The Role of Emotion Regulation in Eating Disorders: A Network Meta-Analysis Approach

Background Previous theoretical models and reviews have documented a strong connection between emotion dysregulation eating disorder (ED) psychopathology among the general and clinical populations. The aim of this review was to build on this previous work by conducting a network meta-analysis to explore associations between adaptive and maladaptive emotion regulation strategies and ED psychopathology trans-diagnostically across the ED spectrum to identify areas of emotion dysregulation that have the strongest association with symptomatology. Methodology A total of 104 studies were included in the meta-analysis and correlation coefficient representing the associations between specific emotion regulation strategies and ED symptomatology were extracted. We ran a Bayesian random effects network meta-analysis and the initial network was well-connected with each emotion regulation strategy being linked to at least one other strategy. We also conducted a network meta-regression to explore whether between-study differences in body mass index (BMI), age, and whether the sample consisted of solely female participants explained any possible network inconsistency. Results The network meta-analysis revealed that ruminations and non-acceptance of emotions were most closely associated with ED psychopathology. There was no significant network inconsistency but two comparisons approached significance and thus meta-regressions were conducted. The meta-regressions revealed a significant effect of BMI such that the associations between different emotion regulation strategies and ED symptomatology were weaker among those with low BMI. Discussion The present findings build on previous work and highlight the role of rumination and difficulties with accepting emotions as key emotion regulation difficulties in EDs. Additionally, the finding that the associations were weaker among ED patients with low BMI may point toward a complex relationship between ED behaviors and emotion regulation. Taken together, our findings call for interventions that target emotion regulation, specifically rumination and difficulties accepting emotions, in the treatment of EDs. Systematic Review Registration https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42021249996, PROSPERO, identifier: CRD42021249996.


INTRODUCTION
Emotion regulation refers to strategies used to identify, initiate and modify the course of emotions (1)(2)(3). Emotion regulation strategies can be automated or controlled methods used to cope with or modify the external expression and/or internal experience when the emotional state has unwanted impact on a desired outcome (1)(2)(3). For example, a young person might suppress their feelings of sadness when moving away from their childhood home in an attempt to not upset their parents thus reducing the external expression of negative emotions. They might then remind themselves they can still visit their parents in an attempt to gain more perspective, reappraise, and reduce the internal experience of sadness. A theoretical framework based on the emotion regulation process model has proposed that emotion regulation strategies can be broadly divided into adaptive and maladaptive strategies (4)(5)(6)(7). The adaptive strategies include reappraisal, active problem solving, and mindfulness strategies based on awareness and acceptance of emotions, while the maladaptive strategies include rumination, avoidance, and suppression of emotions (4)(5)(6). Although the maladaptive strategies can be useful in some scenarios, this framework focuses on dispositional emotion regulation. Thus, for a given strategy to be considered adaptive is needs to be effective and helpful across contexts (6). Excessive reliance on maladaptive strategies has been found to be linked to anxiety symptoms and to impact treatment response in anxiety disorders (4,5). Similarly, people with eating disorders (EDs) have been reported to rely more on maladaptive than adaptive emotion regulation strategies (8)(9)(10). Thus, this review will focus on further investigating the associations between different adaptive and maladaptive emotion regulation strategies an ED psychopathology.
Over the years there has been a great deal of interest in exploring emotion regulation within EDs; it has been proposed that difficulties in emotion regulation is one of the social-emotional factors contributing to the development and maintenance of disordered eating (9,(11)(12)(13). Large scale systematic reviews and meta-analyses have reported links between ED symptomatology and dispositional tendency toward maladaptive emotion regulation and emotion dysregulation among people with anorexia nervosa (AN), bulimia nervosa (BN), and binge eating disorder (BED), both when assessed under laboratory conditions and under naturalistic conditions in studies using ecological momentary assessments (EMAs) (9,11,13). Recently a meta-analytic review by Prefit et al. (12) examined correlations between the use of various emotion regulation strategies and ED psychopathology across all ED diagnoses. The authors focused on examining individual adaptive strategies, including emotional awareness, clarity about emotions, acceptance of emotions, cognitive reappraisal, and problem solving, as well as maladaptive strategies, such as avoidance, rumination and suppression of emotions. The findings showed that across the board maladaptive emotion regulation strategies were positively associated with ED symptoms while the use of adaptive strategies was negatively associated with ED psychopathology. Interestingly, this was the case across ED diagnoses and the authors found no significant evidence of group differences in the analyses. Additionally, no particular strategy came out as more closely associated with ED psychopathology, instead it appeared that all adaptive strategies were equally negatively associated and all maladaptive strategies were equally positively associated with ED symptomatology.
Although the findings from previous reviews highlight the interesting associations between emotion regulation difficulties and ED psychopathology, whether there is a particular aspect of adaptive or maladaptive emotion regulation that has a most important role in EDs remains unclear. This question could be approached using the network meta-analysis technique, which explores comparative associations of several emotion regulation strategies by utilizing existing direct evidence to estimate indirect evidence. This approach has been previously used to compare different interventions to identify most effective treatments (14)(15)(16)(17). More recently the network meta-analysis approach has also been used to compare associations in the field of educations (18). The authors demonstrated a method of applying network meta-analysis to investigate which test was the most closely associated with pupils reading fluency, in an attempt to identify the most useful reading test to be used in school setting. Such a methods could also be applied to emotion regulation in EDs to identify if there is a particular strategy that is most closely associated with ED symptomatology. This approach could inform the development of novel interventions for EDs rather than aiming to generally increase the use of adaptive strategies and reduce reliance on maladaptive strategies.
The aim of this meta-analytic review was to build on and synthesize previous literature by comparing findings from previous work examining the association between different aspects of adaptive and maladaptive emotion regulation and ED psychopathology among people with a diagnosis of an ED. Thus, this review aimed to build on the recent work by Prefit et al. (12) and conduct a network meta-analysis. We compared the use of adaptive emotion regulation strategies, including acceptance of emotions, awareness of emotions, problem solving, and cognitive reappraisal, as well as maladaptive emotion regulation strategies, such as avoidance, rumination, and emotion suppression. The objective was to identify aspects of emotion regulation that have closest association with ED symptomatology to identify useful targets for therapeutic interventions. As previous work has suggested that difficulties in emotion regulation may be trans-diagnostic (12,13), we assessed the associations between the use of different emotion regulation strategies and ED symptomatology across ED diagnoses. Because findings from previous meta-analyses have reported very similar associations between different emotion regulation strategies and ED symptomatology, we did not have a priori expectations regarding which one or more emotion regulation strategies would emerge as being most closely associated with ED psychopathology.

Literature Search
The following keywords were used to search electronic databases including PsychINFO, PsychARTICLES, Medline, Scopus, Pubmed, and Web of Knowledge: ("eating disorder" OR "anorexia nervosa" OR "bulimia nervosa" OR "binge eating disorder") AND ("emotion regulation" OR "emotion dysregulation" OR "affect regulation" OR "affect dysregulation" OR "avoidance" OR "suppression" OR "rumination" OR "problem solving" OR "problem coping" OR "awareness" OR "mindfulness" OR "acceptance" OR "clarity" OR "appraisal" OR "reappraisal"). Where possible filters were used to exclude animal studies, case reports, reviews, and studies not published in English. The initial search was conducted in April 2021 and was updated in August 2021. Bibliographies of included papers and the previous review by Prefit et al. (12) were searched to identify any studies missed in the initial search.

Inclusion Criteria
To be included in the review the studies were requires to meet the following inclusion criteria: (1) include a group of adults or adolescents [10-19 years old (19)] with a diagnosis of an ED, (2) measure some aspect of emotion regulation, (3) measure ED symptomatology, (4) be published in English, and (5) be published in a peer reviewed journal. In the case of longitudinal or interventional studies, only emotion regulation and ED symptomatology measures taken at baseline or pretreatment were included. Additionally, studies that only assessed emotion regulation and/or ED symptomatology at a later stage, post-intervention or at treatment follow-up, were excluded. As we were interested in comparing different emotion regulation strategies, any studies that reported only overall measures and did not assess a specific aspects of emotion regulation were not included. Studies that instructed participants to use specific emotion regulation strategies or assessed regulation of eating or other aspects of food intake, or emotion regulation related behaviors that were to do with food, eating or bodies, such as food avoidance, or body image acceptance, were also excluded. Finally, studies that included young children or clinical populations with diagnoses other than an ED were excluded.

Study Selection
This review was conducted in accordance with the PRISMA guidelines and the flow diagram detailing the literature search and study selection is presented in Figure 1. The initial search and screening based on title and abstract was conducted by J.L along with H.M. The included full text articles were then assessed for eligibility and decisions about final inclusion or exclusion of articles was made initially by the first author. All excluded papers were screened again based on title and abstract by H.M. to ensure no relevant studies were missed. In case uncertainty regarding whether a given study should be included or excluded, the full text article was brought to the rest of the team for consensus meeting and team discussion.

Data Collection and Synthesis
To conduct the network meta-analysis, we took a similar approach to a recent review investigating associations between oral reading fluency and text complexity measures to identify an assessment method most closely correlated with reading fluency (18). Thus, correlation coefficient between an emotion regulation measure and ED symptomatology was extracted from all included papers along with the ED group sample size, the aspect of emotion regulation assessed, and the type of ED symptomatology measure used. Only correlation analysis conducted within the ED group were included. If only correlations for the whole sample which included healthy participants or participants with other diagnoses were reported, the corresponding author was contacted to obtain access to a within ED group correlation coefficient. If a study included more than one ED group separate correlation coefficients for each diagnostic group or one correlation coefficient for the whole mixed ED group were included depending on what was reported. We did not favor separate correlation coefficients as the meta-analysis included all ED diagnoses. The relevant correlation coefficients were not reported in the main text or Supplementary Materials of 267 studies and the corresponding authors were contacted by D.B. to gain access to the data. Correlation coefficients which were not reported in 28 of the included manuscripts, were obtained through personal correspondence. Additional data regarding the age of the ED group, body mass index (BMI), and percentage of female participants was also extracted where available.

Methods Used to Assess Emotion Regulation
We included studies that assessed the use of adaptive emotion regulation strategies, including reappraisal, acceptance of emotions, emotional awareness, and problem solving, as well as the use of maladaptive emotion regulation strategies, such as avoidance, suppression, and rumination (4-7).

Adaptive Emotion Regulation Measures
Thirty-one studies assessed the association between acceptance of emotions and ED symptomatology. Four different methods were used to assess acceptance of emotions (see Table 1 for details), all of which were self-report questionnaire measures. The most commonly used measure was the Difficulties in Emotion Regulation Scale [DERS; (123)] non-acceptance subscale (N = 24). For the purposes of the meta-analysis all correlation coefficients had to have the same direction, with positive correlations indicating that a greater level of ED symptomatology was associated with more difficulties in emotion regulation. Therefore, the correlations that included measures of acceptance of emotions were reversed, while correlation that included measures of non-acceptance or difficulties in accepting emotions were not reversed. This was done to reflect reverse scoring of adaptive emotion regulation, whereby higher scores indicate less acceptance and, thus, this review assessed associations between non-acceptance of emotions and ED symptomatology.
Thirty-eight studies examined associations between emotional awareness and ED symptomatology. Nine different measures were used to assess awareness of emotions ( Table 1), eight of which were self-report questionnaire measures. The most commonly used measure was the DERS lack of emotional awareness subscale (N = 24). One of the measures used, the Levels of Emotional Awareness, is a performance based measure assessing participants' awareness of their own and other's emotions. Although this review builds on the previous meta-analytic review by Prefit et al. (12) we did not include emotional clarity as a separate category. We found that very few measures assessed clarity and those that did used methods that were very similar to ones used to evaluate emotional awareness.   For instance the Attention and Clarity of One's Feelings and the Feelings of Others (ACF) questionnaire is described as measuring emotional self-awareness and includes questions about a persons awareness of and clarity about their own emotions. Therefore, for the sake of simplicity, emotional awareness and clarity were combined and in the case of the DERS questionnaire, which assesses both separately, only the lack of emotional awareness subscale was used. Studies assessing mindfulness were also combined with those investigating emotional awareness and clarity as the measures used were deemed to be very similar, examining similar underlying construct of paying attention to one's feelings and knowing what are the emotions one is feeling. As above, the correlations that included measures of emotional awareness, were reversed, while those that included measures of lack of awareness or difficulties in emotional awareness were kept as is. Therefore, this review ended up assessing associations between lack of emotional awareness and ED symptomatology. Thirty-four studies examined associations between the impact of emotions on participant's problem solving abilities and ED symptomatology ( Table 1). Six different measures were used to assess problem solving, four of which were selfreport questionnaires and the most commonly used measure of problem solving was the DERS difficulties with goal directed behaviors subscale (N = 24). One of the measures, the Means-Ends Problem-Solving Test [MEPS; (124)], was a performancebased measure used to evaluate the effectiveness of participants problem solving strategies. In this task participants are given the beginning and end of four different scenarios and they are then asked to provide the middle part connecting the beginning and ending. In all but two of the measures a higher score indicated better problem solving and thus, the correlation coefficients were reversed for the meta-analysis to assess association between problem solving difficulties and ED symptomatology. The Social Problem-Solving Inventory-Revised [SPSI-R; (125)] includes the negative problem orientation subscale and the DERS includes difficulties with goal directed behaviors subscale, both of which assess negative approach that prevents effective problem solving. This subscale was used in the present review.
Eight studies examined associations between cognitive reappraisal of emotions and ED symptomatology ( Table 1). Two different measures were used to assess cognitive reappraisal, both of which were self-report questionnaires and the most commonly used measure was the Emotion Regulation Questionnaire [ERQ; (126)] reappraisal subscale (N = 5). In both questionnaires, higher scores indicated greater use of cognitive reappraisal and thus, the correlation coefficients were reversed to assess associations between lack of cognitive reappraisal and ED symptomatology.

Maladaptive Emotion Regulation Measures
Forty-seven studies investigated associations between emotional avoidance and ED symptomatology ( Table 1). Twenty-one different measures were used to assess emotional avoidance, 16 of which were self-report questionnaires; the most commonly used measure was the Temperament and Character Inventory [TCI; (127)] harm avoidance subscale (N = 13). One of the measures, the Anagram solution task, used a behavioral measure of avoidance of threatening words. In this task, participants were given a set of anagrams to solve and the time taken to reach the correct solution was used to measure avoidance. In all measures, higher scores indicated more avoidance.
Seven studies examined associations between rumination and ED symptomatology ( Table 1). Four different measures were used to assess rumination, all of which were self-report questionnaires. The most commonly used measure to assess rumination was the Ruminative Response Scale [RSS; (128)] (N = 5). One of the questionnaires used was adapted from the RSS to assess ED specific rumination (46). In all measures used, higher scores indicated more rumination.
Thirteen studies investigated association between emotion suppression and ED symptomatology ( Table 1). Six different measures were used to assess emotion suppression, five of which were self-report questionnaires and the most commonly used measure was the only task-based assessment, which examined evoked facial affect (N = 5). In the Evoked facial affect task, participants were presented with emotionally provoking stimuli and their evoked facial expressions were analyzed. In this task, higher scores indicated more emotion expression and less suppression, and thus, the correlation coefficients were reversed to reflect reverse scoring of the task. In all self-report measures, higher scores indicated more emotion suppression.

Quality Assessment
Quality assessment was conducted by J.L. and D.B. using the NIH Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies (https://www.nhlbi.nih.gov/healthtopics/study-quality-assessment-tools) to investigate study level bias and the results are shown in Supplementary Data. Question relating to exposures were not considered as they were not relevant for the purpose of the present review, which focused on correlations prior to any potential interventions or exposures. Inter-rater reliability was good [intra-class correlation coefficient = 0.94, 95% CI (0.93, 0.96)]. Any disagreements were resolved through group discussion. All included studies were deemed to be high enough quality to be included in the meta-analysis and the sum of the quality scores were included in a meta-regression to examine whether the study quality could explain any potential network inconsistency.

Statistical Analysis
We conducted a Bayesian network meta-analysis to examine which aspects of emotion regulation were most closely associated with ED symptomatology. The meta-analysis was conducted in R (129) using the packages gemtc (130) and metafor (131). First, all relevant correlation coefficients of the association between an aspect of emotion regulation and ED symptomatology were extracted from the included studies. The Pearson's r coefficients were then adjusted using sample-size weights to approximate population correlation and Fisher's r-to-z transformed was conducted using the function escalc from the package metafor. We converted the correlation coefficients to z-scores for the purposes of the network meta-analysis as z-scores are not bounded and come from a normal distribution. We then calculated standard error for each study from the sample size adjusted variance using the following steps where v is the estimated, sample size adjusted variance, σ is the standard deviation, and n is the sample size. The sample size adjusted correlation coefficients and standard errors were then taken forward to conduct the network meta-analysis. First, we generated an initial network of the data where edge thickness represents the number of studies that reported that correlation. The generated network object was then taken forward to specify and compile the random effects model with four Markov chains and a normal likelihood function with an identity link. Next, a Markov Chain Monte Carlo (MCMC) simulation was conducted to estimate the posterior probabilities. To ensure convergence we specified 1,00,000 iterations with 5,000 burn-in iterations. The Gelman-Rubin plots are presented in Supplementary Figure 1. The parameter estimation is conducted by utilizing Just Another Gibbs Sampler (JAGS). We then generated rank probability and forest plots to visualize the direct comparisons between different aspects of emotion regulation and to examine which aspect was the most closely associated with ED symptomatology. Additionally, we also calculated Surface Under the Cumulative Ranking (SUCRA) score to evaluate which aspect of emotion regulation is most relevant in terms of ED symptomatology and might serve a useful target for interventions.
Network inconsistency was examined using a network splitting method and node comparisons with Bayesian p < 0.05 were considered to be indicating inconsistency in the network. As we included studies with adolescent and adult participants of any gender who had any ED diagnosis, meta-regressions were conducted to explore if any possible network inconsistency could be explained by between-study heterogeneity in study quality score, age, BMI, and whether the studies had female only or mixed samples. The meta-regressions were conducted by taking the above steps with a specified regressor.
This review has been registered on PROSPERO (ID: CRD42021249996) and the code and data used to conduct the meta-analysis is available on the Open Science Framework online repository (https://osf.io/gz3kt/?view_ only=6fe361c208e04817b820c1f3fb2fd2b5).

Study Characteristics
The characteristics of each study included in the review are summarized in Table 1. Altogether, data from 19,734 participants were included in the meta-analysis, with average study sample size of 181 (range = 6-2,405). Majority, 72, of the studies included only female participants and three studies did not report the participants' gender. Additionally, most studies included a mixed ED sample and assessed emotion regulation across ED diagnoses while 50 studies examined emotion regulation within one diagnostic group. The mean age across studies

Network Characteristics
The initial network graph, in Figure 2, shows the eligible comparisons to identify emotion regulation strategy most strongly linked to ED symptomatology. The thickness of the edges indicates the number of studies that evaluated a given association or comparison. In the present review most of the studies (N = 50) examined the association between avoidance and ED symptomatology. Across the 104 included studies, 294 correlations contributed to the network. All emotion regulation strategies were compared with at least one other emotion regulation method in addition to the association with ED symptoms resulting in a well-connected network, which is more likely to produce reliable results (132).

Forest Plot
The forest and SUCRA plots from the Bayesian random effects network meta-analysis are shown in Figure 3. Each emotion regulation strategy represents a unit of analysis and is thus presented on the rows. Larger positive effect sizes indicate a stronger association between a given emotion regulations strategy and ED symptomatology. The meta-analysis showed that the emotion regulation strategies vary in their relationship with

Meta-Regressions
Network inconsistency was evaluated using the node-split method (Supplementary Figure 2). Although there was no evidence of significant inconsistency, the comparisons between problem solving difficulties and lack of emotional awareness (p = 0.076) and rumination and non-acceptance of emotions (p = 0.084) approached significance. Therefore, three metaregressions were conducted to examine whether these slight inconsistencies could be explained between-study differences in study quality score, BMI, age, or whether the study included only female participants or a mixed sample of participants.

DISCUSSION
The aim of the present review was to examine associations between various adaptive and maladaptive emotion regulation strategies and ED psychopathology trans-diagnostically to identify strategies that were most closely linked with psychopathology. We used network meta-analysis approach with a well-connected network, which identified rumination and non-acceptance of emotions to be most closely associated with ED symptomatology. Difficulties in cognitive reappraisal was found to be the least connected with ED symptomatology of all emotion regulation strategies examined in this review. There was some evidence of network inconsistency which approached significance and two meta-regressions were performed to examine if this could be explained by the between-study variability in BMI or age. The meta-regressions were nonsignificant suggesting the findings were consistent across age groups and BMI.
The present review adds to the steady accumulation of evidence highlighting the relationship between maladaptive emotion regulation based on rumination and ED symptomatology (133,134). Over the recent years, studies using ecological momentary assessments (EMAs) have reported that rumination and repetitive negative thinking predicts engagement in ED behavior, including body checking and binge eating (135)(136)(137). One EMA study also reported that excessive rumination also predicted higher levels of ED psychopathology at a 1 month follow-up assessment (135). Moreover, it has been suggested that some ED symptoms, such as excessive focus on food, eating, and body weight and shape, are forms of illness specific rumination (138). Indeed, a longitudinal study using EMA and biological assessments found that food-related rumination was linked to BMI status and leptin levels among people in treatment for anorexia nervosa (139). This suggest that at least certain aspects of illness-specific rumination may be linked to under-nutrition and physiological signaling, explaining why weight restoration can have a positive impact on this type of rumination. Overall, it appears that rumination, whether general repetitive negative thinking or illness-specific, is a key characteristic of the acute stage of an ED.
It is important to note that most of the studies included in the present review used the RRS questionnaire, which assesses brooding, depressive rumination, and reflection. Therefore, it is possible that the observed strong association between rumination and ED psychopathology may be partially explained by known links between depression and ED symptoms (140,141). Indeed, one study reported that although food-related rumination was linked to stage of illness and under-nutrition, other aspects of illness-specific rumination including, body weight and shape, were more associated with negative mood than not ED symptoms (139). Additionally, an experimental study has reported that induction of body shape related rumination had a direct negative impact on mood but not on ED cognitions among people with BED (142). Another interpretation is that the above findings may reflect the mechanism through which rumination influences EDs: due to its repetitive nature of rumination forms a habitual relationship with depressive mood which in turn fuels ED symptomatology (143). In support of this hypothesis, one longitudinal EMA study documented a bidirectional mediation between rumination, negative mood, and ED symptoms (136). The authors found that rumination mediated the association between low mood and ED symptomatology and low mood in turn mediated the association between rumination and ED symptomatology. Taken together these findings highlight the need to further examine the underlying processes through which rumination impacts EDs and how rumination may impact illness progression and recovery.
Difficulties in adaptive emotion regulation, specifically with accepting emotions, was another domain that was highly associated with ED psychopathology in the present review. This is in line with findings from the general population showing that reduced use of adaptive emotion regulation strategies, such as acceptance, and increased reliance on maladaptive methods, including suppression and avoidance, are associated with daily habit of food restriction (144). Similarly, people who report objective binge eating episodes and loss of control over eating also report more difficulties accepting emotions than those who do not engage in disorder eating behaviors (145). Additionally, a recent longitudinal study reported that a poor response to CBT-based ED treatment was associated with lack of change in the patients' self-reported ability to accept emotions and limited access to other effective emotion regulation strategies (146). Moreover, another interventional study found that irrespective of treatment condition, changes in acceptance of emotions were associated with greater improvements in ED-related quality of life post-intervention (27). These findings have led some authors to suggest that emotion regulation in general and acceptance of unwanted emotions in particular should be added to current standard ED treatments (146,147). While it seems that acceptance of emotions may have a strong role in disordered eating and the progression of EDs, further investigation of the underlying mechanism would be of interest to aid the development of evidence-based interventions.
Interestingly, difficulties accepting emotions has also been proposed to be linked to excessive reliance on other maladaptive emotion regulation strategies, including rumination, avoidance, and suppression (148,149). If unwanted emotions are deemed unacceptable, a person might put great effort in avoiding situations that give rise to such emotions or, if the emotions are already present, engage in suppression in an attempt to manage the unwanted emotions. Findings from a review of experimental and self-report studies support this notion reporting links between suppression and non-acceptance of emotion among people with AN and BN (150). Difficulties accepting and general dislike of emotions have been linked to greater general tendency to engage in rumination as well as worry and low mood (151). Interestingly, difficulties in emotion regulation, including non-acceptance of emotions, have been found to mediate the association between experiential avoidance and ED psychopathology among people with a range of ED diagnoses (44). Similar findings have been reported in the general population with reduced access to adaptive emotion regulation strategies, including acceptance of emotions and reappraisal, and increased reliance on maladaptive strategies based on suppression being associated with greater tendency to engage in ED-related behaviors (144). Thus, further investigation of the mechanisms that might underlie this connection as well as examination of the impact of acceptance-focused interventions on the use of other emotion regulation strategies among people with EDs may be of interest.
Interestingly, our meta-regression also found a significant effect of BMI, such that the associations were generally weaker among those with lower BMI. This is somewhat in contrast with previous reviews suggesting that there are no significant differences in the associations between emotion regulation methods and ED psychopathology between different ED diagnostic groups characterized by low and high BMI (12,13). However, some studies have reported that starvation impacts emotion regulation in such a way that those with very low BMI in the acute stage of AN report fewer difficulties (152). Indeed, it has been suggested that self-starvation itself works as an emotion regulation strategy which reduces both the internal experience and external expression of emotions through (153). By suppressing physiological responses and arousal, starvation can help the person escape and avoid unwanted emotions (153,154). If no alternative methods are available, the person may over time become reliant on starvation as their sole emotion regulation strategy due to its numbing effect (155,156). This mechanism could be one of the factors contributing to the present finding and it further highlights the complex relationship between emotion regulation and EDs.

Clinical Implications
Over the recent years, several reviews have recommended the use of interventions aimed at reducing rumination and repetitive negative thinking, Metacognitive Therapy (MCT) and Rumination-Focused Cognitive Behavioral Therapy (RFCBT), such as in the treatment of EDs (10,12,133,134). MCT and RFCBT have been successfully used to treat anxiety and depression among other psychiatric disorders, and there is some evidence suggesting that rumination-focused treatments may help reduce the risk of relapse in depression (157)(158)(159). Furthermore, one of the meta-analytic reviews found a very high correlation between rumination and ED symptomatology among the general population, which led the authors to suggest that rumination may be a useful target for interventions aiming to prevent EDs and other forms of disordered eating (133). To date, very few studies examining the impact of interventional designed to target rumination in EDs have been conducted. To our knowledge only one case series has explored the effects of MCT on binge eating behavior among three people with BED (160). MCT aims to alleviate repetitive negative cognitions by increasing awareness and mindfulness, and modifying the metacognitions that underlie maladaptive behaviors, such as binge eating (161,162). The Robertson and Strodl (160) found that MCT intervention significantly reduced binge eating frequency and improved cognitions related worry and rumination. There improvements were maintained at a 2-month follow-up assessment. Together, these findings highlight the need to develop new interventions or adapt existing treatments to target rumination and repetitive negative thinking in EDs.
Acceptance and mindfulness based interventions, such as Acceptance and Commitment Therapy (ACT), have been proposed to target difficulties in coping with unwanted emotions in EDs (147,163). The purpose of ACT is to encourage people to accept and experience unwanted emotions without attempts to modify them, thus reducing avoidance and suppression of difficult emotions and embracing the use of adaptive emotion regulation strategies (163). A number of small case series have reported that ACT can be effective in reducing ED symptomatology and behaviors among people with AN and BED (164)(165)(166). Another larger longitudinal treatment study found that ACT was more effective than treatment as usual in reducing residual ED symptoms and risk or relapse following standard ED treatment and the effects were maintained at a 2-year follow-up (167). However, another interventional study found that ACT did not lead to greater improvements in ED related quality of life than treatment as usual (27). Furthermore, a systematic review examining the use of ACT to treat body image disturbance and weight dissatisfaction reported that they could not determine the effectiveness of ACT due to the poor quality of current evidence (168). These findings suggest that acceptance and mindfulness based interventions may be promising in the treatment of EDs, but more research into the mechanisms that underlie the relationship between emotion acceptance and ED psychopathology is needed to create evidence-based treatment strategies.

Limitations
The main limitation of this review was the use of the network meta-analysis due to its use of indirect evidence, which relies on the assumption of transitivity (169). According to the transitivity assumption, a given associations is exchangeable between studies even if a given study did not assess that association. This assumption can be violated by individual differences in the samples between studies, which can be difficult to control resulting in network inconsistency. Although we did not observe significant network inconsistency, there was evidence of near significant inconsistency in two comparisons. Moreover, since we used the network approach to meta-analysis we were able to only include studies that reported correlation coefficients, which led to the exclusion of a substantial number of otherwise relevant studies. Although we attempted to contact the corresponding authors of all papers which did not report the relevant data, we were able to gain access to correlation coefficients from only 28 studies through personal correspondence. Including only a subset of the available literature to the meta-analysis may have impacted the findings.
Another limitation of this review is that majority of the studies included adult women with normal BMI. Although the metaregressions indicated that age did not have a significant impact on the results, findings from the present review may not be fully generalizable to all age groups. Additionally, even though we did find a significant effect of BMI, we were not able to examine the impact of diagnostic group as the gemtc package used in the present analysis does not presently handle categorical covariates. To truly examine the potential transdiagnostic nature of emotion regulation difficulties in EDs, a direct comparison of different diagnostic groups would be needed rather than solely focusing on BMI. Furthermore, the findings may not be fully generalisable to other genders. Gender identity can impact a person's experiences with the world around them and thus influence their emotion regulation habits (170,171). Thus, further exploration of the impact of gender and particularly minority gender identity on emotion regulation in EDs is needed. Finally, it is also important to note that we did not have equal number of studies examining each association. Indeed, only eight studies reported correlations involving difficulties with cognitive reappraisal and this method was found to be most weakly associated with ED psychopathology with largest between study variance. It is possible that this finding may have been impacted by the small number of studies included. However, only seven studies reposted correlations involving rumination and this maladaptive strategy was found to be most strongly associated with ED psychopathology. Still, equal and large number of studies examining each association would enable us to draw stronger conclusions.

CONCLUSIONS
This review aimed to build on previous work by conducting a trans-diagnostic network meta-analysis to identify emotion regulation strategies most closely associated with psychopathology among those with ED. The meta-analysis revealed that rumination and difficulties accepting emotions were most closely associated with ED symptoms, while the weakest association was between difficulties with cognitive reappraisal and ED symptoms. The meta-regressions showed that BMI had significant impact such that the associations between various emotion regulation strategies and ED psychopathology were weaker among those with low BMI. The present findings add to the steady accumulation of evidence highlighting the relationship between ED psychopathology and reliance on maladaptive emotion regulation strategies based on rumination and non-acceptance of emotions. Together with previous longitudinal observational studies and ecological momentary assessments, these findings suggest that these two maladaptive strategies may have a key role in maintaining and perpetuating ED. The meta-regression finding also emphasizes the complex relationship between ED symptoms and emotion regulation. It is possible that some people with ED may use starvation and malnutrition to escape and avoid unwanted emotions. Thus, there is pressing need to explore and develop interventions targeting emotion regulation difficulties in ED with particular focus on rumination and non-acceptance of emotions.

DATA AVAILABILITY STATEMENT
Publicly available datasets were analyzed in this study. This data can be found here: https://osf.io/gz3kt/?view_only= 6fe361c208e04817b820c1f3fb2fd2b5.

AUTHOR CONTRIBUTIONS
JL: conceptualization, methodology, investigation, formal analysis, visualization, writing-original draft, and writingreview and editing. DB and HM: methodology, investigation, validation, and writing-review and editing. SW and KT: conceptualization, writing-review and editing, and supervision. All authors contributed to the article and approved the submitted version.