and the lower row of panels, two representative subjects walked at 0.96 and 1 m s−<sup>1</sup> , respectively. Subject TG (upper row column) was a man, 23 years of age, 1.78 m in height, 74 kg in weight with a leg length (superior iliac spine to ground) of 0.91 m. Subject BG (lower row) was a man, 30 years of age, 1.73 m in height, 66 kg in weight with a leg length (superior iliac spine to ground) of 0.88 m. Six subsequent strides were recorded per subject. In the (A) column, the curve gives the CoM path and thickened segments correspond to the double foot-ground contacts. The human form outline (arbitrary size) facilitates recognition of the spatial orientation of the figure. In columns (B–D), the scaling of the curves is modified for graphic clarity. (B) Foot-ground contact phases. The arrow tip marks the conventional beginning of the stride; i.e., the point where the forward speed reaches a maximum after the right foot strike. (C) Thickened tracts mark the segments of the trajectory where the curvature is classified as low, at or below 0.05 mm−<sup>1</sup> (i.e., radius of curvature at or above 20 mm). (D) Thickened tracts mark the segments of the trajectory where the instantaneous recovery of mechanical energy, Rinst, is at or below 0.7. It can be seen that the phases of high curvature (thinner tract in C) largely overlap the single stance phases (thinner tracts in B), and phases with low values of energy transfer (R, thicker tracts in D). Taken from Tesio et al. (27), used with permission.

# Trajectory of the Center of Mass and Balance

The peaks of curvature of the CoM path and its lateral oscillations provide an interesting focus for the assessment of balance during walking. Balance deficits can result in falls mostly during walking, and mostly into the lateral direction (29). Lateral instability is, therefore, a likely marker of future falls. A relevant study demonstrated that in forward-induced stepping the main difference between elderly faller and young or old non-fallers is the greater sideways motion of the body toward the stepping side and a more laterally directed foot placement. Potential fallers thus seem to pre-plan this protective behavior (30). The hypothesis of careful neural control of lateral displacement of the body is in line with findings on treadmill walking, where healthy subjects adopt a wider base of support, compared to ground walking (see **Note S3**). This notwithstanding, the lateral margins of stability (the distance between the outer margins of the base and the vertical CoM projection, see below) are kept unchanged (31). Therefore, an understanding of the CoM trajectory and its lateral width during the stride has high clinical relevance. A more precise analysis of the risk for fall may be possible through knowledge of the spatial relationships between the position of the CoM and the base of support (both unior bi-pedal) in dynamic conditions. A seminal paper (32), defined the vector quantity, extrapolated CoM (XCoM), using Equation (1):

$$XCoM = CoM\_{position} + CoM\_{velocity} \sqrt{\frac{l}{g}} \tag{1}$$

where l is the lower limb length and g is gravitational acceleration. The margin of stability (MoS) is defined as the minimum difference between the perimeter of the base of support and XCoM. In adult walking, the lateral MoS is much lower than would be predicted by considering the CoM alone (say, 25 vs. 50 mm) (33). Computing the absolute position of XCoM with respect to the base of support requires knowledge of the base of support itself. The needed information can be available through kinematic analyses as will be detailed later in this review. Since the publication of Hof's paper in 2005, there have been plenty of studies investigating the size of the MoS, and even the "temporal" MoS, i.e., the time remaining before the MoS is trespassed, during walking in various populations of healthy subjects of all ages and under various paradigms, implying perturbations of the walking surface (30, 31, 34–40). From this literature, the general principles seem to emerge that lower margins of stability after perturbation indicate an actual risk for balance, while increased

margins of stability indicate a protective attitude, reflecting a latent instability.

The potential clinical usefulness of the XCoM analysis, capturing the balance variable during walking, cannot be underestimated. Not surprisingly, technical efforts are ongoing to simplify its estimation during walking, e.g., through a limited set of optical markers (41) or even through wearable sensors placed under heel and toes (42).

# The Path of the Center of Mass in Children and Adolescents

A recent study aimed to relate the CoM path to age in children (5–13 years old) and adults (23–48 years old) using metric distances normalized to subjects' height. Walking velocity was also normalized to height through the Froude number (43). Both the bow-tie shape of the CoM path and its U-shape relationship with walking velocity were preserved across age groups, indicating that these parameters decrease relative to body elongation. Increased age and height were associated with shorter height-normalized paths, due to narrowing of the lateral dimension. This indicates that reduction of the lateral oscillations of the CoM relative to body height may be taken as an index of the neural maturation of gait, as already foreshadowed in a previous study (23).

# THE TRANSPORT OF THE CoM AS A MATTER OF COST

# Mechanical Energy Changes in the Inverted Pendulum Motion

**Figure 5** gives a modern representation of the mechanical energy changes of the CoM in a healthy adult during a series of two

FIGURE 5 | Changes in mechanical energy of the center of mass (CoM) during one stride. This subject (sketched on top) was a woman, 36 years of age, 1.5 m in height, 50 kg in weight, walking at 1.53 m s−<sup>1</sup> . From top to bottom, the curves refer to the mechanical energy changes of the CoM due to motion in the forward, vertical and lateral directions (Ekf, Ev, and Ekl, respectively), and their sum, Etot = Ekf + E<sup>v</sup> + Ekl (note the scale difference for Ekl). The stride is considered to begin when Ekf reaches a maximum. Curves from three different strides, performed at about the same average speed (±10%), are superimposed. The bottom horizontal lines mark the time intervals in which the right or the left foot is on ground. The a and b labels indicate the increments of Etot, necessarily sustained by positive muscle work (external work, Wext). Adapted from Tesio et al. (20), used with permission.

subsequent steps (strides), beginning with the heel strike of the right, leading lower limb (ground contacts are marked by the interrupted horizontal segments below the curves). Data were obtained through the "double integration" method [(20), see **Note S2**].

The pendulum-like mechanism subtending the motion of the CoM is revealed from the time-course of the mechanical energy during a step. The mechanism allows the kinetic energy of the CoM, due to the forward velocity (Ekf), to be transformed into the sum (Ev) of gravitational potential energy (Ep) and kinetic energy due to the upward velocity (Ekv). The changes of kinetic energy due to the lateral motion (Ekl) are much smaller than those observed in the sagittal and vertical directions. In an ideal pendulum the friction forces of the air and at the hinge are neglected, the time course of Ekf is the mirror image of the time course of Ev, and the sum of these energies, Etot, remains constant; the oscillation would be perpetual and no work would be necessary to keep it in motion with respect to the ground ("external" work, Wext) (44). Of course, the mechanism is not ideal in walking, modeled as an inverted pendulum. **Figure 2** shows that the energy exchange is not complete, although oscillations of Etot are very limited. Actually, the mechanism allows a remarkable saving (Recovery, R) of Wext. R can peak at 100% (no muscle work required) and reach 60% as a mean along a step, at "optimal" velocities around 1.3 m s −1 . Increments of Etot, necessarily sustained by muscles, occur around the double stance (increment a) and during the single stance (increment b) (fourth curve from the top in **Figure 5**). These concepts are recalled in the **Note S4**, and further on in the main text.

Minimizing the cost of transport of the CoM seems a primary constraint of the walking mechanism. The average velocity selected spontaneously and, for any velocity, the step cadence and length, are very close to those allowing to minimize the total energy expenditure and Wext, and to maximize R (see **Note S4** for details). Also, the inverted pendulum mechanism is tenaciously maintained across all ages (see following paragraphs).

# The Inverted Pendulum During Walking in Children

Children aged 1–4 years are rarely capable of maintaining a constant average velocity during walking. This makes the estimation of velocity-dependent energy changes of the CoM very difficult. In general, it is agreed that a pendulumlike motion of the CoM appears around the age of 2 years. Estimates with large margins of uncertainty suggest that this mechanism achieves adult-like characteristics very soon thereafter (45). Above 4 years of age, mass-specific metabolism and weight-adjusted mechanical external power are higher for children than for adults when compared at the same walking velocity (46). However, this difference vanishes when velocities are compared after adjustment for the subject's height through the non-dimensional Froude number (46, 47). This suggests that the assumption of childadult geometric similarity, justifying the use of the Froude number, is tenable despite some concerns raised in the Literature (48, 49).

# The Inverted Pendulum During Walking in the Elderly

The effects of aging in subjects 65 years or older on various kinematic and dynamic walking parameters have been studied extensively. In two recent and robust meta-analyses, the impact of aging appeared modest (50, 51). In general, elderly people tend to spontaneously adopt a lower speed and a lower step length for any given speed compared with younger adults. This means that disentangling the effects of aging from those of forward velocity and step length with respect to dynamic parameters is challenging (52). In general, lower propulsion from the trailing leg, a lowering of the forward CoM accelerations, a decrease in braking the fall of the CoM at foot strike and a restriction of the mediolateral CoM sway seem to characterize the CoM behavior in individuals above 65 years of age (53, 54). The increase in forward speed precedes that of the upward velocity, as happens at lower speeds. The graph of vertical to forward velocity, the "hodograph" (28), was proposed as a potential index of age-related decline in gait performance. It should be noted, however, that in reality, elderly subjects compensate their limits by adopting a shorter step length (see the above considerations on this point). A study of dynamic balance during walking demonstrated that the mediolateral displacement of the CoM was comparable in elderly and young subjects (55), with a difference that elderly subjects exhibited a cautious narrowing of this displacement when the base of support was restricted. Also, the variability in CoM mediolateral velocity was found to be higher in elderly subjects. The search for a safer motion of the CoM has been suggested as the reason that elderly people, but not younger ones, decrease their average forward velocity, stride length, and minimum forward CoM velocity during the single-stance period when wearing socks compared with walking barefoot. This cautious behavior seems to counteract the known age-related decrease in plantar foot sensitivity (56).

# MATCHING THE SYSTEMIC AND SEGMENTAL VIEWS OF WALKING. MUSCLES AND THE TRANSLATION OF THE CoM

Given that minimizing the cost of transport is a primary constraint of the CoM motion, the neural control of gait must be designed accordingly.

# When Is Wext Produced?

In recent years, a small number of studies have measured the instantaneous changes of Etot (i.e., Wext) and Rinst (27, 57). These studies paved the way to the understanding of the role of individual muscles in generating the motion of the CoM. This information first requires knowledge of the timing of Wext output.

An early seminal study (44) demonstrated that Wext is provided in two reproducible phases of the CoM oscillations along the sagittal plane; specifically, when increases in Ekf (named increment a) occur and when the CoM is lifted (increment b). Both phase shift and size differences between its components may lead to increments of Etot. In general increment a is due to the fact that Ekf increases more than Ev decreases, while the opposite is true for increment b (44). A successive study demonstrated that the peak external power (work/time ratio) during increment a is about four times higher than that required to sustain increment b (20).

# Matching Segmental and CoM Power Production

Unfortunately, reports of the synchronous recording of motions of body segments and the CoM are rare. One practical reason is that parallel independent platforms are needed to record joint moments and torque during left and right leg stance, and platforms which allow at least one entire stride are required for the double-integration recording of CoM motion. In addition, a conventional segmental gait analysis must be also available simultaneously. In general, research questions were mostly focused on segmental or CoM motion rather than on their interaction, and some key questions have only started to be answered recently. The segmental approach conventionally defines step periods as the time interval between successive footground contacts, whereas Cavagna's studies considered a step to be the interval between successive peaks of Ekf, thus presenting a fully CoM-based concept of "cycle." This makes a comparison of segmental and systemic events between studies difficult.

In recent years, a correlation between the Etot changes and temporal step phases was formally proposed. Neptune et al. (58) identified four temporal phases dubbed "Regions." These were based on the zero-crossing of the external power applied to the CoM during one step: Region 1, the step-to-step transition of the leading leg as it absorbs the mechanical energy of impact (approximately the first phase of double support); Region 2, the raising of the CoM in early single-limb support; Region 3, the lowering of the CoM in late single-limb support, and Region 4, the step-to-step transition of the trailing leg as it propels the body (approximately the second region of double support) (**Figure 6**).

By comparing **Figures 5**, **6**, it can be seen that Cavagna's increments a and b (see also **Figure 2**) occur approximately during the step-to-step transition of the trailing leg (Region 4) and the early single-limb support (Region 2) phases, respectively. Hereinafter, the labels "increment a" and "b" will be adopted preferentially, although they will be also referred to as the Neptune's stride phases of their occurrence. In line with the literature (59), the term "push off phase" will also be used as a proxy of "step-to-step transition of the trailing leg."

# The Dominant Role of Plantar Flexors in Generating Wext

Clinical applications of knowledge of the CoM motion requires inferences on the underlying role of segmental impairments. In this view, the synchronous observation of changes of Etot (i.e., Wext, which is positive when Etot increases) and of muscle recruitment appears highly relevant. The sequence of activation of lower-limb muscles can be recorded using surface electromyography or fine-wire implanted electrodes (60). This sequence is well-known for adults (61, 62) and children (63) and is reported in many manuals on gait analysis [such as Figures 8.17, 8.18 of (64)]. These studies provided evidence that the plantar flexors are active, both mechanically and electrically, from midstance to mid push-off, Neptune's Regions 3 and 4, respectively (**Figure 6**), while the ankle dorsal flexors are active during early stance and late push-off, Regions 1, 2, and 4, respectively. Only minor activity is recorded from the knee and hip extensors, mainly at foot strike (**Figure 7**) (65).

The mechanical role of plantar flexors seems crucial, given that they are synchronous with the a increment of Etot, which is 3 to 4 times higher than the b increment at all velocities (**Figure 8A**) (20). In the asymmetric impairment caused by above- or below-knee amputation, the a increment is increased when the unaffected limb is trailing (N/P transition in **Figure 8B**); correspondingly, the plantar flexors may generate a peak power 3 times higher, compared to the next a increment sustained by the muscles of the amputated limb (P/N transition in **Figure 8B**) (66).

Recent research confirmed that the plantar flexors provide the highest power output during walking compared with other extensor muscles that act at the knee and hip (64, 67). This power is mostly released during the push-off phase and is thus expected to have a critical role in the advancement of the body system (68, 69). Active stretching of the plantar flexors occurs midstance, immediately before the push. This has led some authors (70, 71) to suggest that the plantar flexors group or at least the soleus (72) are primarily responsible for a braking action with respect to the CoM fall in late stance (**Figure 9A**).

Other authors proposed that the plantar flexors are used to accelerate the leg into the swing (73, 74), thus producing Wint. Given the methods that were used, these segmental interpretations did not consider the systemic role of the plantar flexors; however, they are compatible with a systemic interpretation. For instance, the braking action of the stretched plantar flexors during active lengthening cannot be ignored, yet it can be interpreted simultaneously with a mechanism that facilitates the remarkable production of positive ankle power during the successive shortening (69). Consistently enough, the work produced by these muscles has been shown to be four times more efficient than the work produced by the hip muscles to sustain the b increment of Etot during the single-stance period, the pendular phase of the step (68). This highlights the contribution of the braking action of calf muscles to the next increment a of Etot during Region 4 of double support. Meanwhile, their role in raising the mass of the swinging leg contributes to the lift of the CoM (59), and thus, to increment b of Etot, during the subsequent early single-limb support (Region 2). These studies show that appropriate modeling of positive work (i.e., when force is produced by shortening muscles) at lower-limb joints, considering both their rotational and translational energies, can explain the increments of Etot along a stride (75) (**Figure 9B**). In short, both the segmental and system viewpoints highlight the role of the push-off as the critical phase when muscle power is generated within the walking body system. There is converging evidence that the plantar flexors are the major "engines" involved in sustaining the increment a of Etot. Their contribution to leg swing means that they also contribute (together with knee and hip extensors) to increment b. This is confirmed by the increased power production during ankle plantar flexion compared with other sagittal joint rotations (67) and the increased power production during increment a compared with that of increment b of Etot. Clinical observations

confirm this conclusion, as is discussed in detail later in this review.

# The Role of Lower Limb Muscles in Children

The pendulum-like transfer of the CoM has adult-like characteristics from the age of 4 years, once the body height is conditioned out by transforming the average walking velocity into the non-dimensional Froude number. The only one exception is perhaps the wider lateral size (relative to body height) of the 3D path (the "bow-tie") of the CoM. Timing of joint power and electromyographic patterns of the lower limb muscles also seem independent of age (76).

# The Role of Lower Limb Muscles in the Elderly

Elderly people (say, above 65 years of age) tend to adopt a lower average velocity and, for any given velocity, a shorter step length, compared to their adult counterpart. These features appear to compensate for weakness, in particular, of the plantar flexors, and for subclinical balance impairment. A recent study confirmed the existence of age-related differences in terms of propulsion from the trailing leg (77) by comparing young subjects (mean age 24 years) with a sample of elderly subjects (mean age 74 years). The power of the plantar flexors was reduced in the elderly group at the step-to-step transition, which indicated that at speeds above 1 m s−<sup>1</sup> elderly subjects were unable to simultaneously increase the velocity of the CoM forward and upward. Once the lower velocity and lower step length were taken into account, the lowerlimb joint dynamics and estimated roles of individual muscles in propulsion of the CoM were found to be superimposable between young (24 years old on average) and elderly (74 years old on average) subjects (52). More recently, however, evidence has been reported for a distal-to-proximal redistribution of positive work across the plantar flexors and knee and hip extensors in elderly people (mean age 76 years) compared with younger people (mean age 22.5 years), when tested at the same walking velocities, during both level and uphill walking (78). The proximal shift of work production was higher with higher velocities. This is consistent with the fact that plantar flexors, compared with knee and hip extensors, work at closer to their maximum capacity (79). As a consequence, general age-related strength loss (80) may prevent the plantar flexors from providing the amount of work that they release during walking at a younger age. This notwithstanding, plantar flexors of elderly individuals never provide <50% of the positive work done by the lower-limb muscles during level walking, and thus they remain the principal engines of walking at any age (78).

each column of panels mark the average duration of the stance phase of the stride under examination (thick upper line) and of the simultaneous contralateral stride (thin lower line). (A) From left to right, the y-axis shows sagittal rotations of hip, knee, and ankle joints. (B,C) From left to right, the y-axis shows the joint torques and (Continued) FIGURE 7 | powers, respectively, of the hip, knee, and ankle joints. Joint power is labeled as generated (Gen) or absorbed (Abs) when the velocity of joint rotation and the joint torque have the same or opposite directions, respectively. During swing, torque and power at the lower limb joints are estimated through inverse dynamics according to anthropometric modeling of body segments. In each panel, the black band encases the mean ± standard deviation of the control sample (walking over ground), whereas the white bands encase the mean ± standard deviation of the sample (walking on the treadmill). Gray bands indicate the overlap between the data from the study (white and gray areas) and control samples (black and gray areas). (D,E) The electromyographic (EMG) signal envelopes from the semitendinosus (ST), vastus medialis (VaM), tibialis anterior (TA), and gastrocnemius medialis (GaM) of the right lower limb. For each muscle, amplitudes were standardized as percentages of the maximum voltage recorded from that muscle across all of the recorded strides (in microvolts: GaM, 174; TA, 197; VaM, 64; and ST, 114). Each band encases the 10th to the 90th percentiles of the EMG amplitudes from the corresponding muscle. The black and gray fillings closely follow the shortening and lengthening periods of the corresponding muscle, respectively. For ST, VaM, TA, and GaM shortening is assumed when the hip and knee are extending and when the ankle is dorsally or plantarly flexing, respectively. Sampling frequency for kinematic and dynamic data: 250 Hz. Treatment of the EMG signal: pre-amplification, 16-bit A–D conversion; sampling frequency: 1 kHz, rectification; smoothing through moving average: 80 ms window. Taken from Tesio and Rota (65), used with permission.

In summary, once the inverted pendulum mechanism is established, the role of individual muscles in CoM motion seems to remain largely unchanged from infancy to old age although the plantar flexors seem to become less dominant compared to the proximal muscles at advanced age. The main adaptations of walking to body size and shape, muscle strength and joint mobility are related to velocity and the step length that is adopted and attainable (50).

# CLINICAL SIGNIFICANCE OF THE MOTION OF THE CENTER OF MASS DURING WALKING

Human walking impairments are varied but, for simplicity, here they can be divided into symmetric and asymmetric forms, depending on the mechanical events observed during each of the two steps within a stride. A key concept, nearly a paradox, emerging from the Literature is that walking impairments may be associated with an overall normal total energy expenditure per unit distance, which in turn reflects a normal efficiency of the pendulum-like mechanism as an average between subsequent steps. Evidence is provided in the following paragraphs and is further discussed in **Note S5**.

# Symmetric Pathological Gaits

Most symmetric diseases and impairments, including many forms of Parkinson's disease, infant cerebral palsy, incomplete spinal lesions, and multiple sclerosis, typically involve a lower velocity during walking. Given the velocity adopted, a normal Wext,m and/or a normal 3D path of the CoM (see below) can be observed in symmetric impairments. This has been found to be the case in scoliosis (81), multiple sclerosis (82), obesity (83), non-freezing Parkinson's disease (84), and hemophilic arthropathies (85). In most pathologic conditions, steps tend to be even shorter (hence, their cadence higher) than is physiologically necessary for the lower speed that is adopted. In extreme cases, the feet never leave the ground in a so-called "shuffling gait" (86). Shortening the steps raises Wint,step (87) at all speeds, yet it minimizes Wext,step. In fact, smaller fluctuations in Etot are seen. This means that the power required from the plantar flexors at each step is decreased, as is the need for balance control given that accelerations of the CoM are minimized (see also **Note S4**). For these reasons, shortened, and thus more frequent, steps flag (non-specifically) an underlying difficulty negotiating the advancement of the body system. A very simple, yet reliable clinical index is provided by the ratio of step length to step frequency, which is referred to as the walk ratio, where a higher value indicates greater control (88–90).

# Asymmetric Pathological Gaits

Knowledge of CoM motion is particularly useful for the clinical assessment of asymmetric gaits. Various terms are used to describe gait asymmetries, which are based on a visual inspection and generally defined as limping or claudication. In clinical jargon, "escape" limp indicates that the single-stance period is minimized for the impaired lower limb with the opposite limb being brought to contact the ground quickly to unload the impaired limb in anticipation of its take-off. This allows the patient to keep the impaired limb off the ground for as long as possible. This is perhaps the most sensitive clinical sign of asymmetric impairment (see **Note S6**) (91), although it is nonspecific because pain, limited joint mobility, ataxia, weakness, or spasticity may all result in an escape limp. Furthermore, step length asymmetry may coexist. As a rule, the heel-to-heel sagittal distance during the double-stance period is longer when the affected limb is leading, also known as the anterior step, compared with when the unaffected limb is leading. This provides a visual index of the reduced propulsion from the impaired lower limb as its role in CoM propulsion is fostered by the rear position. Other clinical definitions refer to the most significant segmental alterations: "sickling" gait refers to exaggerated abduction of the swinging leg, usually due to reduced knee flexion (sometimes, however, due to foot drop); "steppage" refers to foot drop during swing, which leads to excessive hip flexion (but sometimes to sickling), and the old Trendelenburg's sign (91) refers to the sideways downward tilt of the pelvis on the swinging side, usually due to weakness of the hip abductor muscles of the supporting limb. All of these signs may coexist in the case of an escape limp. There may be cases, however, in which neither visual inspection nor the instrumental recording of kinematic variables provide sufficient, or any, information despite the presence of large dynamic asymmetries at the levels of the joints and the CoM. The patients presented in the figures above, suffering from hemiparesis, hip arthritis, above- and below-knee amputation and knee rotationplasty, respectively, all walked with a nearly normal appearance. Nevertheless, the motion of the CoM was very asymmetric; the body system "pole-vaulted" almost passively

FIGURE 8 | (A) The upper panel gives the average of the weight-normalized external power provided by muscles to sustain the increments of the total mechanical energy of the center of mass (CoM) (Etot, see Figure 5) during walking at different speeds (on the abscissa). Power was measured during the increments of Etot that occur during the double- and single-stance phases of the step (a and b, see Figure 5). Lower panel: the a/b ratio plotted as a function of speed. Power is invariably higher during the double-compared with the single-stance phase of the step. Taken from Tesio et al. (20), used with permission. (B) The ordinate gives the average power provided by muscles to sustain the a and the b increments (during double and single stance, respectively) of the total mechanical energy of the CoM (Etot, see also Note S5) during one entire stride in three below-knee (dots labeled b1–b3) and four above-knee (dots labeled a1–a4) amputees. The N and P labels indicate the single-stance phases of the stride on the normal and prosthetized lower limbs, respectively. The N/P and P/N labels indicate the double-stance phases of the stride (normal limb behind or prosthetic limb behind, respectively). During the N to P transition, much more power is generated when the unaffected limb is trailing, compared to when it is leading. Taken from Tesio et al. (66), used with permission.

FIGURE 9 | work performed by other ipsilateral lower-limb joints and segments (hip, knee, foot). Work values were obtained by computing the time integral of power during the push-off phase. Mean inter-subject power and work values and work standard deviations are depicted for walking at 1.4 m s <sup>−</sup><sup>1</sup> based on 6-degrees-of-freedom joint mechanics analysis (<sup>n</sup> <sup>=</sup> 9). Taken from Zelik and Adamczyk (59), used with permission. (B) Rate of energy change or power (work rate) estimates for an individual limb during human walking. The integrated area under each curve during push-off (light gray box) represents the magnitude of push-off work or energy change. (a) Ankle power (red line) overlaid on Total rate of energy change (gray line, due to the motion of and about the body's center of mass [CoM]). (b) The majority of total rate of energy change during push-off is attributable to CoM power (blue line, defined here as the rate of energy change due to push-off limb power production), and smaller contributions are from peripheral energy changes (due to segmental motion relative to the CoM; dashed cyan line). (c) The majority of total energy changes during push-off is also attributable to segmental changes from the push-off limb (green line). (d) The contribution of limb segmental energy changes (green line) to overall CoM changes (solid blue line) is shown here as a dashed blue line. During push-off, the majority of the limb power goes into this contribution, which in turn accounts for the majority of CoM power. Data depicted are inter-subject means at 1.4 m s−<sup>1</sup> (n = 9). Taken from Zelik and Adamczyk (59), used with permission.

on the affected limb, whereas extra work was required from the muscles during the "push" from the unaffected limb.

This agreement between segmental and systemic dynamic asymmetries is expected. However, from analyzing the results in the literature, at least three unexpected findings could be identified. The first is that the effects of most asymmetric diseases on the mechanics of the CoM may be surprisingly similar. The second is that when kinematic effects are minor, the asymmetries of the CoM motion may be large nonetheless. For instance, both in post-stroke (92, 93) and multiple sclerosis patients (94) a weak, if not absent, recruitment of the plantar flexors at push-off is observed. This allows sparing of the affected, or, more affected, lower limb. Given the force of the plantar flexors, their work output can be substantially abated with only a minor reduction of ankle rotation. The visual aspect of gait can be minimally affected while a strong asymmetry in Wext and R between the subsequent steps is generated. As a third unexpected finding, there is a tendency for external work and power asymmetries to compensate between the two subsequent steps, with the result that means along the whole stride and hence, per unit distance, can remain normal. This raises a critical challenge to rehabilitation aiming at restoring symmetry, as will be clarified later on.

# The CoM Motion in Impaired Children

In adolescents with unilateral cerebral palsy, Wext,m is only mildly increased (95). As for adults, symmetric impairments in children does not result in sharp differences in CoM motion compared with controls (96). The pendulum-like mechanism of translation of the CoM is less efficient (Rstep is about 30% lower) in children with cerebral palsy compared with their healthy counterparts. However, it must be noted that patients with cerebral palsy walked more slowly and with a higher step frequency, making it difficult to assess the isolated effect of spasticity. A normal Wext,step is obtained through a higher Rstep

in adolescents with mild obesity and Prader-Willi syndrome, as demonstrated by comparison of the Froude-normalized walking velocities of patients and controls (97). A paradoxical decrease of Wext,m has been identified in adolescents who were operated on for scoliosis (81), who were also found to walk with a higher step frequency, hence, a shorter step length. This factor is, in itself, a source of decreased mechanical efficiency (87).

# THE ENIGMA OF "WILLED" ASYMMETRIES IN CENTRAL PARESIS. IMPLICATIONS FOR REHABILITATION MEDICINE

The affected lower limb of hemiparetic patients produces less work than it could potentially deliver. The unaffected lower limb is overloaded, and the CoM transfer motion retains a normal cost per unit distance. Given this successful adaptation, which rehabilitation approach might ever be effective?

# A Hidden Power Can Be Revealed in Various Walking Paradigms

Asymmetric patients can increase muscle work output from the impaired lower limb in at least four different conditions: at increased speeds (98), during uphill walking (99), during splitbelt walking when the affected lower limb is forced to move faster than the unaffected lower limb (see below) (100), and by using crouch gait, also called bent hip-bent knee gait (67). In the latter, the increase in work involved the hip extensors more than the plantar flexors, while in the other conditions the greatest increase was invariably observed for the plantar flexors. More puzzling still is the fact that, whichever the velocity, the unaffected side also produced more work, thus maintaining the asymmetry. Asymmetry thus seems a goal rather than a necessity. This has been interpreted as a form of learned non-use (67), similar to that described for upper limbs after de-afferentation (101, 102) or stroke (103). Actually, this behavior looks automatic and unconscious, so that the adjective "acquired" seems preferable to "learned" (Baldissera, personal communication). A suggested treatment for upper-limb paresis is the forced use of the affected hand, which is induced by applying a constraint (e.g., a muffle glove) to the unaffected arm, which is termed constraint-induced movement therapy-CIMT (104). This approach is quite generalizable; its application to upper limb non-use was first described on monkeys in 1894 (101), and the analogous practice of eye patching to prevent amblyopia in strabismus is at least 2 centuries old (105). In the case of upper limb paresis, various forms of CIMT have been tested, associated with concurrent exercises and/or transcranial direct current stimulation (106). Despite the long-established research on upper limbs, the hypothesis of learned non-use of the lower limbs, and of tailored "forced-use" exercises, have only been proposed <2 decades ago (107). It is possible that the phenomenon has been overlooked for such a long time because unilateral impairments of the paretic lower limb are compatible with wide oscillations of this limb while walking. In the case of spasticity, the impaired limb may even appear overactive. Despite a visible swing, however, the affected lower limb always presents with a deficit in power production at the trailing ankle during late double stance, which lasts for about 10% of the step period (i.e., some 40 ms). The short duration and modest ankle rotations involved make it difficult to detect this dynamic impairment by visual inspection (however, see **Note S6**), although the recent introduction of force platforms has greatly enhanced detection.

In the authors' opinion, the learned/acquired non-use phenomenon is unspecific across asymmetric gait impairment, and independent of the underlying disease. This is indicated by the adaptation in both orthopedic and neurologic diseases, in which a normal value of Wext,m can be achieved. This may explain the clinical observation that lower voluntary activation is associated with hypotrophy of the affected lower limb in hemiparesis (108), and with the invariable hypotrophy after amputation, including stump and thigh hypotrophy after transtibial amputation (109), contrary to the hypertrophy that would be expected from compensatory overuse. In support of this hypothesis, there has been at least one report on knee rotationplasty where hypotrophy was accompanied by a decrease in the cortical motor map area on the contralesional hemisphere, while an increase of this area was observed in the homolesional hemisphere (110).

# Restoring Symmetry Through Exercise: A Conceptual Challenge for Rehabilitation in Walking Impairments

Is forced-use of the impaired lower limb both a feasible and useful therapeutic strategy in asymmetric gaits? Both in stroke and lower limb amputee patients the impaired limb retains the capacity to provide the power sufficient to achieve a symmetric gait (67, 69, 111): nevertheless, power asymmetry is preferred. By sparing the affected side, asymmetry may minimize local consequences of limited joint mobility, paresis, and sensory deficits causing weakness, pain, and/or instability. Perhaps unfortunately, this adaptation is also compatible with efficiency in terms of Wext (112) and total energy cost per unit distance (111). There seems to be no energetic motivation to restore symmetry. Power asymmetry, however, does have a clinical cost in the long run, making recovery of symmetry a desirable goal. Such clinical cost may occur in terms of impaired postural control at the "spared" lower limb, as demonstrated for standing (113) and gait initiation (114). Orthopedic troubles usually follow: a relevant example is provided by the frequent knee hyperextension (genu recurvatum) syndrome seen in various unilateral impairments, spanning from hemiparesis to cerebellar ataxia, to poliomyelitis, peripheral neuropathies, and myopathies. (115). Why are these long-term sequelae neglected? A pseudo-explanation is: because power (hence, kinematic) asymmetry can be achieved easily and sooner, thus fostering survival and therefore reproduction in a wild environment (105). Asymmetric walking may become soon compatible with normal velocity and cost of transport of the body system. Teleological explanations, including those resting on the Darwinian model of evolution, are not testable experimentally and therefore are pure conjectures yet, they are still part of the process of generating hypotheses (116). Whatever the reasons, reverting the tendency to retain asymmetry seems an appropriate goal, yet a very difficult challenge, for rehabilitation medicine. Original forms of exercise are needed, making at great disadvantage the asymmetric gait, perhaps under both the energetic and the balance perspectives of the CoM motion, compared to symmetric gait. Various approaches have been proposed based on orthotic hindrances (107), lower limb perturbations (117), and split-belt walking (118). The latter approach will be dealt with in the following paragraphs.

# AN EMERGING IDEA: THE SPLIT-BELT TREADMILL AS A SYMMETRIZING TOOL

# Acquired Non-use of the Impaired Side May Affect Walking

Sparing (or "non-use") of the impaired lower limb in the case of asymmetric walking has a deeply studied precedent in strabismus. A child that squints will commit visual acuity to the unaffected eye. Amblyopia will be much more likely than realignment. On the other hand, monocular vision will be more effective, from a behavioral standpoint, than diplopia. The analogy with the non-use observed in case of upper or lower limb impairments is quite strong (105). Maintaining vision of the squinting eye is based on the wellestablished approach of patching the unaffected eye, adopting a constraint-induced/forced-use rehabilitation paradigm (104). Split-belt walking represents a therapeutic analogy, which thus deserves discussion.

# Forcing the Use of the Impaired Lower Limb Through Split-Belt Walking

In split-belt walking, the subject walks on a force treadmill constructed with two parallel independent belts that can rotate at different velocities, hence the terms split- or dual-belt treadmill. Velocity ratios between the belts range from 1.5 to 4, depending on the study. Splitting causes a sudden escape limp in healthy subjects; within a few strides, the stance time and the "anterior" step length are reduced on the faster belt. The split-belt paradigm was introduced in a study on the capacities for early adaptation of the (supposed) spinal circuitries which underlie human walking (119). More recently, the cerebellum has been suggested to have a key role in consolidation of the treatment effect, specifically the after-effect (see below) (120). The idea of using this walking modality as a treatment for hemiparesis and Parkinson's disease has been proposed [for a review, see Reisman et al. (121)]. Step length was the main focus of most studies. The asymmetry in belt velocity may force symmetry between the length of subsequent steps, if the impaired (or the more impaired) limb which usually shows a longer anterior step—is placed on the faster belt. When belts return to an equal velocity, or ground walking is commenced, an after-effect develops: the original asymmetry is attenuated or even reversed. For this reason, in agreement with the error-augmentation pedagogic principle (122), most researchers consider the "paretic slow/anterior step longer" configuration to be more promising (123).

# A False Analogy: Split-Belt Walking in Healthy Subjects Is Not Equivalent to Real Pathologic Claudication

The literature on split-belt walking as a rehabilitation treatment is increasing rapidly, despite the fact that some basic issues remain controversial. Initial studies were based only on kinematic observations. The analogy between experimental claudication in healthy subjects and true pathological claudication during ground walking is only valid for the stance duration, which is shorter for the faster leg of healthy subjects on the treadmill and for the paretic leg of patients during ground walking. This analogy is not valid for step length (see **Video S1**). Differences are even more relevant with respect to dynamic asymmetries between the lower limbs. The "escaping" leg, when placed on the faster belt, is required to produce extra power to avoid being dragged backward by the faster belt. In contrast, during ground walking, the paretic "escaping" lower limb provides a much lower power than the unimpaired lower limb (69). The technicalities and questions concerning the therapeutic use of this walking paradigm are discussed in **Note S7**.

Additional insight may come by raising the question: does increased symmetry of the power of the plantar flexors translate into a more symmetric translation of the body system as a whole (i.e., of the CoM)? Despite the recent shift of interest from purely kinematic to dynamic analyses at the level of the lower limb, energy changes of the CoM during split-belt walking have not been directly addressed to date. Evaluation of this factor might clarify whether the primary aim of late adaptation is the minimization of Wext and/or of the plantar flexors' power and how these are achieved. Unfortunately, the time courses of Etot and R during split gait have not yet been described.

# A Visible CoM Used to Steer Gait Control

There is only one report that describes an experiment in which the subject is given visual feedback of the oscillations of the CoM in the frontal plane during treadmill walking (124). This was achieved by projecting the vertical motion of a reflective marker placed on the sacrum onto a screen in front of the subject who is facilitated by "knowledge of results" (125). This enhanced feedback stimulates the search for an implicit neuro-mechanical solution. After a short series of rehabilitation sessions, the symmetry of the vertical CoM motion in hemiplegic subjects, analyzed with the double-integration method, was improved, and the metabolic cost of walking decreased. However, these isolated results are still awaiting independent confirmation.

# DISCUSSION. THE COM MOTION FROM BENCH TO BEDSIDE: A BUMPY ROAD

Gait analysis is still striving to become a routine method of assessment of walking impairments. This goal is far from being achieved despite continuous educational efforts by dedicated scientific societies (e.g., SIAMOC in Italy, ESMAC at a European level).

# Instrumental Gait Analysis: Why Not Yet a Routine Clinical Approach?

The main obstacles to clinical applications of gait analysis, in the authors' opinions, are neither the cost nor the technical skills required. Many other diagnostic technologies common in medicine are complicated and expensive (e.g., those applied to imaging and neurophysiology). The key point is that, in most clinical conditions, gait analysis seems to add little to a skilled visual assessment, with respect to conventional clinical decision making (126). By greater force, the analysis of the CoM, not yet considered as a standard component of gait analysis, seems doomed to remain confined in the research domain. This is a particular case of a more general problem. Energy, work and power (three related physical variables) look abstract concepts to most clinicians, who are much more familiar with tangible muscle forces and, of course, with kinematics and visible electromyographic tracings. Abstract, however, here does equal neither imaginary nor non-existent. Forces alone do not explain the movement, which implies displacement in a given time, hence power (the product of force time shortening velocity). The neural sequences of muscle activation follow the logic of optimizing the power requirements and are adapted to mechanical requirements. Walking as a movement makes no exception. Its peculiarity lies in the skillful interplay between muscle power and gravity, culminating in the extremely advantageous pendulum-like mechanism, which is the result of a long biological evolution. It is clear, however, that any changes in the engines (the muscles), and/or in their neural control, and/or in the machinery (the joints) do have an impact on the motion of the body system as a whole. The many degrees of freedom of the human body, however, makes the type and amount of this impact hard to predict. In some cases, adaptation will aim at minimizing energy expenditure, hence the overall efficiency of the pendulum mechanism per unit distance. Sometimes the sparing of affected segments will be privileged, and in other cases, the best trade-off will be looked for between segment and system needs. In any case, the sources of kinematic impairments (will they be at a single joint or at the system level in claudication) can only be found in the changes of the patterns of power production.

# Neglecting the CoM Contributes to Make Gait Analysis Marginal

Having said that, the issues of the low clinical success of standard gait analysis based on segmental motions and the neglect toward CoM motion, look related. In the authors' opinion, the available technologies for the analysis of segmental kinematic, electromyographic, and dynamic events are informative and within reach, both technically and financially, of most in- and out-patient services. An international collaboration on the topic is very intense. The analysis of the CoM motion does not raise any insurmountable difficulties. When realized through the indirect kinematic method, studying the CoM motion only requires a dedicated software routine to be added to the software package for kinematic analysis provided by the manufacturers. The weak points remain the clinical questions, not the technical answers. The bulk of data emerging from a contemporary gait analysis

should not be explored within a descriptive logic, which can turn into a fishing expedition with little chance of success. Rather, a priori hypotheses should be conceived, focused questions should be raised, and the analysis should be run according to a confirmatory (or, dis-confirmatory) logic, as one would for any diagnostic procedure. Unfortunately, walking in general and the CoM motion in particular are rarely considered as specific, trans-disciplinary chapters of medical education. By contrast, understanding the CoM motion might help to interpret the focal alterations. First, the CoM cannot be directly observed, but signs of alterations of its path can be visually detected (see **Note S6**). The clinical questions to be raised to gait analysis thus become (a) which are the segmental motions related to this "system" impairments? and (b) will any correction of these focal impairments through exercise, drugs, orthoses, or surgery lead to a more physiologic motion of the body system? Second, extremely simple and affordable techniques, such as those based on wearable accelerometers, might provide a rough quantitative analysis of the CoM motion, sufficient to help monitoring progress during a rehabilitation program in any clinical environment.

# CONCLUSIONS AND PERSPECTIVES

Complete understanding of human walking will be facilitated if the issue is considered within the wider perspective of legged terrestrial locomotion.

# Legged Locomotion in Biology: Hints to the Clinical Interpretation of Walking Impairments (and of Their Resistance to Rehabilitation)

All walking animals fight against both gravity and ground shearing forces. In this context, it appears that the critical function of locomotion is made possible by mechanical strategies that are widespread within the animal realm. Oscillating legs are compatible with the nerve and vascular supply to the legs, and thus with life, in a way that rotational appendages, in the manner of wheels, would not be. Unlike wheeled locomotion, however, this solution entails costly repositioning of the limbs at every step, and costly upward and forward accelerations of the CoM at each cycle. The latter are the main causes of muscle work at low and intermediate walking velocities. It is not surprising that the inverted pendulum compromise, which reduces the otherwise exaggerated energetic cost of transport of the body system, is observed in the majority of walking species, despite their huge variety of sizes, shapes, locomotory structures and nervous systems. Among apes, human walking is special, as far as it is four times more efficient than either the quadrupedal or bipedal walking of Chimpanzees (127). This partly explains the capacity of the modern Homo genus to spread across the whole world during the "great human expansion" outside of Africa, from about 60,000 to 12,000 years B.C. (128). Given the vital role of walking and the extreme efficiency of the human pendulum-like mechanism, it is not surprising that humans tenaciously keep their CoM motion mechanism substantially unchanged from 4 years of age to senility, despite a trend toward joint degeneration and muscle weakening. Again unsurprisingly, the efficiency of CoM motion per unit distance, and often, per single step, is spontaneously preserved in the majority of human motor impairments: this explains their resistance to therapeutic attempts aiming to modify an otherwise successful and overwhelming adaptation strategy.

# Neural Gait Control Is Always Targeting the Invisible CoM

These considerations seem to support the conclusion that the motion of the CoM itself, not less than the motions of anatomical segments, is a constant target of neural control and should also be a primary target of clinical diagnostics. Segmental alterations should be suspected to be both causes and effects of alterations of CoM motion. If one lower limb has an altered motion, the rest of the body system will strive to adapt in order to optimize the balance between sparing the affected segment and maintaining the highest possible efficiency and safety of the pendular motion of the CoM. Relating the CoM motion to the segmental perspectives will allow original impairments to be distinguished from adaptive alterations.

# Future Directions, Between Adaptation and Restoration

Alterations of the 3D CoM pathway may be a sensitive index of pathology yet, there are few specifics with respect to segmental alterations: diagnosis will thus require clinical diagnostic skills. Therapeutic options will increase if clinicians consider walking from the perspective of the CoM motion, too. In symmetric pathological gaits, increasing velocity and step length (however obtained) may be rewarded by a more efficient pendulum mechanism. In asymmetric gaits the CoM perspective is particularly enlightening. On one hand, it seems clear that any attempt to restore symmetry by treating the affected side will be at high risk for failure, as the successful adaptation of CoM motion will be jeopardized. On the other hand, asymmetry leads to troubles in the long run, so that treatments that aim to rebalance the work between the affected and unaffected lower limbs are justified. These treatments need a conceptual advance in rehabilitation medicine and should be driven by monitoring the effects on CoM motion. The idea of forced-use through spilt-belt treadmills looks a promising paradigm of a systemand-segments approach for the diagnosis and treatment of asymmetric walking impairments. Other approaches that directly target the CoM, cited above, might be effective as well. As a final speculation, going perhaps beyond the scope of the present review, the potential role of non-invasive electric or magnetic brain stimulation can be alluded to. A variant (and perhaps one of the neurophysiologic bases) of the "acquired non-use" model is represented by the model of interhemispheric inhibition. Evidence is growing that the preference for an asymmetric locomotion is reflected by the asymmetry in excitability between the two hemispheric motor cortices, both in orthopedic and neurological impairments (110, 129, 130). No matter whether brain functional asymmetries are effects or causes of walking

mechanical asymmetries, combining neurophysiological and behavioral treatments might make sense in the near future.

# AUTHOR CONTRIBUTIONS

LT and VR discussed and contributed to writing the article. LT designed and wrote the content of the manuscript. All authors read and approved the final manuscript.

# FUNDING

This research was funded by the Ministry of Health, Ricerca Corrente 2011, GAITCORR project.

# REFERENCES


# ACKNOWLEDGMENTS

The authors are indebted to Fausto Baldissera for helpful comments and suggestions during writing, and Giovanni A. Cavagna for insightful comments on the manuscript.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fneur. 2019.00999/full#supplementary-material

Video S1 | Video showing a healthy subject walking on a split belt treadmill. Further details are provided in the video.


balance control in young and older adults. J Biomech. (2016) 49:1264–7. doi: 10.1016/j.jbiomech.2016.03.011


speed in obese subjects. Med Sci Sports Exerc. (2009) 41:426–34. doi: 10.1249/MSS.0b013e31818606e7


Neurorehabil Neural Repair. (2009) 23:735–44. doi: 10.1177/1545968309 332880


practice: general requirements, methods and appropriateness. Results of an Italian consensus conference. Gait Posture. (2017) 58:252–60. doi: 10.1016/j.gaitpost.2017.08.003


**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 © 2019 Tesio and Rota. 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.

# Motor Adaptation in Parkinson's Disease During Prolonged Walking in Response to Corrective Acoustic Messages

Mattia Corzani<sup>1</sup> \*, Alberto Ferrari<sup>1</sup> , Pieter Ginis<sup>2</sup> , Alice Nieuwboer<sup>2</sup> and Lorenzo Chiari<sup>1</sup>

<sup>1</sup> Department of Electrical, Electronic, and Information Engineering, University of Bologna, Bologna, Italy, <sup>2</sup> Department of Rehabilitation Sciences, Neurorehabilitation Research Group, KU Leuven, Leuven, Belgium

Wearable sensing technology is a new way to deliver corrective feedback. It is highly applicable to gait rehabilitation for persons with Parkinson's disease (PD) because feedback potentially engages spared neural function. Our study characterizes participants' motor adaptation to feedback signaling a deviation from their normal cadence during prolonged walking, providing insight into possible novel therapeutic devices for gait re-training. Twenty-eight persons with PD (15 with freezing, 13 without) and 13 age-matched healthy elderly (HE) walked for two 30-minute sessions. When their cadence varied, they heard either intelligent cueing (IntCue: bouts of ten beats indicating normal cadence) or intelligent feedback (IntFB: verbal instruction to increase or decrease cadence). We created a model that compares the effectiveness of the two conditions by quantifying the number of steps needed to return to the target cadence for every deviation. The model fits the short-term motor responses to the external step inputs (collected with wearable sensors). We found some significant difference in motor adaptation among groups and subgroups for the IntCue condition only. Both conditions were instead able to identify different types of responders among persons with PD, although showing opposite trends in their speed of adaptation. Increasing rather than decreasing the pace appeared to be more difficult for both groups. In fact, under IntFB the PD group required about seven steps to increase their cadence, whereas they only needed about three steps to decrease their cadence. However, it is important to note that this difference was not significant; perhaps future work could include more participants and/or more sessions, increasing the total number of deviations for analysis. Notably, a significant negative correlation, r = −0.57 (p-value = 0.008), was found between speed of adaptation and number of deviations during IntCue, but not during IntFB, suggesting that, for people who struggle with gait, such as those with PD, verbal instructions rather than metronome beats might be more effective at restoring normal cadence. Clinicians and biofeedback developers designing novel therapeutic devices could apply our findings to determine the optimal timing for corrective feedback, optimizing gait rehabilitation while minimizing the risk of cue-dependency.

Keywords: Parkinson's disease, motor adaptation, continuous gait, auditory cue, verbal feedback, wearable sensors

Edited by:

Paolo Cavallari, University of Milan, Italy

#### Reviewed by:

Gammon Earhart, Washington University in St. Louis, United States Ioannis Ugo Isaias, Julius Maximilian University of Würzburg, Germany

> \*Correspondence: Mattia Corzani mattia.corzani@unibo.it

Received: 11 June 2019 Accepted: 10 September 2019 Published: 24 September 2019

#### Citation:

Corzani M, Ferrari A, Ginis P, Nieuwboer A and Chiari L (2019) Motor Adaptation in Parkinson's Disease During Prolonged Walking in Response to Corrective Acoustic Messages. Front. Aging Neurosci. 11:265. doi: 10.3389/fnagi.2019.00265

# INTRODUCTION

fnagi-11-00265 September 23, 2019 Time: 15:54 # 2

Parkinson's disease (PD) is a neurodegenerative disorder predominantly characterized by the depletion of dopamine and dopaminergic neurons in the basal ganglia (BG) (Mazzoni et al., 2012). The disease affects different neural networks and neurotransmitters, leading to impaired ability to learn and express automatic actions, such as walking (Redgrave et al., 2010). The use of external sensory cues (e.g., auditory, visual) to reinforce attention toward the task (Lee et al., 2012) is an effective gait-rehabilitation strategy for persons with PD; the cues stimulate the executive voluntary component of action (Morris, 2006; Morris et al., 2008; Ferrazzoli et al., 2018) by activating the attentional-executive motor control system and bypassing the dysfunctional, habitual, sensorimotor BG network (Morris, 2006; Morris et al., 2008, 2010; Redgrave et al., 2010; Shine et al., 2014; Tard et al., 2015; Arnulfo et al., 2018; Pozzi et al., 2019). This strategy helps people with PD improve gait consistency and rhythmicity. In the past, auditory cueing during gait has typically been provided continuously in an open loop (regardless of gait performance). However, continuous cueing may result in cue-dependency and habituation on external stimuli (Nieuwboer et al., 2009; Spildooren et al., 2012; Vercruysse et al., 2012; Bohnen and Jahn, 2013).

One of the most innovative developments in the quantitative assessment and management of PD symptoms is the use of wearable technologies during gait (Sánchez-Ferro and Maetzler, 2016), which are able to provide customized cueing: stimuli are triggered when gait deviates from normal, thus providing patients with immediate feedback on their performance. These closedloop stimuli [audio (Ginis et al., 2016; Ginis et al., 2017a,b), visual (Ahn et al., 2017; Chong et al., 2011), audio-visual, (Espay et al., 2010) or proprioceptive (Mancini et al., 2018)] are known as intelligent inputs (Ginis et al., 2017a,b). In contrast to openloop systems, in closed-loop systems the external information does not necessarily become part of the participants' movement representation (as explained by the "guidance hypothesis"), thus possibly decreasing the development of cue-dependency (Nieuwboer et al., 2009). Wearable systems also permit data collection in a more naturalistic environment (Espay et al., 2010; Ginis et al., 2016).

Two previous studies (Ginis et al., 2017a,b) compared the effects of intelligent auditory cueing (IntCue) and intelligent verbal feedback (IntFB) on gait as alternatives to traditional open-loop continuous cueing (ConCue) (see **Figure 1**). Those studies showed that both IntCue and IntFB conditions were at least as effective as ConCue for optimizing gait in PD. For example, the first study showed that IntFB was most effective at maintaining normal cadence at the end of a 30-minute-long gait exercise, although it also increased the gait variability (deviations from the target pace) in persons with PD compared to healthy controls. Furthermore, during IntCue, the number of deviations was actually smaller than during the no-input condition in PD (Ginis et al., 2017b).

In the second one, persons with and without freezing of gait (FOG+ and FOG-) were compared (Ginis et al., 2017a). The results show that the FOG+ group benefits less from intelligent inputs than the FOG- group, probably due to more affected motor and cognitive functions (Ginis et al., 2017a). In addition, the former had significantly more gait deviations than the latter during IntCue and IntFB conditions, but not when continuously cued. Although these findings suggest that ConCue was more effective in supporting prolonged gait in the FOG+ group, the majority of these persons favored the IntFB condition (Ginis et al., 2017a).

These two works (Ginis et al., 2017a,b) adopted a macro approach to analyze the effect of wearable sensors and external inputs on continuous gait in PD: they did not quantify the individual motor responses to the corrective messages. However, since many factors are at play during a prolonged walking trial, such as fatigue and learning, a micro-analysis is more appropriate, because it quantifies the motor adaptations during the participants' immediate response to the IntFB and IntCue conditions. Thus, the cadence of the subjects' first steps following each corrective acoustic message can be quantified.

The aim of this study is to characterize motor adaptation in response to corrective acoustic messages during prolonged walking in order to gain insight on how to better design novel therapeutic devices for gait re-training in PD (FOG- and FOG+). To this end we propose a new model for fitting the short-term motor responses to external inputs (collected with wearable sensors). Using this model we determine the number of steps needed to adapt gait pattern following corrective acoustic messages. We investigated adaptation speed during IntFB and IntCue conditions for the following groups: healthy elderly (HEg), persons with PD (PDg), and PD subgroups with (FOG+g) and without (FOG-g) freezing of gait. We hypothesized that IntFB would lead to a more effective adaptation than IntCue, due to its verbal nature. In fact, the IntFB has an explicit nature with a clear direction of change to adapt the gait, compared to IntCue, which requires some processing time to elaborate the direction of adaptation leading to a delay in the motor response. Furthermore, because persons with PD struggle to maintain normal gait (Hausdorff, 2009), we expected that slowing down would be easier than speeding up. For both conditions, they would adapt more quickly when they were directed to slow back down to their reference cadence (because they had speed up) than when they were directed to speed up (because they had slowed down).

To assess the effect of the clinical characteristics of the participants, we investigated the relationship that links motor adaptation with their disease severity and their cognitive status. Furthermore, to match our micro-analysis with the macro results of previous work (Ginis et al., 2017a,b), we also evaluated the relationship between motor adaptation speed and the number of deviations of each subject, to determine if persons who struggled more to walk consistently were slower to adapt to the corrective stimuli.

# MATERIALS AND METHODS

The present study consists of a sub-analysis of another study which compared persons with PD to age-matched healthy

of the reference walk. (B) Schematic representations of the different intelligent inputs used in the protocol. Green, blue and red bars represent the periods during which the cadence is good, deviates below the threshold, or deviates above the threshold, respectively. NoInfo: no external information was given during the entire walk; ConCue: during the entire walk, participants received the auditory rhythm set at the mean cadence of the reference walk; IntCue: participants received the same auditory rhythm as in ConCue, but only for ten beats and only when the cadence deviated from the reference cadence; IntFB: participants received verbal feedback to "Increase the rhythm" or "Decrease the rhythm" when the cadence was more than 5% slower or faster (respectively) than the reference cadence.

subjects on several gait characteristics throughout 30 min of walking during four different auditory input conditions (Ginis et al., 2017b). These sections briefly describe the participants and protocol of the previous study before presenting the motor adaptation model and statistical analysis.

# Original Study – Participants, Protocol, and Materials

Twenty-eight persons with PD were recruited from the Movement Disorders clinic of the University Hospitals Leuven based on the following inclusion criteria: (Mazzoni et al., 2012) idiopathic PD, diagnosed according to the United Kingdom Brain Bank criteria; (Redgrave et al., 2010) Hoehn and Yahr stage I–III; and (Lee et al., 2012) stable PD medication for the past month and anticipated for the following 2 months. Exclusion criteria were: (Mazzoni et al., 2012) cognitive deficits (Mini Mental State Examination <24/30); (Redgrave et al., 2010) subjectively unable to walk unassisted for 30 min; (Lee et al., 2012) fluctuating response to levodopa; (Morris, 2006) musculoskeletal or neurological conditions other than PD affecting gait; and (Morris et al., 2008) severe hearing problems precluding headphone use for auditory information. Participants were categorized into freezers, FOG+ (n = 15), and non-freezers, FOG- (n = 13), based on a score of one or higher on the New Freezing of Gait Questionnaire (NFOG-Q). All persons with PD were tested in their subjective ON-medication state, an average of 1 h after medication intake. It is important to note that no freezing episodes occurred during the study.

Thirteen age-matched HE were recruited from a database of voluntary study participants. The study design and protocol were approved by the local Ethics Committee of the KU Leuven and performed in accordance with the requirements of the International Council of Harmonization (Declaration of Helsinki, 1964). Written informed consent was obtained from each participant prior to the experiment. Over a period of 6 weeks, participants performed four 30-minute walks, with at least one week between walks, around an elliptical track measuring 24 m by 9 m. Prior to each 30-minute walk, the reference walk consisted of a fixed duration of a 1-min walk

at a comfortable pace was recorded, to obtain the reference cadence. Participants started the 30-minute walk randomly in a clockwise or anti-clockwise direction, after which the starting direction was kept identical per person over the four sessions. After 15 min of walking, participants changed their walking direction (by crossing the trajectory diagonally) to counteract possible effects of disease dominance. In a randomized order, participants experienced one of the following conditions for the entire 30-minute walk: (Mazzoni et al., 2012) continuous cueing (ConCue); (Redgrave et al., 2010) intelligent cueing (IntCue); (Lee et al., 2012) intelligent feedback (IntFB); and (Morris, 2006) no information (NoInfo). Cueing and feedback were provided by an adaptive wearable system (Casamassima et al., 2014) through headphones (Sennheiser RS160, Sennheiser, Germany). During IntCue, for every deviation, participants received an auditory rhythm, consisting of ten beats at the reference cadence whether it was a DOWN event (cadence over the threshold) or an UP event (cadence below the threshold). The threshold which triggered the stimulus was set as a variation of more than 5% from the reference cadence, calculated from the mean cadence of five consecutive steps. During IntFB, participants received a verbal instruction to "increase rhythm" or "decrease rhythm" based on the same criteria as during IntCue. The values for the IntCue and IntFB settings, as well as the duration of the 1-min reference walk, were based on user acceptability, determined during pilot testing prior to the study. All the conditions are shown in **Figure 1**. All walks were performed in the same hall at the same time and day of the week to minimize the effects of time and PD medication. Demographic information and clinical test results were collected: in particular, the Movement Disorders Society Unified Parkinson's Disease Rating Scale—Motor Part (MDS-UPDRS III) (Goetz et al., 2008), Scale for Outcomes in Parkinson's Disease-Cognition (SCOPA-Cog) (Marinus et al., 2003), and Montreal Cognitive Assessment (MoCA) (Gill et al., 2008). All the clinical tests were collected during the ON phase of medication before the start of the walking task to avoid potential influence of fatigue. Clinical tests were evenly distributed over the different assessment days.

Participants wore two foot-mounted inertial measurement units (IMUs) attached to the tops of their shoes using Velcro straps. The IMUs (EXLs1, EXEL srl, Italy) contained a tri-axial accelerometer, gyroscope, and magnetometer, sampled at 100 Hz and wirelessly streaming via Bluetooth to a computer. A custom Matlab (Mathworks Inc., United States) software application, using the algorithms currently implemented in the commercially available system Gait Tutor (mHealth Technologies, IT), processed the signals in real time during each 30-minute walk. The algorithm (validated for PD (Ferrari et al., 2016; Ginis et al., 2016) and described elsewhere (Casamassima et al., 2014)) computed cadence from the raw IMU data and registered any deviations from the pre-recorded reference cadence.

# Motor Adaptation Model

This section describes the analysis we performed for IntCue and IntFB conditions in order to evaluate motor adaptation after intelligent inputs.

Cadence for all conditions was calculated by combining the data from both feet. The system in the original study only retained the average cadence for every five steps of each foot, while to obtain better sensitivity we recreated the original cadence for every trial.

Next, adaptation (in response to both UP and DOWN events) was quantified by fitting a single-term exponential model (Eq. 1) to the cadence of the ten steps following a deviation. In Eq. (1), y is the fitted cadence expressed as a percentage of the difference with respect to the reference cadence, k is the exponential decay/growth rate [step−<sup>1</sup> ], x is the number of steps after intelligent input (from 0 to 9), and M is the under-/overthreshold value [%].

$$
\chi = \pm M e^{-k\chi} \tag{1}
$$

The primary outcome was the exponential decay/growth rate k estimated for each UP or DOWN deviation (for all participants). **Figure 2** shows a representation of the mathematical model in response to both DOWN and UP events. To better illustrate the role of k, each graph in **Figure 2** reports three responses with different k values.

A higher value of k corresponds to a faster adaptation, as is clear in **Figure 2**. Next, we evaluated the relative step constant (intrinsic to an exponential decay/growth model), τ = 1 / k, which (as shown in the graph) intercepts the curve at a value for M of 63%. This characteristic parameter is defined in our study as the number of steps required to reduce M sufficiently that participants are in the correct range. (Note that this percentage supposes a reasonable M). Therefore, τ represents what we can call a refractory period: the number of steps needed to bring the cadence back within the reference range following verbal/acoustic feedback, during which providing a new corrective message may have no effect.

# Statistical Analysis

A preliminary qualitative analysis compared the average responses to corrective stimuli between PDg, HEg, and FOG-g, FOG+g. We used our fitting model to quantify motor adaptation in terms of k, the exponential decay/growth rate during all the corrective acoustic messages received by the participants. We calculated the absolute median values and the relative interquartile range among all groups and subgroups.

Next, we used paired non-parametric statistics (Wilcoxon Signed Rank test) to evaluate the condition effect (IntCue vs. IntFB) and the task effect (UP event vs. DOWN event), analyzing differences in the average k rate of each subject only within the PDg, due to the small sample available. However, the resulting average differences do not say anything about specific motor responses (Rispens et al., 2015). On the other hand, situations where people show a high level of motor adaptation reflect their best possible performance and are thus of specific therapeutic interest. To identify the best performances, we investigated the 90th-percentile values of each subject in addition to the average k rate. Clearly, a paired test requires subjects who had corrective messages in both conditions or in both tasks (depending on the analysis).

Unpaired non-parametric statistics (Mann-Whitney U tests) were used to examine differences in the value of k between groups HEg and PDg (group effect) and subgroups FOG+g and FOGg (subgroup effect). In addition, we performed an exploratory analysis to assess whether subjects who had only UP events responded differently than those who had both UP and DOWN events. We assumed that for those subjects who tend to slow down, it may be more difficult to increase their rhythm in response to corrective acoustic messages—compared to subjects who tend to both slow down and speed up.

The relations between adaptation speed and clinical data were explored by correlating the participants' scores on SCOPA-Cog and MoCA (cognitive aspect) and MDS-UPDRS III (disease severity) with their median k rate using Spearman rank correlation coefficients. The median k rate was also correlated with the number of deviations for each subject. Matlab (Mathworks Inc., United States) was used for all statistical analyses, with α = 0.05.

# RESULTS

# Demographics

For simplicity, the demographic analysis (available from previous studies) is reported in **Table 1** (Ginis et al., 2017a,b). PDg and HEg were well matched for age, body height, body weight, cognitive ability (MoCA), total self-reported daily physical activity (LAPAQ Total), and self-reported daily walking time (LAPAQ Walking). The PD group had significantly lower cognitive scores (SCOPA-Cog). Freezers (FOG+) and nonfreezers (FOG-) were well matched for age, body morphology (weight, height, and leg length), self-reported daily walking, and total daily activity time (LAPAQ), as well as for Hoehn and Yahr stage. The FOG+g had a significantly longer disease duration, lower cognitive scores (MoCA), and more reported gait difficulties on the 12-item gait scale (12G) than the FOG-g.

# Qualitative Adaptation Plots

In this preliminary analysis we qualitatively highlighted the average behavior of the participants following the corrective acoustic messages. We compared the average cadence between HEg and PDg and subgroups FOG+g and FOG-g, from five steps before the deviation until 20 steps after. **Figure 3** shows the average original cadence responses to all corrective acoustic messages received by HEg and PDg during both conditions.

**Figure 4** reports the same analysis as **Figure 3**, comparing FOG- and FOG+ subgroups in both conditions.

# Number of Deviations

To improve interpretability and readability of our analysis, in **Table 2** we reported the total number of deviations,

TABLE 1 | Results are reported as mean (standard deviation) in the case of parametric statistics and as median (quartile 1– quartile 3) in the case of non-parametric statistics.


Bold numbers indicate significant differences between groups and subgroups. MDS-UPDRS III Movement Disorders Society Unified Parkinson Disease Rating Scale-Motor Part, LEDD levodopa equivalent daily dosage, MoCA Montreal Cognitive Assessment, LAPAQ LASA Physical Activity Questionnaire, 12G 12 item gait scale, SCOPA-Cog Scale for Outcomes in Parkinson's Disease-Cognition. All the clinical tests were collected during the ON phase. <sup>a</sup> Chi-squared statistics. <sup>b</sup> Non-parametric statistics were applied.

already presented and discussed in the original study (Ginis et al., 2017a,b).

# Condition Effect: IntCue vs. IntFB

As can be seen in **Table 3**, although the fastest median k rate (i.e., larger median k rate) occurred during IntFB, the differences in k rates for the two types of conditions were not significant for PDg. It is important to note that, to perform a paired analysis, we had to exclude some subjects: only 12 of the 28 PD subjects received at least one UP message during the two conditions and only four subjects received at least one DOWN message.

Next, a subgroup evaluation was carried out (see **Table 3**). No statistical analysis was performed for HEg, FOG+g or FOG-g, due to the small number of paired samples, which would increase the possibility of a type II error. In line with the PDg findings, this analysis suggested a slightly faster adaptation during IntFB than IntCue looking at the absolute median value.

# Group Effect: PDg vs. HEg; SubGroup Effect: FOG+g vs. FOG-g

Differences in motor adaptation between the groups (HEg vs. PDg) and subgroups (FOG+g vs. FOG-g) were analyzed during IntFB (**Tables 4A-1, A-2**) and IntCue conditions (**Tables 4B-1, B-2**). During the IntCue condition, PDg had a significantly faster adaptation than HEg in response to the UP event (p-value < 0.000) and FOG+g adapted significantly faster than FOG-g in response to the DOWN event (p-value = 0.006).

# Task Effect: UP Event vs. DOWN Event

Although the fastest median k rate occurred following a DOWN event, the differences in k rates for the two types of event were not significant for PDg (**Tables 4A-1, B-1**). This finding also holds true within the HEg, FOG+g and FOG-g (**Table 4**), but here no statistical analysis was performed because of the small number of paired samples. Moreover, only eight persons with PD experienced at least one of each event type during the IntFB condition and only five during the IntCue condition.

We performed an exploratory analysis to assess whether members of PDg who had only UP events responded differently than those who experienced both UP and DOWN events. The results indicate a different trend for each condition. During IntCue, those who only had UP events adapted faster (p-value = 0.039). In contrast, during IntFB those who experienced both DOWN and UP events adapted faster than those who experienced only UP events (p < 0.000) (**Table 5**).

# Correlation Analysis

The participants' SCOPA-Cog, MoCA, and MDS-UPDRS III scores did not correlate significantly with the median k rate during either condition or either task. However, as can be seen in **Table 6**, the median k rate correlated significantly with the number of deviations during the IntCue UP event (r = −0.57; pvalue = 0.008), indicating that those with the slowest adaptations had the most deviations throughout the 30-minute walk. No significant correlations were observed for the IntCue DOWN events or any of the IntFB events.

TABLE 2 | Total number of deviations received by the groups in response to UP (A) and DOWN (B) messages.


PD 62 88 FOG+ 54 81 FOG- 8 7 IntCue, intelligent auditory cueing; IntFB, intelligent verbal feedback; UP event,

cadence under the threshold – Increase rhythm; DOWN event, cadence over the threshold – Decrease rhythm; PD, people with Parkinson's disease; HE, healthy elderly; FOG+, people with Parkinson's disease with freezing of gait; FOG-, people with Parkinson's disease without freezing of gait.

# Refractory Period During IntFB in PDg

Focusing on the PD group, the relative step constant τ = 1 / k, with the k values reported in **Table 3**, indicates the refractory period of about seven steps for the UP event and three steps for the DOWN event during the IntFB condition. Higher values of τ are observed during the IntCue condition: about eight steps for the UP event and about five steps for the DOWN event. In **Table 7** we report the values of τ .

# DISCUSSION

This study investigated the effects of intelligent auditory cueing (IntCue) and verbal feedback (IntFB) on motor adaptation in HE, PD, and PD subgroups (with and without FOG.) We introduced an innovative model to quantify motor adaptation speed following the two different acoustic messages. Thanks to our novel adaptation model, which applies a decay/growth exponential model to gait biofeedback for the first time, we can define the refractory period as the value of the relative step constant τ. This value indicates the steps needed to bring the cadence within the reference range following verbal/acoustic feedback.

Our results from the IntFB condition indicate a refractory period of about seven steps for the UP event and about three steps for the DOWN event for PD subjects. We found a similar (only slightly higher) refractory period for the IntCue condition. Clinicians and biofeedback developers designing novel therapeutic devices could apply our findings to determine the optimal timing for corrective feedback, optimizing gait rehabilitation while minimizing the risk of cue-dependency. In this way, the system could provide optimal corrective feedback in maintaining the proper gait pattern.

TABLE 3 | Condition effect. Index of adaptation k is reported as median (quartile 1– quartile 3) among all groups in response to UP (A) and DOWN (B) messages.



IntCue, intelligent auditory cueing; IntFB, intelligent verbal feedback; UP event, cadence under the threshold – Increase rhythm; DOWN event, cadence over the threshold – Decrease rhythm; #deviators participants who have at least one deviation; PD, people with Parkinson's disease; HE, healthy elderly; FOG+, people with Parkinson's disease with freezing of gait; FOG-, people with Parkinson's disease without freezing of gait; Mean Condition effect, paired analysis using the average k rate of each subject; 90th Condition effect, paired analysis using the 90th percentile value k rate of each subject. n, number of participants included in the paired-analysis.

We hypothesized that the verbal and explicit nature of IntFB could speed up the motor response (Taylor et al., 2014), compared to IntCue which is more implicit and may thus requires some processing time to elaborate the direction of adaptation. In contrast to what assumed, our analysis could not detect different adaptations between IntFB and IntCue conditions. Nevertheless, in line with our expectations, the absolute median values of the decay/growth rate k for IntFB are larger than for IntCue in all groups and subgroups. This is consistent with the qualitative indications of the adaptation plots and is in line with the visual exploration performed in a previous study (Ginis et al., 2017a). Furthermore, when looking at the adaptation plots (**Figure 3**) it can be observed that there is an overshooting in IntFB only, which may be explained by the reference cadence indicated only during the IntCue condition.

We also expected that increasing the pace to the reference level would be a more difficult task than decreasing it. However, this trend is not confirmed within PDg by the statistical analysis.

TABLE 4 | Task and group effect. Index of adaptation k is reported as median (quartile 1– quartile 3) among all groups for (A-1, A-2) IntFB condition, (B-1, B-2) for IntCue condition.


IntCue intelligent auditory cueing; IntFB intelligent verbal feedback; #deviators participants who have at least one deviation; PD people with Parkinson's disease; HE healthy elderly; FOG+ people with Parkinson's disease with freezing of gait; FOG- people with Parkinson's disease without freezing of gait; Mean Task effect paired analysis using the average k rate of each subject; 90th Task effect paired analysis using the 90th percentile value k rate of each subject. Group effect unpaired analysis between HE and PD groups. SubGroup effect unpaired analysis between FOG+ and FOG- groups. Bold numbers indicate significant differences in the task (UP vs. DOWN event), groups (PD vs. HE) and subgroups effect (FOG+ vs. FOG-). n, number of participants included in the paired-analysis.

This lack of significance could be due to the small number of deviations recorded.

The group effect analyses yielded two results during IntCue: PDg had faster adaptation than HEg in response to UP events and FOG+g adapted faster than FOG-g in response to DOWN events. These results could be unexpected because HEg and FOG-g have better gait stability (fewer deviations) than PDg and FOG+g, respectively, as indicated in previous work (Ginis et al., 2017a,b). However, this is in line with previous work that showed a higher reliance on external input in PD compared to healthy subjects (Petzinger et al., 2013).

There were no differences in motor adaptation between the groups (HEg vs. PDg) or subgroups (FOG+g vs. FOG-g) during IntFB. In this regard, contradictory results can be found in literature. Roemmich et al. (2014) found that PD and HE adapted similarly, during the first strides after exposure to a split-belt gait pattern. On the other hand, Mohammadi et al. (2015) showed that FOG+ have more difficulties than FOG- and HE to adapt their gait to a split-belt treadmill over a short time period.

Our exploratory analysis revealed that during 30 min of walking, subjects who had only UP events adapted more slowly than those who had both UP and DOWN events during IntFB condition. This result is in agreement with our hypothesis: for those subjects who tend to slow down, it may be more difficult to increase their rhythm in response to corrective acoustic messages—compared to subjects who tend to both slow down and speed up. Instead, for IntCue we had the opposite trend, maybe because the metronome cues were more difficult to understand for those who received both up and down messages.



In the first column the group who received only UP events, while in the second column the group who received both UP and DOWN inputs during the 30 min of walking. Bold numbers indicate significant differences between groups.

TABLE 6 | Correlation between the number of messages received (#deviations) and the median k rate of each subject during IntFB and IntCue conditions in response to UP-event and DOWN-event.


Spearman rank correlation coefficient r and relative p-value are shown in the table. Bold numbers indicate significant correlation.

TABLE 7 | Refractory period. The relative step constant τ = 1 / k [step] in the PD group.

fnagi-11-00265 September 23, 2019 Time: 15:54 # 10


τ represents the refractory period in response to UP and DOWN events and indicates the number of steps needed to adapt gait pattern and bring the cadence within the reference range following verbal/acoustic feedback.

No significant correlations were found between our adaptation speed results and cognitive ability. In this protocol the subjects had to walk maintaining a determined cadence, without turning or avoiding obstacles. This steady state walking, with a reduced cognitive load, may be the reason for the lack of correlations found. In addition, the fact that all participants had an MMSE ≥ 24/30, conform the study's inclusion criteria, suggests that further study is needed in a cohort with a wider cognitive spectrum. On the other hand, the original study found a correlation between gait stability (number of deviations) and the MoCA scale (with the same dataset) (Ginis et al., 2017a).

The relationship between our micro-analysis and the macro approach adopted in previous works (Ginis et al., 2017a,b) was evaluated by correlating the adaptation speed with the number of deviations. We found a negative trend only, during IntCue in response to UP events. During IntCue in response to DOWN events, as well as during IntFB, no significant correlations were observed. This finding seems to indicate that motor adaptation might be more effective during IntFB for all subjects, including those who require a lot of assistance through the intelligent messages. In fact, the previous work reported that subjects preferred verbal feedback (Ginis et al., 2017a). The negative trend found could also be explained by the slower adaptation leading to an increased number of deviations. In fact, participants may not yet be within the threshold values, triggering the feedback again. As expected, in response to DOWN events we did not find correlations in any conditions, suggesting that decreasing cadence is a relatively simple task which can be done fairly quickly.

A critical re-analysis of our results might suggest that an interesting solution could be the use of a combined-cue system, i.e., verbal feedback to increase or decrease cadence followed by the rhythmic cues to specify the target rate. This combined solution could trigger adaptation similar to the IntFB system, because of its explicit nature. On the other hand, due to the target rate indicated by the cueing, the combined-cue system may avoid the overshooting observed in part during the IntFB condition in the preliminary qualitative analysis. In any case, the joint use of IntFB and IntCue conditions could increase the overload of sensory and cognitive functions.

FUTURE WORK

Effective tools for PD rehabilitation should allocate attention appropriately and lighten cognitive load (Stefan et al., 2005). The use of multisensory stimuli improves the learning process (Lehmann and Murray, 2005; Shams and Seitz, 2008), thanks to a reduced cognitive load and easier storage in shortterm memory (von Kriegstein and Giraud, 2006; Schmitz et al., 2013). A multisensory approach also enhances perceptual processing (Shams and Seitz, 2008), known to be reduced in PD subjects with FOG (Davidsdottir et al., 2005). Mezzarobba et al. (2018) demonstrated the effectiveness of this approach in a study which used video and synthesized sounds to help PD subjects with FOG relearn gait movements and reduce freezing episodes.

Following this principle, it might be useful to add another sensory input to the IntFB condition. Proprioceptive feedback (such as vibrational stimuli) which require little or no cognitive processing or attention (Peterson and Smulders, 2015), might greatly improve motor adaptation in response to intelligent inputs. In fact, proprioceptive stimuli, in a closed loop system (Mancini et al., 2018), are already commonly used for gait rehabilitation in PD.

Future work also needs to address the long-term effect of gait rehabilitation on adaptation speed. It could be important to quantify the dynamics of adaptation during a trial and any possible motor-learning effect [i.e., the formation of a new motor pattern, in response to intelligent inputs, that occurs via long-term practice (Bastian, 2008)]. It would also be useful to evaluate motor adaptation through a prolonged, homebased training period, which would provide naturalistic data. Finally, it would be interesting to explore the use of different parameters (stride length, gait speed) instead of cadence to trigger the feedback.

It should be noted that further exploration of our model would benefit from ensuring that sufficient data are obtained to validate the qualitative and quantitative findings.

# 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 The local Ethics Committee of the KU Leuven and it was performed in accordance with the requirements of the International Council of Harmonization (Declaration of Helsinki, 1964). The patients/participants provided their written informed consent to participate in this study.

# AUTHOR CONTRIBUTIONS

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

# REFERENCES

fnagi-11-00265 September 23, 2019 Time: 15:54 # 11


in people with Parkinson's disease. Sci. Rep. 8:12773. doi: 10.1038/s41598-018- 31156-4



**Conflict of Interest:** AF (one of the founders) and LC (shareholder) have a significant financial interest in mHealth Technologies s.r.l., a company that may have a commercial interest in the results of this research.

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 Corzani, Ferrari, Ginis, Nieuwboer and Chiari. 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.

# Acute Effects of Whole-Body Vibration on the Postural Organization of Gait Initiation in Young Adults and Elderly: A Randomized Sham Intervention Study

Arnaud Delafontaine1,2,3 \*, Thomas Vialleron1,2, Matthieu Fischer 1,2, Guillaume Laffaye1,2 , Laurence Chèze<sup>4</sup> , Romain Artico1,2,3, François Genêt <sup>5</sup> , Paul Christian Fourcade1,2 and Eric Yiou1,2

<sup>1</sup> CIAMS, Univ. Paris-Sud., Université Paris-Saclay, Orsay, France, <sup>2</sup> CIAMS, Université d'Orléans, Orléans, France, <sup>3</sup> ENKRE, Saint-Maurice, France, <sup>4</sup> LBMC, Université de Lyon, Villeurbanne, France, <sup>5</sup> UMR End:icap équipe 3, UFR des Sciences de la Santé Simone Veil, UVSQ, Montigny le Bretonneux, France

#### Edited by:

Giovanni Abbruzzese, University of Genoa, Italy

# Reviewed by:

Manuela Galli, Politecnico di Milano, Italy Alberto Ranavolo, Istituto Nazionale per l'Assicurazione Contro gli Infortuni sul Lavoro (INAIL), Italy

> \*Correspondence: Arnaud Delafontaine arnaud.delafontaine@u-psud.fr

#### Specialty section:

This article was submitted to Neurorehabilitation, a section of the journal Frontiers in Neurology

Received: 21 June 2019 Accepted: 09 September 2019 Published: 24 September 2019

#### Citation:

Delafontaine A, Vialleron T, Fischer M, Laffaye G, Chèze L, Artico R, Genêt F, Fourcade PC and Yiou E (2019) Acute Effects of Whole-Body Vibration on the Postural Organization of Gait Initiation in Young Adults and Elderly: A Randomized Sham Intervention Study. Front. Neurol. 10:1023. doi: 10.3389/fneur.2019.01023 Whole-body vibration (WBV) is a training method that exposes the entire body to mechanical oscillations while standing erect or seated on a vibrating platform. This method is nowadays commonly used by clinicians to improve specific motor outcomes in various sub-populations such as elderly and young healthy adults, either sedentary or well-trained. The present study investigated the effects of acute WBV application on the balance control mechanisms during gait initiation (GI) in young healthy adults and elderly. It was hypothesized that the balance control mechanisms at play during gait initiation may compensate each other in case one or several components are perturbed following acute WBV application, so that postural stability and/or motor performance can be maintained or even improved. It is further hypothesized that this capacity of adaptation is altered with aging. Main results showed that the effects of acute WBV application on the GI postural organization depended on the age of participants. Specifically, a positive effect was observed on dynamic stability in the young adults, while no effect was observed in the elderly. An increased stance leg stiffness was also observed in the young adults only. The positive effect of WBV on dynamic stability was ascribed to an increase in the mediolateral amplitude of "anticipatory postural adjustments" following WBV application, which did overcompensate the potentially destabilizing effect of the increased stance leg stiffness. In elderly, no such anticipatory (nor corrective) postural adaptation was required since acute WBV application did not elicit any change in the stance leg stiffness. These results suggest that WBV application may be effective in improving dynamic stability but at the condition that participants are able to develop adaptive changes in balance control mechanisms, as did the young adults. Globally, these findings are thus in agreement with the hypothesis that balance control mechanisms are interdependent within the postural system, i.e., they may compensate each other in case one component (here the leg stiffness) is perturbed.

Keywords: whole-body vibration (WBV), gait initiation, elderly, young adults, anticipatory postural adjustment (APA)

# INTRODUCTION

Whole-body vibration (WBV) is a training method that exposes the entire body to mechanical oscillations while standing erect or seated on a vibrating platform. This method is nowadays commonly used by clinicians to improve specific motor outcomes in various sub-populations such as elderly and young healthy adults, either sedentary or well-trained. Specifically, it has been shown that WBV application might be beneficial to numerous lower limbs motor outcomes, such as endurance, power, strength, neuromuscular activity, flexibility, and knee stability [for young healthy subjects, see for example (1–7); for elderly, see for example (8–13), for review see (14)].

Balance and locomotion might also be improved by WBV application, as quantified (i) with center of pressure (COP) measures in static upright condition [for young healthy adults, see for example (3, 15, 16); for elderly, see for example (17–20)], and (ii) with commonly-used clinical tests such as the singleleg-stance test, the Tinetti test, the Timed-Up-and-Go test, the Berg Balance Scale, and the Limits of stability. However, it is well-recognized that these classical clinical tests are subjective, show ceiling effects, and are usually not responsive enough to measure small progresses or deteriorations in a subject's ability to balance (21, 22). Mancini and Horak (22) further stressed that the greatest limitation of this clinical approach to rating balance is that it cannot specify what type of balance problem a subject suffers in order to direct a treatment. Upright balance maintenance during locomotor tasks is indeed a highly complex task since it necessitates the coordination between many complementary postural mechanisms, including dynamic postural phenomena before the swing foot clearance (23–25), the regulation of stance leg stiffness (26), the swing foot placement on the support surface (27), the braking of center of mass fall (28– 30) etc. The investigation of these mechanisms [cf. description below; see also (31) for a recent review] requires an in-depth biomechanical analysis with force plate recordings. Such analysis is not provided in the classical clinical tests reported above. In fact, the classical clinical test, as Timed Up and Go test for example, could be limited to determined individual risk of falls in elderly (32). Furthermore, Topper et al. (33) show that falls happened most frequently in elderly subjects scoring poorly on clinical tests emphasizing transfer of quasi-static to dynamic situations.

Gait initiation is a functional task that is classically used in the literature as an experimental paradigm to investigate how the central nervous system (CNS) controls balance in dynamical conditions in both healthy adults [e.g., (31–35) for recent articles] and pathological patients [e.g., (36–40) for recent articles]. Gait initiation appears to be the major biomechanical model, complementary to the classical clinical test, in order to predict postural disorders in elderly fallers and non-fallers (41– 44). Therefore, it seems highly appropriate for an investigation on the effects of WBV on dynamic balance. Gait initiation corresponds to the transient period between static erect posture and steady state walking (23–25). It can be divided into two successive phases: a postural phase that precedes the onset of the swing heel off (the so called "anticipatory postural adjustments", APA) followed by an execution phase that ends at the time of swing foot contact (FC) with the ground.

During APA, the COP is shifted backward, creating the initial forward propulsive forces that are necessary to reach the intended center of mass (COM) velocity and step length (23, 25, 45). At the same time, the COP is shifted laterally toward the swing leg, which acts to propel the COM toward the stance leg (23, 26, 35, 46, 47). This anticipatory mediolateral postural dynamic acts to attenuate the COM fall toward the swing leg during the execution phase under gravity (26, 48, 49). It has therefore a stabilizing function. In case mediolateral APA are not sufficient to ensure stabilization, a strategy of lateral swing foot placement that enlarges the base of support size has been documented (27, 46, 47, 50, 51). The development of APA and the lateral swing foot placement during locomotor tasks are known to be dependent on somatosensory inputs from the lower limbs (52– 55). The perturbation of these inputs by the applications of WBV might thus alter these two mechanisms and increase instability and/or reduce motor performance.

In addition to mediolateral APA and lateral swing foot placement, recent modeling study showed that increasing the stance leg stiffness along the mediolateral direction would result in a larger COM fall toward the swing leg side, thus leading to an increased mediolateral instability. The application of WBV may trigger a "tonic vibration reflex" (TVR) via the activation of muscle spindles of the lower limbs (56, 57). Although controversial (8, 58–60), this TVR might potentially be responsible for an increase in lower limbs stiffness and may thus have a negative impact on stability.

Globally, this brief literature review suggests that WBV may potentially perturb the balance control mechanisms involved in gait initiation and may therefore induce instability. This statement thus markedly contrasts with the results of the abovereported clinical tests, which rather suggest that WBV may be effective in improving relatively basic balance ability and locomotion. Thus, the purpose of this study is to clarify the effects of acute WBV application on the balance control mechanisms during gait initiation in young healthy adults and elderly. It is hypothesized that the balance control mechanisms at play during gait initiation may compensate each other in case one or several components are perturbed following acute WBV application, so that postural stability and/or motor performance can be maintained (or even improved). It is further hypothesized that this capacity of adaptation is altered with aging, with a consequent increased instability following WBV application.

# METHODS

# Participants

The study is a randomized investigation that includes 81 healthy adults (i.e., 41 young and 40 elderly). The non-probability convenience method is used, i.e., participants were randomly assigned to one of the four following treatment groups using the "envelope method" (see below): (1) 20 young adults (11 men and 9 women, mean age 25.3 ± 3.5 years, height 1.71 ± 0.05 m and body-mass 71.4 ± 5.3 kg) were assigned to the Young WBV training group (YWBV); (2) 21 matched young adults were assigned to the Young Sham Group (YSG) (11 men and 10 women, mean age 26.6 ± 4.2 years, height 1.74 ± 0.04 m and body-mass 72.1 ± 4.9 kg); (3) 20 elderly (9 men and 11 women, mean age 83.5 ± 2.8 years, height 1.69 ± 0.06 m and body-mass 70.4 ± 4.3 kg) were assigned to the Elderly WBV training group (EWBV) and (4) 20 matched elderly (9 men and 11 women, mean age 84.2 ± 3.7 years, height 1.71 ± 0.07 m and body-mass 71.7 ± 3.9 kg) were assigned to the Elderly Sham Group (ESG; see **Table 1**). In the "envelope method," each participant chooses a closed envelope in which his/her treatment group is indicated in a code form. The participant does not have access to this code but the experimenters do (single blinded condition).

The two WBV groups (young adults and elderly) received a single WBV training and the two sham groups received a placebo treatment (cf. below for description). As stated above, participants were blinded to their allocation group. Participants had no recent history of trauma, known metabolic disorders, inflammatory infectious arthropathies, bone malignancies or neurological disease. In addition, participants were all naïve about WBV treatment. They all gave written consent after having been informed of the nature and purpose of the experiment which was approved by local ethics committees from the CIAMS Research Unit, Equipe d'Accueil (EA) 4532. The study complied with the standards established by the Declaration of Helsinki and was assigned the following trial registration number: 2017-002539-40.

# Experimental Task and Conditions

Experiments took place in the Biomechanics laboratory of the Paris-Saclay University located within the Kremlin-Bicêtre Hospital (Paris, France). Physical conditions (room temperature and time of the day) were common to all treatment groups, and the same before/after the WBV treatment.

Participants initially stood barefoot in a natural upright posture on a force plate (0.9 × 1.80 m, AMTI, Watertown, USA) embedded at the beginning of a 6 m walkway track. The feet

TABLE 1 | Anthropometric data of subjects in both WBV and sham groups (young adults and elderly).


Reported values are means ± standard deviation.

NS, non-significant difference.

were shoulder-width apart, with the arms alongside the trunk and the gaze directed forward to a small target at eye level (2 cm diameter, 5 m distant). The locations of the heel and big toe of each foot in the initial posture were marked with sections of adhesive tape placed on the force plate and were used as a visual reference on which participants positioned themselves under the supervision of the experimenters. From this initial posture, participants performed two series of 10 gait initiation trials: one series just before, and a second series immediately after a specific treatment (pre- and post-treatment conditions, respectively) depending on their group (WBV or sham). In both pre- and post-treatment conditions, participants initiated gait at a spontaneous velocity and at their own initiative following an auditory signal delivered by the experimenter, and then continued walking straight until the end of the track. Participants initiated gait with their preferred leg in all trials. One blank trial was provided in the pre-manipulation condition (not recorded) to ensure that the instructions were well-understood by the participant and that the material was operational. The rest time was ∼10 s between trials.

# WBV and Placebo Treatment

For both the WBV group and the sham group, participants stood upright on a vibration platform (referee Power Fitness Double D <sup>R</sup> ) with knees flexed ∼30◦ and barefoot. During the treatment, knee angle was monitored on-line by both the subject him/herself and the experimenters thanks to a visual feedback provided by the signal of a monoaxial electrogoniometer (Biometrics <sup>R</sup> ) fixed on the right leg. Participants were instructed to maintain the signal within a 25–35 range degrees.

For both young adults and elderly, the WBV application included four 45 s bouts of vibration, intercalated by 1 min of seated rest without vibration between bouts (61–63). Vibrations were delivered by the vibration platform at a frequency of 50 Hz [this frequency is known to enhance locomotor task and intralimb coordination (61)] with a vertical displacement of 2 mm and a 2 g acceleration.

The sham group received a treatment that followed exactly the same procedure as the WBV group except that no "real" vibrations were applied to participants. In order to maintain the most realistic placebo conditions as possible, participants were instructed by an experimenter that they could not physically feel the vibrations because they were too small to be sensed.

# Mechanical Model

The same mechanical model as in a previous study was used to determine the stance leg stiffness along the mediolateral direction during the execution phase of gait initiation [for details on this model and the related equations of motion, see (26)]. In brief, the human body was modeled as a single conic inverted pendulum which rotates around a fixed point corresponding to the stance ankle. It was considered that the COM falls laterally under the influence of two forces: the gravity force P = mg (where m is the mass of the whole body, and g is the gravitational acceleration) and an elastic restoring mediolateral force T (in Newtons) that reflects active muscular control of the movement (64, 65), with T = k|yM| where k is the stance leg stiffness (Newtons/meter) and |yM| is the absolute value of the mediolateral COM shift (in meters), systematically oriented toward the swing leg (positive values) during the execution phase. The initial position and velocity of the cone corresponded to the position and velocity of the participant's COM at toe off.

# Raw Data

Force-plate data were low-pass filtered using a second order Butterworth filter with a 10 Hz (30) cut-off frequency. The mediolateral (yp) and anteroposterior (xp) coordinates of the COP (in meters) were computed from force-plate data as follows:

$$\mathbf{y\_p} = \frac{\mathbf{M}\mathbf{x} + \mathbf{F}\mathbf{y} \times \mathbf{dz}}{\mathbf{F}\mathbf{z}} \tag{1}$$

$$\mathbf{x\_{p}} = \frac{-\mathbf{M}\mathbf{y} + \mathbf{F}\mathbf{x} \times \mathbf{dz}}{\mathbf{F}\mathbf{z}}\tag{2}$$

where Mx and My are the moments (in Newtons.meters) around the anteroposterior and mediolateral axes (at the forceplate origin), respectively; Fx, Fy, and Fz (in Newtons) are the mediolateral, anteroposterior, and vertical ground reaction forces, respectively; and dz (in meters) is the distance between the surface of the force-plate and its origin which is located below this surface.

Instantaneous COM acceleration along the anteroposterior and mediolateral axes was determined from the ground reaction forces according to the Newton's second law. COM velocity and displacement were computed by successive numerical integrations of COM acceleration using the rectangles method (66) and integration constants equal to zero, i.e., initial velocity and displacement null (23).

### Experimental Dependent Variables

The biomechanical traces and the representations of the main experimental variables obtained for one representative young subject of the WBV group initiating gait (one trial) in the pretreatment condition and post-treatment condition are presented in the **Figure 2**. The following instants were determined from the biomechanical traces: gait initiation onset (t0), swing heeloff, swing toe-off, swing foot-contact, and rear foot-off. These instants were determined from force-plate data (23, 67, 68). Two t<sup>0</sup> times were estimated, one for the mediolateral axis and one for the anteroposterior axis. The t<sup>0</sup> times corresponded to the instants when the mediolateral or anteroposterior COP trace deviated 2.5 standard deviations from its baseline value.

The mediolateral and anteroposterior COM position in the initial upright static posture were estimated by averaging the COP position during the 250 ms period preceding the "all set" signal. Gait initiation was divided into APA (from t<sup>0</sup> to heel-off), swing foot-lift (from heel-off to toe-off), and execution phase (from toe-off to foot-contact). The duration of APA along the mediolateral and anteroposterior axes were computed separately, because the t<sup>0</sup> times for these two axes do not necessarily occur simultaneously. The amplitude of APA was characterized by the peaks of the backward and mediolateral COP shift obtained during the APA time window. COM velocity and displacement along the mediolateral and anteroposterior axes were quantified at heel-off and foot-contact.

The spatiotemporal features of gait performance during the execution phase included: the swing phase duration, the anteroposterior COM velocity at foot-contact (progression velocity), and the step length. Step length corresponds to the difference between the most backward COP position during gait initiation and the COP position at the time of rear foot-off (69). Rear foot off time was marked with the mediolateral COP trace.

The spatiotemporal features of postural stability during the execution phase included: the step width, the ML COM position at FC, the ML COM velocity at FC and the margin of stability (MOS). An adaptation of the MOS introduced by Hof et al. (70) was used to quantify the mediolateral dynamic stability at foot-contact (thereafter referred to as "stability"). The MOS corresponds to the difference between the mediolateral boundary of the base of support (BOSymax, in meters) and the mediolateral position of the "extrapolated COM" at swing foot-contact (yCOMFC, in meters). Thus:

$$\text{MOS} = \text{BOSymax} -\_{\text{Y}} \text{COMFC} \tag{3}$$

Participants systematically landed on the force-plate with the swing heel first, then the toe. Under such foot landing strategy, it is known that BOSymax could be estimated with the mediolateral COP position at the time of rear foot-off (69, 71). Based on the study by Hof et al. (70) and the results from our previous studies (26, 27, 69, 71, 72) the mediolateral position of the extrapolated COM at foot-contact (yCOMFC) was calculated as follows:

$$\text{y}\_{\text{\textdegree}}\text{\textdegree\text{COMP}\text{\textdegree}} = \text{y}\text{M}\_{\text{\textdegree C}} + \frac{\text{y}^{\prime}\text{M}\_{\text{\textdegree C}}}{\alpha\_0} \tag{4}$$

where yMFC and y'MFC are respectively the mediolateral COM position (in meters) and velocity at foot-contact (in meters/second), and ω<sup>0</sup> is the eigenfrequency of the body (Hz), modeled as an inverted pendulum and calculated as follows:

$$
\omega\_0 = \sqrt{\frac{\mathcal{g}}{l}} \tag{5}
$$

where g = 9.81 m/s<sup>2</sup> is the gravitational acceleration and l is the length of the inverted pendulum, which in this study corresponded to 57.5% of body height (73). Mediolateral dynamic stability at foot-contact is preserved on the condition that <sup>y</sup>COMFC is within BOSymax, which corresponds to a positive MOS. A negative MOS indicates a mediolateral instability and implies that a corrective action is required to maintain balance. Hence, variables related to the mediolateral stability included the MOS and its components, i.e., step width, mediolateral COM velocity and position at swing foot-contact. Step width corresponds to the difference between the most lateral position of the mediolateral COP trace obtained during the first plateau of the trace and the mediolateral COP position at the time of rear foot-off (69).

# Statistics

Mean values and standard deviations were calculated for each variable in each condition. The normality of data was checked using the Kolmogorov–Smirnov test and the homogeneity of variances was checked using the Bartlett test. Repeated measures (RM) ANOVAs with the treatment (pre-treatment condition vs. post-treatment condition) as within subject factor and the group (sham group vs. WBV group) as between subjects' factor were used on each experimental variable. Because this research focuses on the effect of WBV application on gait initiation in young and elderly adults, merely than on the effects of age on gait initiation, the RM ANOVAs were conducted for young and elderly adults taken separately. When necessary, a significant outcome was followed-up with the Tukey post-hoc test. Finally, linear correlation between variables was assessed with the Pearson coefficient r. The threshold of significance was set at p < 0.05.

# RESULTS

# Biomechanical Traces

The time course of the biomechanical traces was globally similar in the two groups (young adults and elderly) and in the different treatment conditions (WBV and sham). The traces obtained in a representative young adult subject, before and after the WBV treatment, are reported in the **Figure 1**.

As can be seen from this figure, swing heel off was systematically preceded by postural dynamics that corresponded to APA. During these APA, the COP displacement reached a peak value in a backward direction (see the negative variation of the xP trace) and toward the swing leg side (negative variation of the yP trace), while the COM velocity was directed forwards (positive variation of the x'M trace) and toward the stance leg side (positive variation of the y'M trace). The mediolateral COM velocity trace reached a first peak value toward the stance leg side at around heel off. This trace, then fell toward the swing leg side and a second peak value toward this side was reached a few milliseconds after foot contact. The anteroposterior COM velocity increased progressively until it reached a peak value a few milliseconds after swing foot contact. The mediolateral COM shift trace was bell-shaped and reached a peak value toward the stance leg side at the beginning of the swing phase, while the anteroposterior COM was continuously shifted forward (see **Figure 2**). Differences across the conditions are reported in the paragraphs below.

# Initial Posture, Anticipatory Postural Adjustments, and Foot-Lift Phase

In both the young adults and the elderly, no baseline differences were found between the WBV and sham groups (p > 0.05) for all the variables (see **Table 2**).

In both the young adults and the elderly, statistical analysis showed that there was no significant effect of the group (sham group vs. WBV group) or treatment (pre-treatment condition vs. post-treatment condition) nor any significant interaction between these two factors on the initial COM position along the mediolateral or anteroposterior direction.

In the elderly, there was also no significant effect of the group or treatment nor any significant interaction between these two factors on the spatiotemporal parameters of APA along

the mediolateral or anteroposterior direction (APA duration, peak COP shift, COM shift, and velocity at heel-off), nor on the swing foot-lift phase duration. In contrast, in the young adults, there was a significant effect of the treatment [F(1, 39) = 17.94, p < 0.001] on the mediolateral peak of anticipatory COP shift, along with a significant group X treatment interaction [F(1, 39) = 8.83, p < 0.01]. Similarly, there was a significant effect of the treatment [F(1, 39) = 8.89, p < 0.01], along with a significant group X treatment interaction [F(1, 39) = 4.38, p = 0.04] on the mediolateral COM velocity at heel-off. Post-hoc analysis further revealed that these two interactions could be ascribed to the result that none of these two APA related-parameters were significantly different between the pre and post-treatment condition for the sham group, while both were significantly larger in the post-treatment condition than in the pre-treatment condition for the WBV group (p < 0.001 for the mediolateral peak of anticipatory COP shift

and p < 0.05 for the mediolateral COM velocity at heel-off; cf. **Table 2**).

# Motor Performance

In both the young adults and the elderly, no baseline differences were found between the WBV and sham groups (p > 0.05) for all the variables (see **Table 3**).

In both the young adults and the elderly, statistical analysis showed that there was no significant main effect of the group or treatment nor significant interaction between these two factors on the execution phase duration, step length, and progression velocity at foot contact.

## Stability

In both the young adults and the elderly, no baseline differences were found between the WBV and sham groups (p > 0.05) for all the variables (cf. **Table 3**).

In the elderly, statistical analysis showed that there was no significant effect of the group or treatment nor any significant interaction between these two factors on the MOS and related components (i.e., step width, mediolateral COM position, and velocity at foot-contact). In contrast, in the young adults, there was a significant effect of the treatment, along with a significant group X treatment interaction on the following variables: MOS [F(1, 39) = 5.14, p = 0.03; F(1, 39) = 5.62, p = 0.02, respectively], mediolateral COM position [F(1, 39) = 7.32, p = 0.01, F(1, 39) = 5.19, p = 0.03, respectively], and mediolateral COM velocity at foot-contact [F(1, 39) = 10.14, p < 0.01, F(1, 39) = 4.51, p = 0.04; respectively]. Post-hoc analysis further revealed that these interactions could be ascribed to the result that none of these variables were significantly different between the pre and the post-treatment condition for

TABLE 2 | Comparison of APA and foot lift parameters between the pre- and post-treatment condition in the WBV and sham group (young adults and elderly).


Reported values are means ± standard deviation. APA, anticipatory postural adjustments; COM, center of mass; COP, center of pressure; AP, anteroposterior; ML, mediolateral; HO, heel off; NS, non-significant interaction. \*, \*\*, \*\*\*Significant difference between the pre and post-treatment condition as revealed with the Tukey post-hoc test, with P < 0.05, P < 0.01, and P < 0.001, respectively.

TABLE 3 | Comparison of motor performance, stability and stiffness parameters between the pre- and post-treatment condition in the WBV and sham group (young adults and elderly).


Reported values are means ± standard deviation. COM, center of mass; COP, center of pressure; AP, anteroposterior; ML, mediolateral; FC, foot contact; NS, non-significant interaction. \*, \*\*, \*\*\*Significant difference between the pre and post-treatment condition as revealed with the Tukey post-hoc test, with P < 0.05, P < 0.01, and P < 0.001, respectively.

the sham group, while they were statistically different for the WBV group. Specifically, the mediolateral COM position (p < 0.01) and the mediolateral COM velocity at foot-contact (p < 0.01) were both significantly smaller in the post-treatment condition than in the pre-treatment condition for the WBV group, while the MOS was significantly larger (p < 0.05; cf. **Table 3**).

# Stance Leg Stiffness

In both the young adults and the elderly, no baseline difference was found between the WBV and sham groups (p > 0.05; cf. **Table 3**). In the elderly, statistical analysis showed that there was no significant effect of the group or treatment nor any significant interaction between these two factors on the leg stiffness. In contrast, in the young adults, there was a significant effect of the treatment [F(1, 39) = 4.15, p < 0.05) along with a significant group X treatment interaction [F(1, 39) = 4.19, p = 0.04] on this variable. Post-hoc analysis further revealed that this interaction could be ascribed to the result that leg stiffness was not significantly different between the pre and the posttreatment condition for the sham group, while it was statistically larger for the WBV group (p < 0.05). In addition, results showed that, in the WBV group post-treatment condition, the stance leg stiffness in young adults was significantly correlated with the amplitude of mediolateral APA (r = 0.61, p < 0.01), i.e., the larger the leg stiffness, the larger the COP shift (**Figure 2**).

# DISCUSSION

In the paragraph that follows, the coherence of biomechanical data and their implication in terms of advanced postural control knowledge in young adults and elderly are discussed.

# Effect of Acute WBV Application on APA in Young Adults

In young adults, results showed that the amplitude of the mediolateral APA, expressed in terms of peak COP shift, was increased in the WBV group post-treatment. The difference between the pre- and the post-treatment condition reached up to 16.1%, which may be considered as clinically significant. The result that no difference was observed between the pretreatment and the post-treatment condition in the sham group discards a placebo or a practice effect due to the repetition of the task. The increased mediolateral APA could neither be ascribed to an increase in the progression velocity (27) or to an increase in the duration of the execution phase of gait initiation (26, 46, 47, 72). Results indeed showed that there was no significant effect of the WBV application on these two parameters. In contrast to the mediolateral APA amplitude, the mediolateral APA duration did not change in the WBV group post-treatment. Participants were therefore able to generate larger APA within the same duration, i.e., the APA efficacy was increased post-treatment. In line with this statement, results showed that the mediolateral COM velocity at the time of swing foot-off, which is known to depend on the spatiotemporal parameters of APA (27, 72), was larger following the WBV application. As this initial COM velocity is directed toward the stance leg, it acts to minimize the COM fall in the opposite direction (i.e., toward the swing leg) during the execution phase under the gravity effect (26, 49). The stabilizing function of APA was thus improved by acute WBV application.

The spatiotemporal parameters of APA are also known to be dependent on the somatosensory inputs arising from the lower limbs and trunk (52–55, 74–78). One can therefore wonder whether the positive effect on APA amplitude reported in this study could be ascribed to a change in the somatosensory inputs induced by the WBV application. While the current literature on healthy subjects repeatedly reported a detrimental effect of acute WBV application on the cutaneous lower limbs sensitivity (16, 79–81), it seems that acute WBV application does not influence lower limbs and trunk proprioception (17, 80–83); note that a detrimental effect on trunk proprioception was reported by Li et al. (84). Now, plantar sensitivity and limb proprioception were not investigated in the present study. However, an alteration of these inputs in the WBV group posttreatment (if any) may likely not be responsible for the changes in the mediolateral APA amplitude reported in the present study. For example, Lin and Yang (85) showed that the amplitude and the duration of mediolateral APA for gait initiation were both reduced (and not increased or unchanged, respectively, as in the present study) following plantar desensitization induced by cold water immersion. A decrease in the mediolateral APA for stepping was also reported when the ankle muscles acting in the mediolateral direction were vibrated to induce alteration of the proprioceptive afferent inflow from Ia fibers (53). In the present study, participants were exposed to WBV but, because of the standing posture with the feet on the vibrating force-plate, the ankle muscles were strongly submitted to the vibrations as in this latter study. The authors proposed that the proprioceptive information induced by vibration and afferent inflow related to body movement exaggerated sense of movement at the ankle. This biased perception led to a decrease in the APA spatiotemporal features. Similarly, a decrease in the postural response during balance recovery following cutaneous and muscular deafferentation experimentally induced by foot anesthesia and leg ischemia, respectively, was reported by Thoumie and Do (77). Finally, proprioceptive perturbation induced by experimental pain applied to the tibialis anterior (86) or acute fatigue of the same ankle muscles (78) were both shown to induce a decrease in the APA amplitude for gait initiation.

Globally, the above mentioned studies suggest that the increased mediolateral APA observed in the present research could probably not be ascribed to an (eventual) alteration of the somatosensory inputs induced by acute WBV application. Although not measured in these latter studies, it is possible that a major difference between these studies and the present one lays in the changes in the leg stiffness post-treatment. The stance leg stiffness was indeed increased by 16.8% during the execution phase of gait initiation following the WBV application, which may not be the case in the above mentioned studies due to the different treatments. In the following paragraph, it is argued that this change in the mechanical feature of the stance leg might be responsible for the increased mediolateral APA.

# Effect of Acute WBV Application on Stance Leg Stiffness and Relationship With APA and Stability in Young Adults

As for APA, the increased stance leg stiffness observed in the WBV group post-treatment could not be ascribed to a placebo or a practice effect since no change in this parameter could be detected in the sham group pre-treatment vs. posttreatment. This increased leg stiffness might possibly be related to a "TVR" induced by the activation of leg muscle spindles sensitive to vibrations (56, 57). The effect of WBV on limb stiffness is however controversial in the literature. In human, no effect of acute WBV on stiffness has been reported on patellar tendon (87), hamstring, quadriceps (88), and triceps surae (89) and in lower limbs during hoping (90). In contrast, WBV has been shown to increase area and stiffness of the flexor carpi ulnaris tendon in rats (91) and in the Achille tendon of older women (8). Colson and Petit (90) concluded that further studies should be undertaken to ascertain the effectiveness of WBV on lower limbs stiffness. It is possible that the effect of acute WBV application on muscle stiffness depends on the task to be done and the muscle tested (and particularly its content in spindles), but also on the mechanical features of the vibrations (amplitude, frequency, duration).

Whatever its origin, previous modeling study showed that increasing stance leg stiffness has the potential to induce instability during gait initiation (26). The current results showed that acute WBV application induced two opposite effects on postural stabilization: a stabilizing effect via an increase in the mediolateral APA amplitude, and a destabilizing effect via an increase in stance leg stiffness. Results showed that the combined action of these two opposite effects resulted in an increased stability during gait initiation, as assessed with the larger mediolateral margin of stability post-treatment. This increase reached 21.4%, which can be considered as clinically relevant. This finding suggests that the negative effect of WBV application on leg stiffness and on related-stability was (over)compensated by the larger mediolateral APA. Furthermore, results showed that these two balance control parameters were positively correlated, i.e., the larger the stance leg stiffness, the larger the mediolateral APA amplitude. Based on these findings, the authors propose that the increased stance leg stiffness induced by the WBV application was taken into account in the programming of the mediolateral APA. In this hypothesis, WBV application would induce a postural constraint in the form of an increased stance leg stiffness requiring motor adaptation to maintain stability. This interpretation is in agreement with the hypothesis that balance control mechanisms are interdependent within the postural system, i.e., they may compensate each other in case one component (here the leg stiffness) is perturbed (31, 92– 94). This interdependence would be necessary to maintain an optimal control of stability in situations where a constraint is applied to the postural system. This statement is supported by recent studies. For example, during stepping over an obstacle in reaction to sudden force-plate translation, Zettel et al. (46, 47) reported that the drastic reduction of the mediolateral APA amplitude due to the urgency to clear the swing foot from the support surface did not result in a greater instability because it was compensated by larger step width, which corresponds to another form of balance control mechanism. In the same vein, Artico et al. (94) reported that the amplitude of mediolateral APA for gait initiation with the goal to clear an obstacle depended on whether participants had to strike the ground with a "rear foot strategy" (where the swing heel strikes first, RFS) or a "forefoot strategy" (where the toe strikes first, FFS). More specifically, it was found that the amplitude of the mediolateral APA was lower in the FFS than in the RFS condition. Striking the support surface with the toe first mechanically enlarged by a few centimeters the mediolateral base of support compared to striking with the heel first, thus making the control of stability in the final posture easier. As a consequence, the need to develop large mediolateral APA was lessened in the FFS condition as compared to the RFS condition to maintain stability. Because APA has an energy cost (46, 47, 95, 96), it was proposed that the CNS would reduce them in the former condition for economical purpose. Similar statement was proposed to explain why APA for upper limb task are depressed when the postural stability is made higher by an additional thoracic support (97). The mediolateral APA would be programmed according to the swing foot strike strategy and related-stability, thus revealing an interdependence between these two mechanisms. Globally, the current results thus add to the growing evidence that dynamic stability during gait initiation may share a principle of homeostatic-like regulation similar to most physiological variables, where separate mechanisms need to be coordinated to ensure stabilization of vital variables, here postural stability (31).

# Effects of Acute WBV on Postural Organization of Gait Initiation in the Elderly

In contrast to young adults, no significant effects of acute WBV application for any of the biomechanical variables recorded was observed in the elderly. This finding may a priori contrast with current literature that instead tends to show a beneficial effect of WBV application on postural control in static position and on various clinical locomotor tests [for review (98)]. Now, it is noteworthy that these latter studies used long-term WBV treatments and not single bout as in the present study. It is therefore not excluded that long-term WBV application would result in dynamic stability improvement in the elderly. The result that no increase in stance leg stiffness was observed in the WBV group post-treatment may possibly be ascribed to the progressive alteration with aging in the central structures (e.g., spinal motoneurons) (99, 100), somatosensory receptors of the lower limbs (e.g., spindles and mechanoreceptors of the plantar foot sole) (101, 102), and/or sensory-motor pathways involved in the TVR reflex (103). This alteration would make elderly's stance leg stiffness less sensitive to acute WBV application than young adults. As a consequence, no motor adaptation e.g., in the form of larger APA or larger step width would be required to maintain stability since it was no further challenged by the WBV application. The current results however suggest that studies focusing on the effects of long-term WBV application in elderly should take into consideration the possibility that leg stiffness may be increased by this treatment. In case participants are not able to develop motor adaptation, instability may occur during locomotor tasks with increasing risks of falls.

# CONCLUSION

The present results showed that the effects of acute WBV application on the postural organization of gait initiation depend on the age of participants. In young healthy adults, a positive effect was observed on dynamic stability, while no effect was observed in the elderly. The positive effect was ascribed to an increase in the mediolateral APA following WBV application which overcompensated the potentially destabilizing effect of the increased stance leg stiffness. WBV application may thus be efficient to improve dynamic stability but at the condition that participants are able to develop adaptive changes in balance control mechanisms. In elderly, no anticipatory (nor corrective) postural adaptation was required since acute WBV application did not elicit change in the stance leg stiffness. Globally, the present finding is therefore in agreement with the hypothesis that balance control mechanisms are interdependent within the postural system, i.e., they may compensate each other in case one component (here the leg stiffness) is perturbed.

# 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 CIAMS Research Unit, Equipe d'Accueil (EA) 4532. The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

# AUTHOR CONTRIBUTIONS

AD, TV, MF, GL, LC, RA, FG, PF, and EY contributed with project creation, data collection, data analysis, drafted the manuscript, discussed the results, and participated in the revision of the manuscript.

# REFERENCES


of a systematic review and meta-analysis. Arch Gerontol Geriatr. (2017) 73:95–112. doi: 10.1016/j.archger.2017.07.022


in healthy young adults. Proc of The Phys Soc. (2010) 19:3069–77. doi: 10.3138/ptc.2014-77


**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 © 2019 Delafontaine, Vialleron, Fischer, Laffaye, Chèze, Artico, Genêt, Fourcade and Yiou. 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.

# An Immersive Motor Protocol for Frailty Rehabilitation

Elisa Pedroli <sup>1</sup> \*, Pietro Cipresso1,2, Luca Greci <sup>3</sup> , Sara Arlati 3,4, Lorenzo Boilini <sup>5</sup> , Laura Stefanelli <sup>5</sup> , Monica Rossi <sup>5</sup> , Karine Goulene<sup>5</sup> , Marco Sacco<sup>3</sup> , Marco Stramba-Badiale<sup>5</sup> , Andrea Gaggioli 1,2 and Giuseppe Riva1,2

<sup>1</sup> Applied Technology for Neuro-Psychology Lab, Istituto Auxologico Italiano - Istituto di Ricovero e Cura a Carattere Scientifico, Milan, Italy, <sup>2</sup> Department of Psychology, Università Cattolica del Sacro Cuore, Milan, Italy, <sup>3</sup> Institute of Intelligent Industrial Technologies and Systems for Advanced Manufacturing, National Research Council, Milan, Italy, <sup>4</sup> Department of Electronics, Information and Bioengineering, Politecnico di Milano, Milan, Italy, <sup>5</sup> Department of Geriatrics and Cardiovascular Medicine, Istituto Auxologico Italiano - Istituto di Ricovero e Cura a Carattere Scientifico, Milan, Italy

#### Edited by:

France Mourey, Université de Bourgogne, France

#### Reviewed by:

Frederic Merienne, ParisTech École Nationale Supérieure d'Arts et Métiers, France Federica Piras, Santa Lucia Foundation (Istituto di Ricovero e Cura a Carattere Scientifico), Italy

> \*Correspondence: Elisa Pedroli e.pedroli@auxologico.it

#### Specialty section:

This article was submitted to Neurorehabilitation, a section of the journal Frontiers in Neurology

Received: 09 April 2019 Accepted: 24 September 2019 Published: 15 October 2019

#### Citation:

Pedroli E, Cipresso P, Greci L, Arlati S, Boilini L, Stefanelli L, Rossi M, Goulene K, Sacco M, Stramba-Badiale M, Gaggioli A and Riva G (2019) An Immersive Motor Protocol for Frailty Rehabilitation. Front. Neurol. 10:1078. doi: 10.3389/fneur.2019.01078 Frailty is a pre-clinical condition that worsens physical health and quality of life. One of the most frequent symptoms of frailty is an increased risk of falling. In order to reduce this risk, we propose an innovative virtual reality motor rehabilitation program based on an immersive tool. All exercises will take place in the CAVE, a four-screen room with a stationary bike. The protocol will include two types of exercises for the improvement of balance: "Positive Bike" and "Avoid the Rocks." We will choose evaluation scales related to the functional aspects and subjective perception of balance. Our aim is to prove that our innovative motor rehabilitation protocol is as effective as or more effective than classical rehabilitation.

Keywords: motor rehabilitation, virtual reality, CAVE, frailty, elderly, stationary bike, balance, risk of falls

# INTRODUCTION

The constant increase of the elderly population compared to other age groups is now an evident phenomenon (1) which has led to increased efforts to propose solutions to the problems arising from the physiological condition of the elderly. Aging causes changes in both cognitive and motor functioning, which, depending on the degree of decline, can impact on different aspects of life with repercussions at various levels. In particular, it is possible to outline a condition of particular vulnerability in a part of this population, in patients defined as "frail," which represent 6.9% of adults over 65 years old (2). In this pre-clinical condition, there is a pattern of decline in the functioning of different aspects such as gait, mobility, balance, and cognitive functioning (3). These aspects associated with increasing age place these patients in a particular condition of vulnerability that is directly associated with a high risk of adverse health outcomes, mortality, disability, and more commonly a higher risk of falls (2, 4–6). The diagnostic criteria for this condition are: unintentional weight loss (10 lbs in the past year), self-reported exhaustion, weakness (grip strength), slow walking speed and low physical activity. Three or more of these criteria are needed for diagnosis according to the definition of Fried and colleagues (2).

Among the consequences of frailty mentioned above, the risk of falling is one of the most frequent and critical health problems occurring in the elderly and in particular in the frail population. It is estimated that one out of three elderly people falls at least once a year (7). This event has important consequences both for the autonomy of the individual and for problems in the psychosocial area, with further repercussions for cognitive functioning and quality of life (8, 9). The risk of falling in old age is a phenomenon that can be explained on the basis of the interaction between cognitive and motor factors. In fact, correct locomotion presupposes the possibility of simultaneously managing gait performance and one or more cognitive tasks (10). The presence of cognitive activity during the execution of motor tasks often occurs in daily activities (11, 12). The possibility of performing both tasks concurrently could be compromised in the elderly, who often show a decrease in attention skills and executive functions, thus leading to an increased risk of falls.

This phenomenon is interpreted according to the cognitivemotor interference (CMI) theory (13, 14), which states that the simultaneous execution of a motor task and a cognitive task represents a kind of dual-task (DT) interference that requires great cognitive resources and in particular attentive abilities and executive functioning (15). Depending on the complexity of the cognitive task, its simultaneous execution with motor performance may compromise the execution of motor performance, of cognitive performance or both (1, 15). However, on the basis of the information we have, DT tasks have been tested and proved to be an excellent tool for the simultaneous treatment of motor and cognitive abilities, and consequently for the recovery of abilities (such as gait) and the reduction of the risk of falls.

On the other side, work on balance separately is important in order to provide more focused exercises. A recent review (16) underlay the importance of balance in reducing the risk of falls in the elderly. In particular, old subjects with deteriorated balance fall more frequently than seniors with unimpaired postural control, which emphasizes the need for balance and postural training in this specific population (17). Almost all the studies that have investigated the prevention or the treatment of the risk of falling in the elderly conclude that different kinds of physical activity are effective for balance control and fall prevention (18). Osoba et al. (19) strongly recommend treatments to improve balance and gait in the elderly, in particular with virtual reality.

Considering frailty as a dynamic and reversible process with transition between states over time (20), many studies have focused on the possibility of reducing frailty and the risk of falls with specific interventions and activities that can prevent this condition (21). The motor rehabilitation approach based on physical exercise, both aerobic and to increase strength (22), has proved useful in reducing the risk of falls (23–26) and for the general improvement of cognitive functioning (27). Physical exercise, such as balance, strength, flexibility, and coordination training, is associated with a reduction in the risk of falls not only in healthy elderly people but also in individuals with cognitive impairment (28) and specifically in frail older people (23, 24). A one-size-fits-all program is not suitable, as the intensity of the exercise must be proportional to the patient's capabilities (29). But with regard to the kind of treatment, a recent systematic review showed that exercises to increase strength and postural balance are those most associated with the prevention of falls (25). Moreover, specific treatments for the training and recovery of balance mechanisms, such as treadmillbased systems, therapist-applied perturbations and perturbationbased balance training, would be more effective than general exercises (30, 31).

Several studies suggest the effectiveness of the integration of motor and cognitive training to decrease the risk of falls, and the DT approach seems to be one of the more efficient for the improvement of motor and cognitive abilities (29, 30). The contribution of higher-order cognitive systems such as executive functions makes this approach an effective training for the treatment of fall risk (10).

Virtual reality (VR) has improved the development and implementation of interactive cognitive-motor training programs. Ecological and realistic environments can be created by means of VR, which depict real/daily life situations with beneficial effects on patients' acceptance and adherence (32). Balance and functional mobility are the main domains tackled by VR with promising outcomes, suggesting this tool as an appropriate rehabilitative approach (33).

According to positive technology theory (34), interaction with technology leads to positive emotions and self-growth. The quality of psychological intervention can benefit from what is called "transformation of flow." According to Riva and colleagues, the user is able to exploit the optimal experience with VR and increase his/her involvement to obtain better performances (35, 36). Thanks to VR is possible to create a task both involved and challenging in order to engage patients, leading to promising results in cognitive and physical rehabilitation (37, 38).

VR cycling training for motor rehabilitation has been used in old adults and stroke patients (39–43); however, no one has implemented DT protocol with physical and executive functions. We will describe the rationale, design and usability of a fullyimmersive VR DT biking navigation called the "Positive Bike." To our knowledge, the majority of the research on balance training involves standing posture; fewer studies have focused on sitting posture rehabilitation (42). Stationary cycle exercises have a positive effect on weight shifts and gait, as well as the functioning of lower body limbs and a reduction of fall risk (44–46). Cycling also contributes to maintenance of specific balance coordination patterns and could help to preserve balance control and speed of voluntary stepping in the elderly (47). Walking is very close to cycling; indeed, they are both cyclical and activate agonist-antagonist muscles (48–50). Additionally, stationary cycling provides a controllable workload and safer equipment compared to the treadmill, leading to lower risk of injury in frail users (51).

Accordingly, in this paper an innovative VR-based protocol is proposed. The aim of the training will be to increase balance in frail people so as to reduce the risk of falls. This protocol will be developed within a national financed project with the purpose of creating both high- and low-end tools for motor (52, 53) and cognitive rehabilitation (54, 55). In this paper we will focus on the high-end motor part.

# MOTOR REHABILITATION TRAINING

# Inclusion/Exclusion Criteria

The eligibility criteria will require the participants to be 65 years of age and older, to match at least three of the five frailty criteria of Fried and colleagues (2) and to have an MMSE (56) score between 30 and 27. Fried's criteria include unintentional weight loss, self-reported exhaustion, weakness, slow walking speed and low physical activity. Exclusion criteria include significant vision impairments, presence of depression or anxiety without medications and hemianopsia or hemiplegia. The presence or absence of these criteria will be assessed during the initial clinical assessment performed by a physician. If patients report some depression or anxiety symptoms or the clinician suspect one of this problems he will be given specific tests, like Beck Depression Inventory [BDI-II (57)] or the State-Trait Anxiety Inventory [STAI Y1-Y2 (58)]. The final sample will be composed of 64 patients; in order to achieve this goal, at least 80 subjects will be assessed. To evaluate the size of the involved samples, we will use a Sample Size Calculation (Power Analysis) using the software GPower<sup>∗</sup> 3. The recent randomized trial assessing reduction in frailty (59) found that changes in frailty and mobility are similar in magnitude and represent medium effect sizes. Using their data, we estimated a minimum of 64 subjects to be included in the physical rehabilitation experiment in order to achieve a minimum power of 90%, considering a medium effect size of 0.4, a 15% dropout/non-compliance rate and a significance level of 0.05. Possible side effects connected with VR systems, such as nausea or dizziness, are referred to as cybersickness.

# Outcome Measures and Data Analysis

We will choose evaluation scales related to the functional aspects (points 1, 2, and 3) and the subjective perception of balance (point 4). We will also take objective data using the Neurocom Balance Master (point 5). A general muscle strength assessment (point 6) will take using a hand grip dynamometer, used also for testing the frailty of the patients. A trained physiotherapist performed the assessment in order to avoid low reliability of the data. All the information are included in the **Table 1**.



1. The Tinetti Balance Scale (60) is considered a gold standard for the validation of balance tests. It is a simple clinical that Our hypothesis is that our VR rehabilitation program is more effective than classic treatment in improving objective and subjective outcome measures. In order to confirm our hypothesis, we will perform Mixed Model ANOVA to compare the difference between the groups (VR VS NN-VR) and also between the time (T0, T1, T2) for each outcome measure collected. Also, the Bayes Factor will be used to determine if our program is more effective than classic treatment.

# Protocol

During the first medical examination, the inclusion and exclusion criteria were assessed by a physician. If the subject was considered suitable for the clinical protocol, outcome measures will be collected (T0) by a trained physiotherapist. Patients will be then randomly assigned to a control or experimental group using a randomization sequence obtained from the site randomizer.org. The first group will undergo classical physiotherapy, while the other one started a VR protocol. After 5 weeks without physical treatments, patients will return to the hospital to undergo a second evaluation (T1). Then, 10 biweekly rehabilitation sessions will start, and at the end a new assessment will be done (T2). The workflow is presented in **Figure 1**. Each session will last approximately 45 min and included both cycloergometer and dynamic exercises. To consider the treatment valid, patients will have to participate in at least 8 of 10 rehabilitation sessions and all the assessments; patients who will execute fewer than eight sessions will be considered drop-out (**Figure 1**). All participants will sign the written informed consent, which was approved by the Ethical Committee of IRCCS Istituto Auxologico Italiano. The study was conducted in compliance with the Helsinki Declaration of 1975, as revised in 2008.

# VR SETTINGS

The training will take place in a Cave Automatic Virtual Environment (CAVE). The CAVE system consists of a roomsized cube in which a combination of four stereoscopic projectors (Full HD 3D UXGA DLP) is used to obtain a 3D visualization of the virtual environment (VE) scene onto three walls, plus the floor. The projected right-eye and left-eye images are combined together by active goggles, making the perception of depth possible. In addition to the visualization devices, CAVE is equipped with an optical tracking system (VICON). Such a system allows the tracking of passive reflective markers and enables the correction of the spatial distortion of the simulated environment, which is eventually displayed in the CAVE with a 1:1 scale ratio. In our study, both CAVE goggles and an Xbox joystick are equipped with an asymmetrical set of markers allowing for the retrieval of their position and heading in the space. These pieces of data are used, respectively, to adjust the user's point of view and to enable the use of the Xbox joystick as a pointer for the interaction with 2D interactable elements (i.e., buttons) in the CAVE.

All the CAVE functionalities are handled by a cluster system composed of two HPZ620 Graphics Workstations, mounting Nvidia Quadro K6000 GPU with dedicated Quadro Sync cards.

Both VEs described in the following paragraphs were developed using Unity 3D and MiddleVR Unity plug-in. Thanks to this plug-in, the application deployed from Unity can communicate with all the CAVE system modules: the scene can be projected onto the CAVE walls, and the motion data retrieved from the VICON system can be exploited as inputs. The parts of the system are highlighted in **Figure 2**.

# Stationary Bike

The Positive Bike application requires a stationary bike (Cosmed EuroBike 320) placed inside the CAVE. Bike velocity and workload can be, respectively, read and set–via a serial cable– thanks to an ad-hoc developed protocol exploiting the bike manufacturer's Software Developing Kit (SDK). A pushing button is anchored on the cycloergometer handlebars for the detection of user interaction, and an Arduino2 board is used to connect the button to the computer.

Besides the GUI (graphical user interface), which is dedicated to the operator for the exercise parameters setting, the application is composed of two parts. The first one represents a trail in a park that flows according to the pedals' velocity (measured by the cycloergometer in revolutions per minute, RPM). The user bikes along the predefined path, which is created thanks to the placement of subsequent nodes on the route; the interpolation of such nodes is performed in real time using quaternion spherical linear interpolation (Slerp).

Since the user cannot deviate from the predefined route, the park is designed to discourage any desire to turn: there are no

road forks, and around the unpaved path there is just grass. To avoid boredom, some elements of the landscape change throughout the exercise, e.g., different species of plants and trees, lakes, buildings, etc. appear in the background. The path has some bends to increase the realism of the scene, but they are all very slight to avoid the occurrence of cybersickness related to the expectations of a lateral acceleration. Some tests made before this study ensured that no cybersickness arose in healthy subjects because of the bends.

During the exercise, participants are asked to keep their cycling velocity between 55 and 65 RPM. The bike workload is set by the therapist at the beginning of the exercise according to the subject's physical status. If the biking velocity is too low or too high, audio feedback is provided to the user: an acute sound is reproduced to signal to the user that he/she has to slow down; a grave sound is used to ask to speed up. The choice of signaling errors related to the physical part of the DT training was made to avoid distracting the elderly from the cognitive task by introducing an additional visual feedback.

The cognitive task of the exercise foresees the recognition of targets (**Figure 3**) appearing randomly on either the left or right side of the biking path. Targets are animals whose names start with a predefined letter that is communicated to the patients prior to the exercise beginning; other animals are considered distractors. The time elapsing between two subsequent targets is decided by the therapist who sets the exercise parameters. All the targets appear when the user is at a distance of 20 meters, so that he/she can clearly discriminate the targets' features. Target selection occurs by pressing the button placed on the cycloergometer handlebar while the target is still in the subject's visual field (i.e., it is displayed on the right or left wall of the CAVE).

Each time the user presses the button, he/she receives visual feedback regarding the correctness of the choice. No feedback is given if the user does not press the button, either if the choice is

FIGURE 3 | A frame of the "Positive Bike" environment.

correct (the displayed object is a distractor) or if the target has been missed. All data related to the cognitive exercise execution, as well as all the parameters set by the therapist, are stored in a user-dedicated folder in XML format.

The second part of the training occurs at the end of the biking: the screen displays a written question asking the subject how many targets he/she remembers having picked. The therapist types the answer on the CAVE computer's keyboard and saves this piece of information together with the exercise data saved at the end of the previously described scenario.

# Avoid the Rocks

The aim of this VE is training balance in frail people by using a virtual environment running in a CAVE. The VE simulates a walk on a straight road. Along the road, the user encounters obstacles (i.e., different-shaped rocks) and has to avoid them. Obstacles are positioned on the road so that the user is stimulated to perform

lateral movement (left and right) or to bend down to avoid hitting the rocks (**Figure 4**).

The user does not need to walk to proceed forward since his/her walking is simulated by displacing the user's point of view in the forward direction. The speed of the displacement can be increased or decreased by the operator pushing the "+" or "–" buttons on the keyboard.

To allow for the detection of collisions between the user and the rocks along the path, a virtual model of the player (i.e., a capsule-shaped object) is used. Such a model follows the user's head displacement in the CAVE, which is measured in real time thanks to the tracking of the 3D goggles. When the virtual model of the player is detected as colliding with an obstacle, it triggers the reproduction of a sound signaling the error. Similarly to the previous scenario, data regarding users' performance are stored in an XML file in a user-dedicated folder.

# CLASSICAL REHABILITATION

The training will take place in the rehabilitation gym of our hospital under the supervision of a physical therapist. In order to replicate the protocol proposed in the virtual rehabilitation, we developed two groups of tasks, one with the stationary bike and the other with classic balance tasks, as described below. Each part require 15 min, all the sessions are about 30–40 min according to the needed of the patients.

# Stationary Bike

We will use a stationary bike (the same model used in the VR protocol) placed in the gym. The therapist set the workload, increasing it session by session according with the training level gained by the subject. During the exercise, participants are asked to keep their cycling velocity between 55 and 65 RPM. No dual task was required.

# Balance Training

We will use a training protocol specific to balance in every subject. In literature, no specific tasks for increase balance in frail elderly people are reproted. We will use some devices such as a balance pad, proprioceptive footboard, rocking footboard, etc. for exercise mono- and bi-podalic station. We will train the subjects with and without visual deprivation. The workload is regulated according to the physical status and performance ability of the subject. Example the therapist ask to patients to mantein balace standing on one foot with the arms cross on the chest.

# DISCUSSION

The aim of this VR rehabilitation protocol will be to improve balance and reduce the risk of falls in frail elderly people. In order to assess these hypotheses, we will develop an innovative tool using an immersive VR system, the CAVE. We will compare this innovative protocol with a selection of classical physiotherapy exercises with the same purposes.

# Expected Results and Limitations

According to our hypotheses, we would like to prove that our innovative motor rehabilitation protocol is as effective as or more effective than classical rehabilitation. We will include both subjective and objective measures in order to better understand the degree of improvement subjects will obtain. Positive Bike aims to improve the dual-task abilities of frail elderly people using an innovative, engaging and challenging training. We hope that both the subjective and objective measures will increase after our training.

# Future Steps

Several studies (66–68) have showed that continuing rehabilitation activities at home contributes to the maintenance of benefits obtained. Accordingly, we are developing a low-end VR tool to promote physical rehabilitation at home. This new system will be tablet-based and will exploit the potential of 360◦ videos (68–70). To our knowledge, there are no studies that have tested this technology for motor rehabilitation. 360◦ videos are usually enjoyed by using head-mounted displays. We decided to use tablets instead of head-mounted displays to reduce the risk of injuries. Performance of balance exercises excluding patients from the "real environment" could be risky, and tablets are a safer tool for the use of 360◦ videos. We will try to replicate the protocol used for rehabilitation in the high-end VR setting by adapting it to low-end technology. We will also provide the patients with a portable cycloergometer in order to perform the dual-task exercise.

# AUTHOR CONTRIBUTIONS

EP, PC, LB, MR, LS, and KG conceived and designed the protocol. LG and SA participated in the design, developed the environments, and integrated the cycloergometer

# REFERENCES


functionalities. EP and LG wrote the first draft. MR provide the required revisions. MS, MS-B, GR, and AG are supervisors. All the authors revised the final version of the manuscript.

# FUNDING

This work was supported by the Italian-funded project High-End and Low-End Virtual Reality Systems for the Rehabilitation of Frailty in the Elderly (PE-2013-02355948).


Ossébo randomised controlled trial. BMJ. (2015) 351:h3830. doi: 10.1136/bmj. h3830


**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 © 2019 Pedroli, Cipresso, Greci, Arlati, Boilini, Stefanelli, Rossi, Goulene, Sacco, Stramba-Badiale, Gaggioli and Riva. 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.

# Comparison of Step Count Assessed Using Wrist- and Hip-Worn Actigraph GT3X in Free-Living Conditions in Young and Older Adults

Stephane Mandigout <sup>1</sup> \*, Justine Lacroix <sup>1</sup> , Anaick Perrochon<sup>1</sup> , Zdenek Svoboda<sup>2</sup> , Timothee Aubourg3,4 and Nicolas Vuillerme3,5

<sup>1</sup> Université de Limoges, HAVAE, EA 6310, Limoges, France, <sup>2</sup> Faculty of Physical Culture, Palacky University Olomouc, Olomouc, Czechia, <sup>3</sup> Univ. Grenoble Alpes, AGEIS, Grenoble, France, <sup>4</sup> Orange Labs, Meylan, France, <sup>5</sup> Institut Universitaire de France, Paris, France

Background: Walking represents a major component of physical activity (PA), and its restriction could degrade autonomy and quality of life. An important objective for preventive and/or rehabilitative strategies to improve balance and gait in normal and pathological aging conditions is to focus on physical activity. Activity monitors have recently been getting increasingly popular and represent a modern solution to measure—and communicate—PA notably in terms of steps/day. These activity monitors are well-suited for various populations as they can be worn on a variety of locations on the body, including the wrist and the hip (i.e., the two most common locations), in an undifferentiated way according to the manufacturer's instruction. The aim of this study was hence to verify potential differences in step count (SC) by comparing this parameter assessed using wrist- and hip-worn activity trackers over a 24-h period in free-living conditions in young and older adults.

#### Edited by:

Helena Blumen, Albert Einstein College of Medicine, United States

#### Reviewed by:

Christopher Buckley, Newcastle University, United Kingdom Lisa Roberts, University of Alabama at Birmingham, United States

> \*Correspondence: Stephane Mandigout stephane.mandigout@unilim.fr

#### Specialty section:

This article was submitted to Geriatric Medicine, a section of the journal Frontiers in Medicine

Received: 30 January 2019 Accepted: 21 October 2019 Published: 05 November 2019

#### Citation:

Mandigout S, Lacroix J, Perrochon A, Svoboda Z, Aubourg T and Vuillerme N (2019) Comparison of Step Count Assessed Using Wristand Hip-Worn Actigraph GT3X in Free-Living Conditions in Young and Older Adults. Front. Med. 6:252. doi: 10.3389/fmed.2019.00252 Methods: Young adults (n = 22) and older adults (n = 22) voluntarily participated in this study. They were required to wear two commercially-available Actigraph GT3X+ activity monitors simultaneously at two locations recommended by the manufacturer, i.e., one positioned around the wrist and one above the hip, over a 24-h period in free-living conditions. The manufacturer's software was used to obtain estimates of the SC.

Results: For both groups, the wrist-worn activity tracker provided significantly higher SC than the hip-worn activity tracker did. For both placements on the body, older adults exhibited significantly lower SC than young adults. Interestingly, for both young and older participants, the difference between both measurements tended to decrease for longer distances.

Conclusion: The different estimations of the step count provided by the comparison between two identical Actigraph GT3x on the wrist or the hip during the 24-h observation period in free-living conditions in young and older adults strongly suggests that caution is needed when using total step per day values as an outcome to quantify walking behavior. Probably we can suggest the same caution across implementation of different activity Tracker.

Keywords: activity tracker, step count, physical activity, older, free-living

# INTRODUCTION

Walking represents a major component of physical activity (PA), and its restriction could degrade autonomy and quality of life (1). An important objective for preventive and/or rehabilitative strategies to improve balance and gait in normal and pathological aging conditions is to focus on PA. Activity monitors have recently been getting increasingly popular and represent a modern solution to measure—and communicate—the amount of PA performed by its user (2).

A patient's level of PA can be estimated using either the daily energy expenditure or the step count (SC). PA recommendations are generally defined on the basis of energy expenditure (Kcal, MET.min−<sup>1</sup> , MET.h−<sup>1</sup> ) (3). Unfortunately, the accuracy of the estimation of energy expenditure by activity trackers (AT) appears questionable regardless of the population (4). On the other hand, the SC is presented as a relatively stable and reliable indicator (2). Currently, no recommendation have been published regarding this indicator (2). Despite this, the SC may be used as a relevant indicator of a person's amount of PA, lifestyle (active vs. sedentary) and physical inactivity (2).

AT may be worn on a variety of locations on the body, including the ankle, the wrist, the hip, and around the neck. Depending on the model and manufacturer, some AT may be placed indiscriminately at different locations on the body (Actigraph GT3X for instance). The location of the AT must then be registered in the software for the device to correctly define the algorithm used to estimate the SC. In normal use, these algorithms are not publicly available because they are the property of the manufacturer. Moreover, fitting the device with a specific algorithm for its location implies that the same SC should be found regardless of the activity performed by the person.

The literature shows that the most accurate way to evaluate the SC under free-living conditions is to place the AT at the ankle (5, 6). However, for practical and aesthetic reasons, ATs are regularly placed on the hip or the wrist.

Incorrectly positioning the AT may alter its results due to the technology it relies on. Indeed, two main technologies are used based on the internal mechanism used to record steps, i.e., spring-suspended lever arm or accelerometer, among which accelerometry is increasingly used. Depending on the technology used and the position of the AT on the body, the resulting SC may be significantly different (2). Additionally, significant differences tend to appear depending on the type of activity, laboratory conditions (7, 8) or standardized activities (walking, running) (9–12), or free-living condition (5, 13), between the possible AT positions.

The Actigraph GT3X (Actigraph LLC, Penascola, FL, USA) represents the epitome of scientific accelerometers as it is unobtrusive, low-cost, and its sensitive triaxial accelerometers are capable of storing high-resolution, raw, unfiltered acceleration signals over long durations. The Actigraph monitor has been extensively studied in many situations: the validity for the evaluation of PA in healthy or pathological populations and the comparison with other AT (13–17); used as a gold standard in some studies (18, 19). Despite all these studies, the recent systematic review by Migueles et al. (20) conclude that it is necessary to take a cautious approach regarding the accuracy/reliability of Actigraph in estimating the SC in real-life situations as a function of 1-the mainly used positions (wrist and hip) and 2- the age of the subjects.

In order to better assess the accuracy according to the position or the type of AT, some studies have focused on the calculation of the absolute error rate (AER) (21). This calculation allows to determine whether wrist-hip differences could be attributed to a factor inherent to the AT or inherent to the subjects. Some studies have used this parameter in their experimental design (9, 19, 22). To the best of our knowledge, no study has addressed the relationship between the percentage of absolute error between the SC of the two prominent AT positions (hip and wrist) and the age of the wearer with the Actigraph GT3X. Within this context, the aim of this study was to compare the AER of SC assessed using wrist- and hip-worn Actigraph GT3X over a 24-h period in free-living conditions in young and older adults.

# METHODS

# Study Population

Our study population was aged between 18 and 85 years, without medical contraindication, and volunteered to participate after signing a consent form. The exclusion criteria were: any cardiovascular pathologies or mobility issues. The sample was divided into two groups: a group of subjects aged 18–45 years and a group of subjects aged 70–85 years. The protocol was approved by the Comité d'Ethique pour les Recherches Non-Interventionnelles (CERNI) of the Grenoble-Alpes University, France. All subjects gave written informed consent in accordance with the Declaration of Helsinki.

# Materials

The material requirements for the study were two Actigraph GT3X accelerometers (ActiGraph Pensacola, FL, USA, www. actigraphcorp.com). These tri-axial accelerometers are used to record the SC along with various PA data. Accelerometer data were collected at a frequency of 80 Hz and aggregated to 60-s epochs for analyses. Following the manufacturer's guidelines (23), a Low Frequency Extension (LFE) filter was used to increases the device's sensitivity and detect low-frequency accelerations (i.e., slow walking).

# Experimental Design

This study was designed to record the PA of a sample in a freeliving situation for 24 h. The AT were positioned as follows: one Actigraph GT3X at the hip (in the center of the pelvis) and a second one at the non-dominant wrist. Subjects were asked to remove the device before showers and for aquatic activities. To compare accelerometer data according to each location, we limited the data from all devices to the actual wearing time when both devices were worn. The Actigraph was placed in the morning (between 8 and 11 a.m.) and was picked up the next day at the same time. The registration period was programmed using the manufacturer's Actilife software v 6.13.3 (www.actigraphcorp. com/actilife/). A 24-h period of recording allowed us to avoid the risk of human failure (weariness, forgetfulness...). After verification by an investigator, the records appeared to be correct and usable.

The parameter used in this study was the SC. All equipment was activated before placing it on the previously described locations. The minimum required recorded duration of accelerometer data to be included in the analysis was 24 h for both AT. After the 24 h of recording, the subject was requested to return the equipment in order for the practitioner to transfer it using ActiLife and reset the devices for new use.

# Statistical Analysis

The step count data were presented in the form of mean and standard deviation. Firstly, to compare these data in free-living conditions in young and older adults according to trackers location, statistical tests of comparison were selected by testing step count data for normal distribution using the Shapiro-Wilk test. As the dependent variables did not conform to a Gaussian distribution, non-parametric comparative tests were chosen for the statistical analysis process. These tests were performed into two successive steps as follows (**Figure 1**):

(1) Comparison analysis. We compared the differences between the step counts provided by the wrist-worn and hip-worn trackers using two assessment criteria, namely the significance and the effect size of these differences. Significant differences were assessed by means of non-parametric Wilcoxon signed-rank tests. Effect sizes, also known as magnitude, were obtained using Cohen's d. This coefficient was calculated as a ratio of mean difference divided by mean standard deviation in both conditions. Effect sizes were considered small if d < 0.5, medium if 0.5 ≤ d < 0.8 and large with d ≥ 0.8 (24). We completed this statistical procedure by comparing the difference of measurement between hip-worn and wrist-worn AT according to the age category of the participants using Wilcoxon signedrank tests.

(2) Association analysis. The results from step (1) were then complemented with an additional analysis to conclude whether trackers could remain exchangeable despite potential differences of measurement, and if so, to what extent. For this purpose, four assessment criteria were used, namely relation, reliability, agreement and variation. Relation between the step counts provided by the wrist-worn and hip-worn AT was calculated by means of Spearman's rank correlation coefficient rho. Reliability was measured by means of intraclass correlation coefficient (ICC). An ICC value between 0.00 and 0.40 was considered poor, 0.40 and 0.59 was fair, 0.60 and 0.74 was good, and 0.75 and 1.00 was excellent (25). The obtained scores were reported in Bland-Altman plots to visualize the agreement between the wrist-worn and hip-worn AT. Finally, for a comparison purpose, we assumed the hip location could be used as reference to study errors of measurements. To this end, we assessed the variation of the error measurement generated by the hip-worn AT according to the step count measured by the wrist-worn AT by calculating the absolute error ratio (AER). For each method, the absolute error for each estimated parameter [IC, FC, stride time (mean and CV), step time (mean and CV), and swing time (mean and CV)] was hence determined relatively to hip-worn trackers as follows:

$$AER = P - \Pr\_{\text{(1)}} \tag{1}$$

Where pr is the reference value of the parameter p.

The level of significance was set as p < 0.05 in all statistical tests. All statistical calculations were completed using the R software environment (version 3.1.0; R Foundation for Statistical Computing, Vienna, Austria).

# RESULTS

A total of 44 volunteers participated in this study. Participant characteristics are summarized in **Table 1**.

# Comparison Between the SC Provided by the Hip-Worn and by the Wrist-Worn AT

The SC for both AT are summarized in **Table 2**.

Regarding the overall population, measurements at the wrist were significantly higher than measurements at the hip with, respectively, 11,203 (SD = 4,543) vs. 6,866 (SD = 4,655) counted steps in average. A significant difference was also found in the young participants group, with, respectively, 11,347 (SD = 3,258)

FIGURE 1 | Synthesis diagram of the statistical treatment.

#### TABLE 1 | Participant characteristics.


#### TABLE 2 | Step count analysis.


Descriptive statistics, comparison, correlation, agreement, and Bland-Altman parameters for the numbers of steps provided by the wrist-worn and hip-worn AT for all participants and by age group. SD, standard deviation; ICC, Intraclass coefficient correlation; CI, confidence interval.

\*From the regression equation: Mwrist = α + β.Mhip, where Mwrist is the wrist-worn measurements variable and Mhip is the hip-worn measurements variable.

counted steps for wrist-worn AT vs. 7,810 (SD = 3,969) counted steps for hip-worn AT. Interestingly, this contrast increased for older participants: the mean SC provided by the wrist-worn AT was almost twice the value provided by the hip-worn devices, with, respectively, 11,060 (SD = 5,787) vs. 5,922 (SD = 5,172) counted steps in average. We then validated the significance of these measurement differences in the overall population, but also in young participants and older adults separately (p-values < 0.00001 for all three cases using Wilcoxon comparison tests, where the null hypothesis was a similarity between hip-worn and wrist-worn AT measurements). In addition, Cohen's d points out the strong effect size of this phenomenon for the overall population (d = 0.93), but also for both young (d > 1) and older participants (d > 0.8). We may however note that the error measurement between hip-worn and wrist-worn AT is significantly lower in young than older participants (p-value = 0.0065 and absolute Z-score = 2.48).

# Association Between SC Provided by Hip-Worn and Wrist-Worn AT

Correlation analyses showed significant positive relationships between the SC for wrist-worn and hip-worn AT in the overall population (Spearman's rho = 0.76, p < 0.001), for the young participants (Spearman's rho = 0.85, p < 0.001), and for the older participants (Spearman's rho = 0.70, p < 0.001) (**Figure 2**).

The Bland-Altman plot for the SC measured for both AT positions is provided in **Figure 3**.

As indicated in **Table 2**, the estimated bias, i.e., the mean of the differences between the measurements of the wrist-worn and hip-worn AT, is 4,337 counted steps. This result implies that the wrist-worn AT tends to overestimate the SC in comparison to hip-worn AT.

# Variation of the Error Measurement

Finally, we assessed the error rate generated by the location of the AT in order to account for the possibility for this error to decrease at a certain threshold.

The absolute error rate (AER) between the SC provided by the wrist-worn AT and by the hip-worn AT is shown in **Figure 4**.

These results highlight a significant negative correlation between the AER and hip-worn SC in the overall population (Spearman's rho = −0.77, p < 0.001), in young participants (Spearman's rho = −0.87, p < 0.001) and in older participants (Spearman's rho = −0.58, p < 0.001). In other words, the error in the SC provided by the hip-worn AT tends to decrease according to the distance walked.

# DISCUSSION

The objective of our study was to compare the absolute error rate (AER) and the SC assessed using wrist- and hip-worn Actigraph GT3X over a 24-h period in free-living conditions in young and older adults. Our results show that the more the individual walks during the day, the more the error in the SC provided by the hip-worn Actigraph GT3X tends to decrease. Our results further demonstrate an overestimation of the SC, as the SC measured by the Actigraph GT3X at the wrist is 39% higher than the SC measured at the hip (p < 0.05). Moreover, the hip-wrist difference is even more significant in older adults (p = 0.0065). Age could therefore be a factor influencing the measurement difference of the SC recorded by two Actigraph GT3X placed at different locations. To the best of our knowledge, no study has compared the SC difference given by identical Actigraph GT3X positioned at the hip and wrist during a 24 h recording in freeliving conditions in young and old adults. Two recent studies conducted in young adults (13) and in older women (15) reported that the SC recorded by the Actigraph GT3X at the wrist was significantly higher than the number recorded at the hip when recording an activity in a real life situation. These results suggest a difference in the SC between the two positions, without any real

indicators of the precision of one in relation to the other. The effect of age was however not assessed.

# Actigraph GT3X Accuracy Between Hip and Wrist According to Age

One of the parameters which could explain our results is the decrease of accuracy of AT when recording slow activities (11). Indeed, several studies demonstrate that few AT are capable of recording motion slower than 1 m.s−<sup>1</sup> (26).

Older adults (60+) have been shown to self-select a walking speed of 1.18 m.s−<sup>1</sup> (±0.17 m.s−<sup>1</sup> ) (27) which can extend to 1.34 m.s−<sup>1</sup> (±0.21 m.s−<sup>1</sup> ) in healthy older adults (28). Webber et al. (17) carried out a comparative study (ActiGraph vs. Stepwatch, Hip vs. Ankle) during a hallway walk in 38 geriatric rehabilitation patients (83.2 ± 7.1 years of age), walking at a comfortable pace (0.4 m.s−<sup>1</sup> ). Speeds under 1 m.s−<sup>1</sup> are indeed commonly experienced in elderly populations and people with motor disabilities. In the study by Webber et al., the AER was low for slow walking speeds (<3%) and did not significantly differ between the StepWatch and the Actigraph GT3X+ (placed at the ankle); however, error values were higher (19–97%) when the Actigraph GT3X+ was worn at the hip during a hallway walk. In this study, the comparison involved two AT models, among which the Stepwatch was considered as the reference. In our study, the AER values were estimated at 39% between two Actigraph GT3X, positioned at the wrist and hip. Our results suggest that an activity monitor placed at the wrist can potentially overestimate low-speed activities and underestimate high-speed activities. According to Aziz et al. (29), wrist kinematics may represent a relatively small part of total body movements during walking (especially when walking with limited arm swing), and a relatively larger one during some sedentary activities such as simply moving hands while sitting.

Furthermore, Feng et al. (10) compared the accuracy of three commercially available accelerometers (Axivity AX3, Actigraph GT3X, and APDM-Opal) and pointed out the importance of customizing the AT placement and algorithms to maximize the measurement accuracy when selecting accelerometers specifically designed to measure the SC for slower walking speeds. Unfortunately, these algorithms are proprietary information and cannot be accessed by the standard user. Despite the accessibility of information, the Actigraph implements a Low Frequency filter (LFE) in the Actilife software. A normal filter can detect accelerations within a frequency range of 0.25–2.5 Hz, while the LFE filter establishes a lower threshold to capture slower movements. Despite this filter, the results vary drastically depending on the body positions (20).

# SC as a Function of at Locations on the Body

The lower SC provided by the Actigraph GT3X placed at the hip tended to generate greater absolute error between both positions (hip vs. wrist). The traditional AT was designed to be worn at the hip, attached to a belt or waistband (2). Many studies demonstrate that the accuracy of the measurement of the SC

would increase when the AT is worn on the hip compared to the wrist during standardized activities such as walking and running (9, 13, 30). This higher accuracy could be explained by the fact that an AT worn at the hip is closer to the body's center of mass, which would facilitate the detection of the whole body's acceleration. **Figure 4** highlights significant correlations between the SC at the hip and the AER of SC in all samples of the population. This leads us to consider the Actigraph's accuracy with caution depending on its position. **Figure 2** shows a good correlation between the SC of both Actigraph GT3X, but with poor ICC. In addition, the wide range in the limits of agreement ([−1,314; 9,987]) emphasizes the idea that the two device locations should not be used interchangeably without accounting for the existing bias when measuring the SC. In other words, an Actigraph GT3X worn at the wrist will tend to be less accurate than the same AT worn at the hip. The percentage of error can induce a significant difference in the SC according to the position of the AT.

Other factors may be involved in the significant SC differences observed between each position during activities in reallife conditions:



# Practical Application

Our study shows that two identical Actigraph GT3X placed at the hip or at the wrist will generate a consequential AER in real-life situations. We were able to affirm that this error was multifactorial. In a study comparing the accuracy of the Actigraph GT3X and ActivPal, Steeves et al. (8), showed that a 4% difference in SC may amount to an error of 37 extra

minutes of walking per day. Out of a 7-days observation period, the use of an AT with underestimation errors higher than 14% may translate into errors corresponding to more than one entire day of walking activity (30). As a reminder, the World Health Organization recommends 30 min of walking per day, 5 days a week. In our study, the average of the differences between the SC of the wrist vs. the hip is 30% higher in older adults compared to younger subjects (**Table 2**). These 30% represent a large part of their daily activity. In the light of our work, we argue that the accuracy of the sensors, which directly depends on the technology and processing algorithm, will have to be considered differently between young and old subjects or patients with motor disabilities. Indeed, elderly subjects tend to take fewer and slower steps, which strongly influences the variation between the estimated and real SC. Therefore, it seems essential to accurately identify the target population and the intended type of activity. Besides, the scientific literature clearly shows that a consistent use of the same AT is essential. A wide variety of models are indeed available on the market, and the SC obtained for a given activity is often different from one AT to another (33). Moreover, the position of the AT on the body seems to be an even more discriminating criterion of SC. Our work demonstrated that the difference between the SC measured at the wrist and hip

can range from 30% in young subjects to nearly 50% in elderly subjects. The lack of accuracy in the measurement of the SC from an AT placed at the wrist may represent an issue for scientific uses, as it would reflect the number of movements performed by the upper limbs and not a real SC. It may however be a good indicator of a more global amount of PA. To obtain an optimal evaluation of the SC as close as possible to reality, placing the AT at the hip appears to be the most favorable position, with the exception of the ankle. Furthermore, a wide majority of activity tracking devices tend to underestimate the SC at slow speeds (<1 m.s−<sup>1</sup> ) (2). The few AT models which are able to correctly identify slow steps are generally expensive (>\$400) and unaffordable for the general public.

# Limitations of the Study

Our study may have been limited by its small population. However, our results are supported by the many results within the literature and provide a major complement to the use of Actigraph GT3X at the wrist or hip. Besides, as there is no gold-standard solution to evaluate the SC in freeliving situations, our work does not allow us to assert which Actigraph GT3X provides the most precise values, i.e., the closest to reality. Further studies are therefore required to identify optimal AT placements at least for low-to-moderate activity, and in which position these monitors are mostly used by consumers during free-living conditions. In light of our study, it seems necessary to carefully consider the position of the AT, the age of the users and their lifestyle habits to achieve this objective.

# CONCLUSION

Our study showed that wearing the AT at the wrist may provide overestimated SC compared to the same AT model placed at the hip in young and elderly people in free-living conditions. On the one hand, this difference appeared to be accentuated according to the age of the subjects. On the other hand, it seems that the difference between the two positions tended to decrease for higher SC. These results suggest the hypothesis that the gait speed is an essential criterion when estimating the SC using an accelerometer. The assessment of the amount of PA in free-living conditions based on the SC remains uncertain and imprecise. The literature on the subject is extremely abundant and rather difficult to synthesize. Further work will be needed to improve the quality of SC measurement in free-living conditions for all populations (young, old, healthy or patient).

# REFERENCES


# ETHICS STATEMENT

All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the CERNI of the Univ. Grenoble Alpes, France.

# AUTHOR CONTRIBUTIONS

SM: conceptualization, project administration, methodology, formal analysis, and writing original draft preparation. JL and AP: conceptualization, investigation, and methodology. ZS: validation. TA: formal analysis. NV: conceptualization, methodology, and formal analysis. All authors read and approved the final manuscript.

# FUNDING

This work was supported by the Région Nouvelle Aquitaine and the French National Research Agency in the framework of the Investissements d'avenir program (ANR-10-AIRT-05 and ANR-15-IDEX-02). The sponsors had no involvement in the review and approval of the manuscript for publication. This work forms part of a broader translational and interdisciplinary project, GaitAlps.


#### **Conflict of Interest:** TA was employed by the company Orange Labs.

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 Mandigout, Lacroix, Perrochon, Svoboda, Aubourg and Vuillerme. 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 Novel Square-Stepping Exercise Program for Older Adults (StepIt): Rationale and Implications for Falls Prevention

#### Eleftheria Giannouli\*, Tobias Morat and Wiebren Zijlstra

*Institute of Movement and Sport Gerontology, German Sport University Cologne, Cologne, Germany*

The ability to effectively execute compensatory steps is critical for preventing accidental falls, and consequently stepping training is an essential ingredient of fall prevention programs. In this paper, we propose a concept for stepping training that aims to maximize training effects by taking into account recent research evidence and a precise dosing of training ingredients. The concept addresses motor as well as cognitive falls-related aspects, it is suitable for individual as well as group based training, and it does not require costly equipment. Theory and evidence behind all of the training principles is reviewed, and an example of an exercise protocol is described in detail. Participants are presented with stepping patterns which they have to memorize and implement on a mat. In order to enable investigation of dose-response effects, the difficulty level systematically and gradually increases session by session based on four principles: execution speed, pattern complexity, pattern length and execution in dual-/multi-tasking conditions. The presented concept can be used as a framework for the development of further prevention and/or rehabilitation stepping exercise programs. Further studies using this exercise regimen or modified versions of it are encouraged.

Keywords: balance, gait, mind-motor, dual-task training, aging, rhythmic auditory cueing, rhythmic auditory stimulation (RAS), variable practice

# INTRODUCTION

In order to release older adults and healthcare systems from the burden of accidental falls, it is important to timely detect fall risk factors and develop effective interventions for falls prevention. Some of the strongest predictors for falls are deficits in lower-limb strength, balance, and gait performance (1). Specifically, accidental falls in older adults can be frequently attributed to the inability to step precisely on the ground and to incorrect weight-transferring during everyday activities such as (single-task) walking (2, 3) or in situations that require a simultaneous performance of several tasks (4). This can be explained by the age-related changes in spatiotemporal characteristics of stepping for balance; older individuals tend to: take too short steps or steps in the wrong direction (5, 6), to collide one leg against the other during oblique steps (7, 8), have slower stepping reactions (9) and need more attentional resources while walking under dual-task conditions (10).

The most important component of falls prevention exercise programs is balance training (11) since good balance is crucial for maintaining postural equilibrium and thus for the avoidance of

#### Edited by:

*Helena Blumen, Albert Einstein College of Medicine, United States*

#### Reviewed by:

*Lisa Robinson, Newcastle Upon Tyne Hospitals NHS Foundation Trust, United Kingdom Klara Komici, University of Molise, Italy*

#### \*Correspondence:

*Eleftheria Giannouli eleftheria.giannouli@unibas.ch*

#### Specialty section:

*This article was submitted to Geriatric Medicine, a section of the journal Frontiers in Medicine*

Received: *18 April 2019* Accepted: *13 December 2019* Published: *14 January 2020*

#### Citation:

*Giannouli E, Morat T and Zijlstra W (2020) A Novel Square-Stepping Exercise Program for Older Adults (StepIt): Rationale and Implications for Falls Prevention. Front. Med. 6:318. doi: 10.3389/fmed.2019.00318*

**328**

falls. Balance exercise programs often focus on standing balance tasks, where the center of mass has to be statically controlled over the base of support. However, these lack in ecological validity and are not specific enough to cause neuromuscular adaptations that are actually required in balance-threatening situations. As mentioned earlier, in real life, maintaining balance to avoid a trip or a slip requires fast (rather than static/slow) stepping movements as well as high foot placement accuracy in order to initiate a correct step or inhibit a wrong one to quickly avoid an obstacle or an unexpected perturbation. Therefore, effective falls prevention and rehabilitation exercise programs should focus on performing precise, rapid and well-directed steps.

Stepping training is a form of highly specific balance training for falls prevention which directly addresses stepping capacity—a commonly executed protective strategy for maintaining balance in the everyday environment (12) and also an important fall risk factor in older adults (13, 14). A recent systematic review and meta-analysis showed, indeed, that both reactive and volitional stepping interventions reduce falls among older adults by ∼50% (15).

In addition to this, physical activity in general and even more so structured physical exercise has shown to improve cognitive fall risk factors (16). In fact, combined physical and cognitive training (such as a stepping training containing additional cognitive tasks), may lead to larger improvements in cognitive and physical outcomes compared to physical or cognitive training alone (17), possibly with greater impacts on daily functioning. Thus, targeting deficits in both mobility and cognition through dual- and/or multi-tasking exercise programs is likely to represent the most effective strategy to minimize cognitive and physical declines in healthy older adults and therefore reduce the risk of falls and cognitive impairment.

Finally, regardless of type of training (strength/balance/stepping training), difficulty level (i.e., including increasingly challenging exercises e.g., either by adding a cognitive element, having additional movements, or increasing speed) is crucial for the success of falls prevention programs. This has been shown by systematic reviews (17, 18) as well as by a recent study monitoring movement characteristics of stepping exergames (19).

In this paper we propose a concept for stepping training (StepIt) that aims to maximize training effects in stepping capacity by taking into account recent research evidence and a precise dosing of training ingredients. In the next sections, we first review the theory and evidence behind Rhythmic Auditory Stimulation (RAS) used as an overall tool for stepping interventions aiming to prevent falls. Afterwards, we elaborate on all defined key elements of the proposed training by explaining their (neural) mechanisms. Finally, an exemplary protocol of the StepIt exercise program which includes all the aforementioned principles is described in detail and based on that, recommendations are presented on how this exercise program can be adjusted to fit needs of different purposes and populations. The presented concept can be used as a framework for the development of further prevention and/or rehabilitation stepping exercise programs.

# RHYTHMIC AUDITORY STIMULATION

At the StepIt exercise program participants are presented with stepping patterns which they have to memorize and then execute on a grid-like rubber mat using the method of rhythmic auditory cueing as execution is done at paces delivered by a metronome. Rhythmic Auditory Stimulation (RAS) is "a neurologic technique using the physiological effects of auditory rhythm on the motor system to improve rhythmical movements like for example gait" (20). To cue movements, metronomes and music (or a combination of them, i.e., metronome tone embedded into music) are the most frequently used rhythmic auditory stimuli tools. They have been used very effectively in the context of RAS-based motor rehabilitation programs for patients suffering from various movement disorders (21), mostly as a result of neurological diseases like stroke (22, 23), Parkinson's disease (24, 25) and multiple sclerosis (26). It has been consistently reported that stepping in time to a metronome can improve pathological gait in neurological patients e.g., in terms of increased walking speed (27), reduced step time asymmetry and step time variability (28) and increased stride length (29) as well as in healthy, but fallprone, older adults (24), even after only one training session (30).

This profound effect of auditory rhythm to the motor system is due to the rich connectivity between these two systems (auditory and motor) across a variety of cortical, subcortical and spinal levels. In fact, the motor system is so sensitive to stimulation by the auditory system that even simply listening to an auditory rhythm engages motor areas in the brain (31). The auditory system detects and processes temporal information with high precision and speed (it is actually about 20–50 ms faster and more precise than the visual and tactile systems (32) and subsequently projects them into motor structures in the brain, creating entrainment between the rhythmic signal and the motor response (33). In physics, entrainment is defined as the process in which one system's motion or signal frequency entrains the frequency of another system. In clinical language, entrainment refers to the process where the brain's internal timekeeper adjusts to external timekeepers (music/metronome etc.) enabling enhanced motor control in terms of increased anticipatory mapping and scaling of optimal velocity and acceleration gait parameters across the fixed movement interval (34). This movement "reprogramming" results in higher gait speed as well as smoother and less variable movement and muscle activation (35).

Neuroimaging studies have described the neural basis for auditory-motor entrainment. The auditory system has fiber connections from spinal motor neurons and up until the brain stem, subcortical, and cortical levels (34). Brain areas involved in rhythm processing, timing and duration perception, are closely related to those which control movement, such as the premotor cortex, supplementary motor area (SMA), cerebellum and basal ganglia. The cerebellum, which is involved in sensorimotor coupling, may monitor rhythmic patterns and adjust behavior to changing tempos and therefore control rhythmic auditorymotor synchronization. The putamen, and the basal ganglia in general, are associated with rhythmic events and beat perception (36).

Overall, the process of continuous entrainment resulting from the attempt to synchronize the movements with the rhythm requires repeated "error corrections." This process takes place throughout the StepIt exercise program and gradually improves with practice until automatization is reached (37). This beneficial entrainment effect of rhythmic auditory cueing has been suggested to involve a variety of mechanisms: (i) supplement of sensory deficits present in fall-prone persons, (ii) neurophysiological changes, (iii) enhancement of auditory imagery, (iv) reduction in musculoskeletal activation variability, and (v) reduction of cognitive-motor interference (38).

# DIFFICULTY LEVEL PRINCIPLES

One of the basic theories of training is that in general a skill will improve if it is practiced. Practicing a motor task that is varied along some task dimension is referred to as variable practice and it is considered to be a method that enhances the ability to transfer the learned task to a novel variation of the task or to a new environment (39). This happens because, similar to the mechanism just described for the RAS technique, the continuous attempt to adjust to altered conditions creates a trial-and-error mechanism that ultimately maximizes retention (40).

Progression during the StepIt exercise program is achieved by systematically manipulating four training principles, two addressing motor load and two addressing cognitive load. Motor load is increased gradually by increasing speed of execution and complexity of the stepping pattern. Cognitive load is increased gradually by extending the length of the stepping patterns and by adding additional cognitive/motor tasks (**Figure 1**). All four principles (execution speed, pattern complexity, pattern length and execution in [Dual-Task (DT)/Multi-Task (MT)], based on which progression was achieved, have been found to be important components of effective balance training/falls prevention exercise programs in former single studies (14, 19, 41).

# Execution Speed

In order to improve the ability to step quickly, the tempo given by the metronome on which participants have to execute the presented stepping patterns is increased across the training sessions. The ability to step quickly is a critical factor in avoiding a fall (42). Unfortunately, many community-dwelling older adults walk slower than the optimal speed for functioning optimally in the everyday life (e.g., to cross a street safely as a pedestrian).

Stepping exercise programs for fall-prone individuals who often show sensory, neuromuscular, and cardiopulmonary deficits may be enhanced by a variable practice schedule as manipulation of intensity by increasing movement speed increases demands in these systems which in turn may after repeated training result in an increase in physiologic reserve. Furthermore, stepping in high speeds transfers to a functional outcome as it simulates the actions that need to be taken in real life to avoid a fall. Indeed, high-intensity variable stepping training has shown to improve gait speed as well as other gait kinematics in frail older adults (43) as well as in neurological patients (44).

Regarding the neural mechanisms underlying the relevance of high speed training, studies suggest that plasticity associated with locomotor adaptation is speed-specific meaning that there are some neural networks for controlling locomotion that are recruited specifically for fast vs. slow walking (45). Recently, this is has been shown on a muscular level; different locomotor modules are recruited in different walking speeds to presumably meet additional functional demands with a speed increase (46). This separation in the control of fast and slow walking could possibly occur in the cerebellum, which is responsible for locomotor adaptation (45).

# Pattern Complexity

To avoid a trip or a slip in real life, besides speed, direction, and amplitude of the compensatory step is also crucial. Therefore, across sessions, complexity is manipulated by starting with patterns including only shorter, forward/lateral steps and progressing to wider, backward/oblique steps.

Stepping quickly in different directions and/or changing travel direction while walking, often referred to as steering (47) are two very important and also very complex aspects of balance and mobility of older adults. The majority of falls in old age happen while walking (48) due to a trip (40–60%) or a slip (10–15%) (49). Falls are 44% forward, 41% backward, and 33% concurrently or solely sideward (50). Studies have shown that in order to prevent falling, older adults, and especially fallers, prefer to modify their base of support (make a longer step to extend their margin of stability) by using stepping rather than feet-inplace strategies (grabbing nearby objects or rapidly extending

the arms to prevent falling) (12, 51). These compensatory steps are often a response to both forward (82%) as well as sideward (70%) falls (50), indicating that the ability to take rapid steps in various directions could potentially prevent many falls. Studies have found that there is an age-related decrease of stepping speed in the forward, backward and sideward direction (52) and that this is more pronounced for fallers than non-fallers (53).

Most of the existing stepping exercise programs include forward and lateral steps. This is important as age-related deficits in lateral balance recovery are associated with fall incidence (8, 54) and most hip fractures occur after lateral falls (55), often because older adults tend to collide their legs against each other trying to control lateral balance (56). However, backward walking/stepping is essential for functional ambulation as it is required for many common activities of daily living like opening a door, stepping back from a curb to avoid a fast-moving vehicle etc. and backward balance loss often causes serious injuries as well (57) because backward falls are harder to prevent than forward and lateral ones (58). In fact, fallers show greater deficits in backward walking than non-fallers (59) and backward gait velocity identifies fallers more accurately than forward velocity (60). Backward walking has been applied effectively in several studies mostly with patient groups (61, 62).

Like during any motor learning task, multidirectional stepping causes structural and functional changes at the central nervous system. In order to achieve greater motor learning (in our case safer, smoother and more efficient multidirectional stepping performance) it is important to increase complexity and variability (63), which results in a flexible and adaptable motor system (63). In the StepIt exercise program this is achieved through the execution of steps in all directions and with different amplitudes, something which increases motor exploration and requires high inter- and intra-limb coordination, both of which are fundamental to human motor control, especially for foot trajectory and foot placement accuracy.

Finally, considering the limited generalization to other locomotor tasks (64) (stepping in different directions than the trained ones), falls prevention stepping training programs even for high-functioning older adults should incorporate multidirectional steps and also focus on the adaptation of step length (65) in order to be effective.

# Pattern Length

As mentioned earlier, in the StepIt exercise program, participants are presented with stepping patterns that they have to memorize and then execute on a grid-like mat. The number of steps inside the stepping patterns is progressively increased throughout the course of the intervention. Essentially, this is an n-back (1-back) visuospatial working memory task.

Working memory is fundamental to human cognition as it is responsible for maintaining and manipulating goalrelevant information for the performance of complex tasks (66). Working memory training has been shown to improve working memory capacity even in older adults, as a result of cognitive plasticity. Since cognitive plasticity starts when the environmental demands are higher than the demands the cognitive system usually faces (67), it is important that the difficulty level remains challenging (always a little higher than the routine performance level). Indeed, studies have demonstrated that training with variability in working memory task demands leads to greater transfer effects than training with constant task demands (68).

Although the exact procedures of how the neural systems that support working memory are altered through intensive training are not fully elucidated a recent neuroimaging study found that intensive working memory training produces functional changes in large-scale front parietal networks (69).

# Dual-/Multi-Tasking

Besides increasing the amount of step positions to be memorized, another way to increase cognitive load is to step while conducting additional motor or cognitive tasks.

It is well-documented that walking under cognitive load, a situation that is extremely common in everyday life, poses high motor and attentional demands on the central nervous system. Based on the capacity sharing theory, performing an additional task while walking alters gait performance and/or the execution of the secondary/cognitive task (70). Studies have extensively reported benefits of (esp. cognitive-motor) dual-task training on cognitive, motor (71) as well as dual-task performance (17, 72). Especially regarding dual-tasking and RAS, a process applied at the later stages of the StepIt exercise program, it is suggested that dual-tasking with rhythmic auditory cueing frees up cognitive resources (73).

Overall, a combination of physical exercise, sensorimotor stimulation and cognitive engagement may facilitate neurophysiological changes that contribute to cognitive improvement. A very recent review (74) has summarized the neurophysiological mechanisms underlying cognitive improvements following motor-cognitive dual-task training: dual-task training may stimulate similar neurobiological processes which produce a synergistic response: Common increase of cerebral blood flow as well as angiogenesis in the cortex and cerebellum induced from both physical exercise as well as cognitive training.

# TRAINING PROTOCOL

In this section, an exemplary protocol for 9 weeks of the groupbased version of the StepIt exercise program aiming to improve physical, cognitive and psychological fall risk factors will be presented. The difficulty level described here is appropriate for older adults without major mobility or cognitive impairments, thus it can be used as a prevention (rather than rehabilitation) tool. Further recommendations regarding possible adjustments of this protocol to fit other populations, purposes and settings are presented in the "Design Recommendations" section.

# Session Structure and Equipment

The first 10 min of each session is a warm-up phase. First, participants can walk freely across the room at their own preferred speed while performing different joint-mobility exercises such as: arm circles forwards and backwards, walk on tip-toes, walk with butt kicks, walk with alternating knee lifting, side shuffle walking etc. Afterwards, rubber mats (one for each participant) are spread on the floor across the whole room. Their

TABLE 1 | Number of patterns to be practiced for each condition.


*RF, right foot; LF, left foot; BF, both feet.*

size should be approximately 90 × 90 cm and they should be made from extra non-slip yoga mats. They are divided into 9 equal squares (30 × 30 cm) with a 3 × 3 pattern (**Figure 2**). While participants are walking across the room, the instructor calls random numbers (from 1 to 9) and participants have to find the closest square on one of the mats with the number called by the instructor and step on it.

For the session's main phase (45 min) each participant stands behind their mat and starts performing the instructed stepping patterns. All sessions should be supervised by preferably two instructors. The main instructor (MI) leads the sessions and the assistant instructor (AI) stands at the back and offers help/correction when needed.

The MI demonstrates and explains the patterns first on a flipchart, which ensures good visibility also for participants standing at the back, and if needed also performs the pattern on his/her mat. Participants are then given time to practice the pattern until they can memorize and reproduce it without looking at the MI/flipchart. During this time both instructors walk around and provide help in case some participants need it. After a couple of rounds of practicing in their self-selected speed, a metronome is switched on and participants have to execute the stepping patterns in the given pace. Each session includes stepping patterns executed with one leg (left leg remained in the predefined start square and right leg stepped on the given numbered squares, then the same (i.e., mirrored) pattern was executed with the right leg remaining at the start square and the left stepped on the given numbered squares) as well as patterns with both legs moving in turns (to avoid confusion, execution should always with the same leg).

Moving to the new pattern happens once ca. 80% of the participants master the current pattern. In case some participants manage to learn the sequence much earlier than others, in order to avoid longer periods of inactivity and boredom, they can be approached from either the MI or the AI and asked to recall and repeat the patterns that have been taught so far in the session or to learn new patterns (with the same difficulty level) which the instructors come up with spontaneously.

Around 306 patterns should be developed for 18 training sessions. The number of patterns decreases every 2 weeks (because the number of steps within each pattern increases every 2 weeks) in order to have enough time to practice the patterns within the 45 min stepping phase. For a detailed description of the number of patterns for each condition and week, please see **Table 1**.

In case this concept is used as an intervention with fixed duration, the exact content of each session (including the exact stepping patterns, the pace as well as the starting point/square for each of the 18 sessions) should be predefined and put together in a manual (see **Supplementary Material** for some exemplary training plans of week 1, 4, and 7) to be followed by all MIs.

The last 5 min of each session is a cool-down phase, where various stretching exercises in a standing position are performed.

# Progression

The difficulty level is gradually increased within each session (by using first one leg and then both for the execution of the stepping patterns) as well as between sessions.

Between sessions, the difficulty level is increased based on four principles (P); two of them addressing the increase of motor (M) load and two of them addressing the increase of cognitive (C) load (**Table 2**):


Regarding the direction of feet placement, for the first 3 weeks the patterns require forward, sideward and backward steps. In week

#### TABLE 2 | Difficulty level increase between sessions.


*BMP, beats per minute; RF, right foot; LF, left foot; BF, both feet.*

4–6 the patterns include much longer steps which require to skip the middle line of the squares (for example stepping from 7 to 2 etc.), and finally the last 3 weeks of the intervention, the patterns also require steps with crossings of the legs (for example from 6 to 5 or from 6 to 2).

Pacing of the steps is increased by 2 BPM at almost every session both for the RF/LF as well as the BF patterns.

Initially, the patterns include 3 steps and every 1 or 2 weeks the number of steps included in each pattern increases by 1 step, resulting in week 8–9 when patterns consist of 8 steps.

The last way to increase the difficulty level is the progress from single-task (only reproducing the stepping patterns) in week 1–3, to dual tasking in week 4–6, when participants will have to also conduct an additional motor (e.g., balance an object on their hands, clap hands front and behind their backs, browsing magazines, unbutton their shirt, hold ball above their heads/behind their backs etc.) or cognitive task (targeting two executive functions: verbal fluency (naming words associated with certain given words, naming objects, animals, vegetables, professions that start or end with a certain letter etc.) and memory/concentration (counting onwards, serial sevens, serial threes, spelling words backwards etc.) while executing the stepping pattern. Finally, in the last 3 weeks, while executing the stepping pattern, participants have to conduct both a motor and a cognitive task simultaneously (multitasking). **Table 3** gives an overview of the content of all 18 training sessions.

# DESIGN RECOMMENDATIONS

As mentioned earlier, the presented concept can be used as a framework for the development of further prevention and/or rehabilitation stepping exercise programs. In this section, we present recommendations on how this exercise program can be adjusted to fit different purposes, populations and settings. Being a form of highly specific balance training, this stepping training program can be applied to either healthy older adults aiming to improve their overall balance ability, as well as to neurological patients aiming to improve their stepping capacity and/or to any other fall-prone target group aiming to prevent falling.

It can be delivered either as a group-based training program, using mats and at least one instructor (possibly also two, depending on size and homogeneity of the group) or as a homebased training program using an ICT-based solution (e.g., in form of an exergame), which would not necessarily require the presence of an instructor. In clinical settings (e.g., physiotherapy practices or nursing home facilities), it is also possible to offer this mat-based but still in 1:1 training. Depending on the participants' preferences, RAS can be delivered via a metronome, music, or a combination of both. All of them have proven to be equally effective.

There are several ways to make training more variable or more challenging for example by altering material, position and size of the mat. Using a very thin mat (or even "drawing" the 9-square grid on the floor) is suitable for beginners or fall-prone older adults. At later stages or for relatively healthy older adults, thicker/softer mats can be used or the thin mats can be placed on unstable surfaces (e.g., on the Posturomed balance system), which will increase proprioception and neuromuscular demands.

In order to also train step length, the size of the mat can easily be personalized using the Pythagoras theorem by measuring maximum step length (MSL) with the maximum step length test (75) (or alternatively leg length or height). Beginners or fallprone older adults can then start by training with mats sized to fit ca. 60% of their MSL and then progress up until 80% of their MSL. If personalizing mat size is not possible, it is possible to have three standard sizes (Small: 85 × 85 cm, Normal: 90 × 90 cm and Big: 95 × 95 cm) and use them based on participant's height and/or training level.

Regarding tempo and dosage, a recent meta-analysis on the effects of RAS in Parkinson's patients (24) found that training using this method should include tempo variations ±10% with respect to the preferred cadence, for a minimal period of 20– 45 min per day, for at least 3–5 days per week. Although these recommendations cannot be used as is for a stepping training because they are based on gait studies, they can be used as a basis and adjusted to fit needs and fitness levels of other kinds of neurological patients (stroke, MS) as well as healthy older adults.

Furthermore, content/focus of training can also be adjusted to fit needs of different target groups. Participants with memory complaints or mild cognitive impairment can focus on the cognitive load (pattern length and dual-/multi-tasking) whereas stroke/Parkinson's patients can focus on the motor load (execution speed and pattern complexity).

# CONCLUSIONS

To reduce the burden of falls in older adults falls prevention exercise programs that apply new research evidence into practice


TABLE 3 | Detailed overview of the content of all 18 training sessions.

*RF, right foot; LF, left foot; BF, both feet; BPM, beats per minute; P, patterns; FSB, forward, sideward (lateral), backward steps; SML, skipping middle line; OS, oblique steps.*

need to be developed. The ability to execute movements varying in speed, amplitude, complexity and additional cognitive load is critical for preventing falls (76) and thus stepping training programs that incorporate such aspects have resulted in substantial reductions of falls (15).

We have proposed a framework for a stepping training program, reviewed the theory and evidence underlying it and described in detail the implementation of an exemplary 9 week, group-based stepping exercise program applying the suggested concept.

The exemplary training plan does not require costly equipment and has potential for high adherence levels, taking into account the social aspect of physical activity.

However, being a group-based, it also has certain limitations such as that, the use of relevant secondary cognitive tasks (e.g., visual search tasks) is not possible. Moreover, adjustments of the difficulty level cannot happen according to each participant's current personal level. However, recent evidence suggests that even just exposure to varying levels of task difficulty is sufficient for inducing training gains (77), making individually-tailored training not always necessary.

The elements as well as progression rate of the exemplary training program can be easily modified to fit the needs of different samples (e.g., healthy older adults, patient groups), which makes it a useful tool for the development of further stepping exercise programs. Pilot feasibility studies are needed to test its feasibility (safety/adverse outcomes and enjoyment/adherence). If proven feasible, its effectiveness to improve fall risk factors is then to be tested via randomized controlled trials. Due to its multi-domain approach, besides the improvement of stepping capacity, further expected outcomes include the improvement of further physical, cognitive and psychological risk factors. Thus, further studies using this exercise regimen or modified versions of it are encouraged.

# DATA AVAILABILITY STATEMENT

All data for this study have been provided in the article/**Supplementary Material**.

# AUTHOR CONTRIBUTIONS

EG developed the original research idea, designed the study, and was the major contributor in writing this manuscript. TM and WZ contributed significant components to the study design. All authors critically revised and approved the manuscript.

# FUNDING

This research was part of the project MyAHA: my Active & Healthy Aging, which was funded by the European Union's

# REFERENCES


Horizon 2020 research and innovation program under grant agreement Nr 689582.

# ACKNOWLEDGMENTS

We would like to gratefully acknowledge the assistance of Jessica Coenen in generation of the stepping patterns and overall study preparation.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmed. 2019.00318/full#supplementary-material


community-dwelling older women. Arch Gerontol Geriatr. (2017) 73:240–7. doi: 10.1016/j.archger.2017.07.011


**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 Giannouli, Morat and Zijlstra. 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.

# Effect of Underwater Treadmill Gait Training With Water-Jet Resistance on Balance and Gait Ability in Patients With Chronic Stroke: A Randomized Controlled Pilot Trial

#### Chae-gil Lim\*

*Department of Physical Therapy, College of Health Science, Gachon University, Incheon, South Korea*

Objective: The purpose of this study was to determine the effects of underwater treadmill gait training with water-jet resistance and underwater treadmill gait training with ankle weights on balance and gait abilities in chronic stroke patients.

#### Edited by:

*Helena Blumen, Albert Einstein College of Medicine, United States*

#### Reviewed by:

*Alessandro Giustini, Consultant, Italy Iman Akef Khowailed, University of St. Augustine for Health Sciences, United States*

> \*Correspondence: *Chae-gil Lim jgyim@gachon.ac.kr*

#### Specialty section:

*This article was submitted to Neurorehabilitation, a section of the journal Frontiers in Neurology*

Received: *13 February 2019* Accepted: *08 November 2019* Published: *12 February 2020*

#### Citation:

*Lim C (2020) Effect of Underwater Treadmill Gait Training With Water-Jet Resistance on Balance and Gait Ability in Patients With Chronic Stroke: A Randomized Controlled Pilot Trial. Front. Neurol. 10:1246. doi: 10.3389/fneur.2019.01246* Methods: Twenty-two inpatients and outpatients with stroke-induced impairments were randomly assigned into two groups: an underwater treadmill gait training with water-jet resistance group (*n* = 11) and an underwater treadmill gait training with ankle weights group (*n* = 11). Participants received conventional physical therapy for 30 min and underwater treadmill gait training with water-jet resistance or ankle weights for 30 min. Intervention was performed 5 days a week for 4 weeks. The Balance System SD was used to assess static and dynamic balance. The GAITRite system was used to assess gait velocity, cadence, step length, stride length, and swing phase. All measurements were performed at the beginning of the study and 4 weeks after the intervention.

Results: The water-jet resistance group and ankle weights group showed significant improvement in static balance (*P* < 0.00 vs. *P* = 0.01), dynamic balance (*P* < 0.00 vs. *P* = 0.57), gait velocity (*P* < 0.00 vs. *P* = 0.037), cadence (*P* < 0.00 vs. *P* = 0.001), step length (*P* < 0.00 vs. *P* = 0.003), stride length (*P* < 0.00 vs. *P* = 0.023), and swing phase (*P* < 0.00 vs. *P* < 0.00). However, the static and dynamic balance ability score (*P* < 0.00), gait velocity (*P* < 0.00), cadence (*P* < 0.00), step length (*P* < 0.00), stride length (*P* < 0.00), and swing phase (*P* = 0.023) in the group that received underwater treadmill gait training with water-jet resistance improved more than in the group that received underwater treadmill gait training with ankle weights.

Conclusions: Our results demonstrated that underwater treadmill gait training with water-jet resistance is effective in improving static and dynamic balance as well as gait abilities in chronic stroke patients. Thus, training using underwater treadmill gait training with water-jet resistance may be useful in facilitating active rehabilitation in chronic stroke patients.

Keywords: stroke, underwater treadmill gait training with water-jet resistance, balance, gait ability, hemiplegia

# INTRODUCTION

Stroke patients have impaired walking ability due to decreased balance and are prone to falls (1, 2). Falls cause injuries, resulting in decreased mobility, fear of falls, difficulty in resuming activities of daily living, and impaired balance and gait ability (3–6).

Weight-bearing treadmill training using a task-oriented approach is designed to improve balance and gait ability in hemiplegic patients (7–9). However, to provide treadmill training for stroke patients, it is necessary to promote a sense of stability. Therefore, a safety harness may be necessary for walking in water (10, 11).

Underwater treadmill training by using the buoyancy can be promoted to gait disturbance of stroke patients rather than land-based treadmill training because of effectively reducing body weight. In addition, water resistance can increase energy consumption by combining aerobic exercise and resistance exercise (12). Also, Underwater treadmill training reduces the burden on weight-bearing joints and the risk of falling, and provides resistance during movement in multiple directions (13). Thus, underwater aerobic exercise that reproduces the action of walking and running is increasingly popular (14).

Even when walking ability is restored in stroke patients, hip and knee joint flexion and ankle dorsiflexion are reduced (15). As a result, stroke patients have difficulty in dealing with obstacles more complicated than gait or stair climbing, and require proper flexion of the lower limb and dorsiflexion of the foot from the floor (16).

The authors previously reported that when a sandbag equivalent to 5% of body weight was worn on the affected ankle during weight-bearing treadmill training, the walking speed was improved, and the ratio of the swing phase was significantly increased in the hemiparetic lower limb and increased in the nonhemiparetic lower limb (17). A recent study reported that when weight is applied to the affected ankle during underwater walking, the support ratio increases and stability and symmetry increase (18). Underwater backward and forward treadmill walking against a variable-speed current reportedly led to increased step length and frequency (19).

Previous studies on treadmill training used resistance to improve walking ability by applying external weights to the lower extremities in chronic stroke patients. However, studies on underwater treadmill training in stroke patients using waterjet resistance or external weights only included a short training period or a cross-sectional study design.

This study compared the effect of water-jet resistance and ankle load training on balance and walking ability in stroke patients to identify intervention methods that can be applied in clinical practice.

# METHODOLOGY

# Setting, Study Design, and Participants

This study was a two-arm, parallel, and randomized controlled pilot trial with concealed allocation, and researcher and assistants blinding. All procedure of this study was approved by the Institutional Review Board of Gachon University (IRB No: 1044396-201612-HR-096-02) and registered at Clinical Research Information Service (CRiS), Republic of Korea (KCT0003576) and all participants signed an informed consent prior to beginning the study. In addition, this study conforms to all CONSORT guidelines as much as possible.

# Experimental Procedure

Patients with a history of stroke were recruited from a rehabilitation hospital in Incheon, Republic of Korea. Participants were enrolled in this study if 6 months or more had passed since onset of a unilateral hemispheric first stroke; the ability to walk at least 10 m was required (regardless of need for assistance), and the required minimental state examination (MMSE) score was at least 24. All patients were diagnosed with a chronic stroke as defined by computed tomography or magnetic resonance imaging. Patients with a cognitive, visual, or cardiorespiratory disorder (including cardiac pacemaker placement, heart failure, and arrhythmia), orthopedic intervention, hydrophobia, skin disease, and undergoing botulinum toxin injections within the prior year were excluded. Also, Patients with a pulse rate ≥100 beats per minute (bpm), a systolic blood pressure ≥180 mmHg, and a diastolic blood pressure ≥100 mmHg were excluded.

# Randomization and Masking

Twenty-two inpatients and outpatients with stroke-induced impairments were randomly assigned into two groups: an underwater treadmill gait training with water-jet resistance group (n = 11) and an underwater treadmill gait training with ankle weights group (n = 11). Randomization was intended to minimize an order effect. Baseline measurements of abilities were performed prior to randomization. Subsequently, each participant was allocated to one of the two groups via allocation codes included in consecutively numbered, sealed, opaque envelopes. Simple randomization was conducted using Microsoft Excel for Windows software (Microsoft Corporation, Redmond, WA, USA) by a researcher who was not involved in participant recruitment. To ensure masking, protocols and intervention order were not revealed to participants or clinical evaluators.

# Interventions Procedures

Hydrotherapy protocols included shallow water flowing in various combinations of buoyancy, hydrostatic pressure, turbulence, and resistance (depending on water level and treadmill speed). A higher water levels provide higher buoyancy and hydrostatic pressure, but higher velocities produce more turbulence and resistance (20). The slope of the underwater treadmill was horizontal, and the height of the water was the height of the xiphoid process (21). The temperature of water was 34◦C, which is suitable for functional training while minimizing the physiological changes and stabilizing the deep body temperature (18). The treatment room temperature was set at 26◦C to reduce the difference between room temperature and water temperature (22).

Participants received a conventional physical therapy program for 30 min and gait training on an underwater treadmill (Focus, Hydro Physio, Nottingham, Nottinghamshire, UK), with waterjet resistance at 442 L/min against the anterior the shin, or wore an ankle weight equivalent to 5% of body weight for 30 min (**Figure 1**). The initial speed of this program based on the study that treadmill training performed at the fastest speed that stroke patients can perform was significantly increased in stride length, walking speed, and step length than that of fixed speed treadmill training. The training was performed at the maximum speed that the patients could do within the range of the walking speed was 1–4 m/s. At this time, when the patient was breathing or experiencing difficulty during training, the maximum speed was reduced by one step (23). We applied one of these methods to the patients considering of their ability. In all interventions and assessments, if the patient complained of discomfort, immediately stop and take a rest. The intervention was performed 5 days a week for 4 weeks (17).

# Outcome Measures

The general characteristics were collected through file audit and self-report. The primary outcomes were the Static and dynamic balance abilities by measured with the Balance System SD (Biodex Medical Systems, Inc., Shirley, NY, USA). The secondary outcomes were the changes in gait abilities. The GAITRite system (CIR Systems, Inc., PA, USA) was used to assess gait velocity, cadence, step length, stride length, and swing phase. All measurements were performed at the beginning of the study and 4 weeks after the intervention.

# Sample Size Estimation

We estimated a minimum acceptable sample size of 21 patients per group to achieve a power of 0.8 with a significance level (α) of 0.05 using a 1-sided, 2-sample t-test (G∗Power 3.1) but we realistic enrolled 30 patient and was referenced that 21 patients would be necessary based on an inter-groups difference in endurance training with hydrotherapy in a previous trial (12).

# Data Analysis

Statistical analyses were performed using SPSS for Windows Version 18.0 (IBM, Armonk, NY, USA). The Kolmogorov– Smirnov test was used to determine the normality of parameter distributions. For a normal distribution, continuous data were expressed as the mean ± standard deviation, and as a percentage for categorical data; parametric tests such as an independentsamples t-test or the χ 2 test were used to compare baseline characteristics of the two groups. A paired t-test was used for within-group comparisons, and an independent t-test was used for between-group comparisons. The level of significance was set at α = 0.05.

# RESULTS

Between Jun 2017 and Dec 2017, a total of 30 patients were admitted to the rehabilitation center, 22 fulfilled the inclusion criteria. Participants were randomly assigned to an underwater treadmill gait training with water-jet resistance group (n = 11) or an underwater treadmill gait training with ankle weights group (n = 11). All 22 participants completed the study (**Figure 2**). General baseline characteristics are shown in **Table 1**. Recorded

characteristics included gender, age, height, weight, lesion side, post-stroke duration, and MMSE scores. The mean ± SD age of the patients was 51.90 ± 9.62 years, and post-stroke duration was 9.63 ± 2.61 months. Baseline demographic characteristics such as gender (males/females, 7/4 vs. 9/2), age (54.63 ± 7.25 vs. 49.18 ± 12.00 years), lesion side (right/left, 5/6 vs. 3/8), and post stroke-duration (10.18 ± 2.92 vs. 9.09 ± 2.30 months), and MMSE scores (25.63 ± 1.43 vs. 25.81 ± 1.66 score) were not significantly different between to an underwater treadmill gait training with water-jet resistance group and an underwater treadmill gait training with ankle weights group (P > 0.05).

Both group showed significantly improved static balance. In the water-jet group, the static balance score improved from 1.16 ± 0.32 to 0.49 ± 0.17 score (P < 0.00) and the dynamic balance score improved from 3.57 ± 1.45 to 1.78 ± 0.88 score (P < 0.00). In the ankle weight group, the static balance score improved from 1.10 ± 0.42 to 0.95 ± 0.32 score (P = 0.01) and the dynamic balance score little improved from 3.15 ± 0.80 to 3.09 ± 0.91 score (P = 0.57). However, the change in the static balance ability score in the water- jet group improved more than in the ankle weight group (P = 0.00; effect size = 0.73), and the change in the dynamic balance ability score in the water-jet group improved more than in the ankle weight group (P = 0.00; effect size = 0.76; **Table 2**).

In temporal parameter of gait ability, the gait velocity improved (P < 0.00 vs. P = 0.037) and the spatial parameters TABLE 1 | General characteristics of the two groups by randomization assignment.


*Data are expressed as mean* ± *SD or n (%).*

\**MMSE, mini mental state examination.*

*<sup>a</sup>The P-value was obtained using a* χ 2 *.*

*<sup>b</sup>The P-value was obtained using an independent t-tests.*

TABLE 2 | Changes in static and dynamic balance within each group and between the two groups.


*Data are presented as mean* ± *SD.*

*<sup>a</sup>P* < *0.05. The P-value was obtained using a paired t-test.*

*<sup>b</sup>P* < *0.05. The P-value was obtained using an independent t-test.*

as cadence (P < 0.00 vs. P = 0.001), step length (P < 0.00 vs. P = 0.003), stride length (P < 0.00 vs. P = 0.023), and swing phase (P < 0.00 vs. P < 0.00) values in both groups were increased compared to pre-intervention values. However, the velocity (P = 0.00; effect size = 0.87), cadence (P = 0.00; effect size = 0.91), step length (P = 0.00; effect size = 0.88), stride length (P = 0.00; effect size = 0.95), and swing phase (P = 0.023; effect size = 0.54) values in the water jet group improved more than in the ankle weight group (**Table 3**).

# DISCUSSION

The results demonstrate that a 4 week underwater treadmill gait training program with water-jet resistance had a beneficial effect on static and dynamic balance ability scores, gait velocity, step length, stride length, and swing phase in patients with chronic stroke when compared with patients who underwater treadmill gait training with ankle weights. These results are of special


TABLE 3 | Changes in gait ability within each group and between the two groups.

*Data are presented as mean* ± *SD.*

*<sup>a</sup>P* < *0.05. The P-value was obtained using a paired t-test.*

*<sup>b</sup>P* < *0.05. The P-value was obtained using an independent t-test.*

interest because little evidence is available on the effect of aquatic intervention on balance and gait ability in stroke patients.

The typical gait pattern in stroke patients is characterized by unstable weight shifting to the affected side during walking, resulting in asymmetric and slow gait. In a study comparing postures in stroke patients and the elderly, it was reported that the stroke patients had an increased risk of falling because they exhibited asymmetric posture and greater postural sway than healthy elderly subjects (24). Wing et al. (25) suggested that whole-body intensive rehabilitation (3–6 h/day, 4–5 days/week, for ≥2 weeks) is effective for improving postural balance and functional mobility in stroke patients. To address these findings, water has been used in rehabilitation. Water provides a safe environment for patients by reducing the risk of falls. In addition, warm water can exert a therapeutic effect by alleviating pain or spasticity. Warmth can increase skin temperature, dilate peripheral blood vessels, increase blood supply, accelerate muscle relaxation, alleviate pain or muscle cramps, and improve balance (26).

Berger et al. (27) reported that 6 weeks of aquatic therapy improved postural control. Zhu et al. (28) suggested that a relatively short program (4 weeks) of hydrotherapy exercise resulted in a large improvement in a small group (n = 14) of individuals with relatively significant balance and walking dysfunction following a stroke. Mentiplay et al. (29)

demonstrated that the strength of the knee extensors shows a poor-to-moderate correlation with gait velocity. Another study (30) determined that the static strength of lower limb muscles on the paretic side was correlated significantly with gait velocity and cadence in stroke patients. In agreement with a previous study, we found that balance and gait ability improved in chronic stroke patients. Improvements in knee range of motion resulted from repetitive movement in the sagittal plane by the anterior of the shin against water-jet resistance.

In contrast to these studies, Lee et al. (31) reported that a 4 week aquatic treadmill exercise program had a beneficial effect on muscle strength but not motor function and balance in subacute stroke patients, possibly because improvements were observed in all groups with conventional rehabilitation therapy in the subacute phase. In contrast, our study added water-jet resistance and participants had chronic strokes.

We also examined the effects of water-jet resistance on gait abilities. Previous reports showed that treadmill training may be more effective than conventional gait training in improving gait parameters such as functional ambulation, stride length, percentage of paretic single stance period, and gastrocnemius muscular activity (7). A recent systematic review reported that aquatic therapy improves dynamic balance and gait performance in individuals with neurological disorders, especially those with multiple sclerosis, Parkinson's disease, and stroke (32). We found that although velocity, cadence, step length, stride length, and swing phase in both groups tended to improve after 4 weeks of aquatic therapy, the difference in cadence between groups was insignificant. These results reflect the fact that water-jet resistance was applied against all areas below the knee, but that ankle weights seemed to play a role in limiting the effect of other areas and prevented ankle floating.

In our study, we had clinically significant effect sizes and high power for detecting statistically significant changes in the velocity (effect size = 0.87), cadence (effect size = 0.91), step length (effect size = 0.88), stride length (effect size = 0.95), and swing phase (effect size = 0.54).

This study had several limitations. First, the water jet resistance and ankle weights were applied at different locations (anterior of shin and ankle) and the same load (442 L/min) was applied in all subjects, regardless of sex, age, height, and weight. Although various levels of difficulty were provided according to the general characteristics of the subjects and study parameters were changed according to individual balance and gait ability, there was a limit in providing diversity. This led to some difficulty in adapting training for each subject, using the same applications. Therefore, further studies will need to address these limitations. Second, the number of subjects was small, and the subjects were

### REFERENCES

1. Ikai T, Kamikubo T, Takehara I, Nishi M, Miyano S. Dynamic postural control in patients with hemiparesis. Am J Phys Med Rehabil. (2003) 82:463–9; quiz: 470–62, 484. doi: 10.1097/01.PHM.0000069192.32 183.A7

limited to those with chronic stroke, which limited the ability to generalize the results of underwater treadmill gait training with water-jet resistance. Finally, we could not test lower limb strength. Based on these limitations, a more detailed training method should be used to identify the effects of this study.

# CONCLUSION

The present study showed that a 4 week underwater treadmill gait training program with water-jet resistance had a beneficial effect on static and dynamic balance ability scores, gait velocity, cadence, step length, stride length, and swing phase in patients with chronic stroke when compared with patients who underwater treadmill gait training with ankle weights. Future studies should assess results in a larger cohort of subjects with various durations since stroke onset.

# DATA AVAILABILITY STATEMENT

The data used to support the findings of this study are available from the corresponding author upon request.

# ETHICS STATEMENT

All procedure of this study was approved by the Institutional Review Board of Gachon University (IRB No: 1044396-201612- HR-096-02).

# AUTHOR CONTRIBUTIONS

CL makes substantial contributions to conception and design, acquisition of data (two assistant researcher help), data analyzing, substantial contributions to interpreting data, drafting the article and revising it critically for important intellectual content, and final approval of the version to be submitted.

# ACKNOWLEDGMENTS

The author thanks Physical therapist Mr. Je Yong Park for help with the study.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fneur. 2019.01246/full#supplementary-material


and forward treadmill walking in water. Gait Posture. (2009) 29:199–203. doi: 10.1016/j.gaitpost.2008.08.008


**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.

Copyright © 2020 Lim. 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.

digital media

of impactful research

article's readership