Eating disorder risk during behavioral weight management in adults with overweight or obesity: A systematic review with meta‐analysis

Summary This systematic review examined change in eating disorder risk during weight management interventions. Four databases and clinical trials registries were searched in March and May 2022, respectively, to identify behavioral weight management intervention trials in adults with overweight/obesity measuring eating disorder symptoms at pre‐ and post‐intervention or follow‐up. Random effects meta‐analyses were conducted examining within group change in risk. Of 12,023 screened, 49 were eligible (n = 6337, mean age range 22.1 to 59.9 years, mean (SD) 81(20.4)% female). Interventions ranged from 4 weeks to 18 months, with follow‐up of 10 weeks to 36 months post‐intervention. There was a within group reduction in global eating disorder scores (20 intervention arms; Hedges' g = −0.27; 95% CI −0.36, −0.17; I 2 67.1%) and binge eating (49 intervention arms; −0.66; 95% CI −0.76, −0.56; I 2 82.7%) post‐intervention, both maintained at follow‐up. Of 14 studies reporting prevalence or episodes of binge eating, all reported a reduction. Four studies reported eating disorder symptoms, not present at baseline, in a subset of participants (0%–6.5%). Overall, behavioral weight management interventions do not increase eating disorder symptoms for most adults; indeed, a modest reduction is seen post‐intervention and follow‐up. A small subset of participants may experience disordered eating; therefore, monitoring for the emergence of symptoms is important.


Funding information
However, despite psychosocial and mental health being increasingly recognized as important outcomes of obesity care, 3 eating disorder risk remains understudied from an efficacy or safety perspective.
Evidence shows that the prevalence of eating disorders is higher in adults with overweight or obesity, compared to those of lower weight, 4 and there is a potential lifelong comorbidity associated with eating disorders. 5 This amplifies the need to investigate the impact of behavioral weight management interventions on eating disorder risk.
Behavioral weight management interventions aim to improve weight status and health outcomes for adults with overweight or obesity. However, there are concerns that interventions may unintentionally contribute to the onset of disordered eating or the development of eating disorders. 6,7 Several core components of behavioral weight management interventions are considered risk factors or behaviors associated with the development of eating disorders in community samples. For example, although dieting and dietary restraint are frequently employed in weight management interventions, they are also associated with the onset of binge eating, binge eating disorder, and bulimia nervosa in adolescent girls. 8,9 Similarly, excessive dietary restriction, excessive increase in physical activity, and significant and/or rapid weight loss are features of anorexia nervosa. 10 Our systematic review of dietary interventions used for weight management in children and adolescents found that eating disorder risk did not increase, 11 and that dietary restraint may not be a useful marker of risk in this context. 12 However, to our knowledge, there is a lack of understanding on the impact of weight management interventions on eating disorder risk in an adult population. Behavioral weight management interventions conducted in adults may have a greater focus on weight loss and may use more prescriptive interventions, compared to those used in pediatrics, warranting investigation.
Recent systematic reviews have examined the impact of weight management interventions on various dimensions of mental health and disordered eating among adults. A 2020 meta-analysis of 42 trials found behavioral weight management interventions to lead to a greater improvement in symptoms of depression, mental healthrelated quality of life, and self-efficacy than inactive comparators. 13 This review identified only one study reporting results for binge eating, finding no difference between intervention and control arms on the likelihood of reporting any binge eating. 13,14 A 2017 systematic review identified five trials of weight management interventions measuring eating disorder risk. 15 All trials reported improved eating disorder outcomes, including a reduction in binge eating post-intervention compared to baseline. A 2015 systematic review of 10 trials 16 examined restrictive diets including low-energy diets and very low energy diets with mixed findings. Trials including participants with pre-treatment binge eating disorder generally reported a reduction in binge eating behaviors. Yet, among trials including participants with sub-clinical or no binge eating symptoms at baseline, some reported a reduction in symptoms and others reported no change or an increase. Within that review, two studies reported the onset of binge eating or binge eating disorder in 10%-15% of participants who did not report binge eating at baseline. 16 The conflicting findings across previous relevant reviews make it difficult to thoroughly understand the impact of behavioral weight management interventions on a broad range of eating disorder outcomes. Additionally, the present literature has either had a broad focus, that is, including all mental health outcomes 13 or all obesity treatment approaches and outcomes, 15 or a very narrow focus, that is, limited to examining binge eating following restrictive diets. 16 To date, there has not been a comprehensive synthesis focused on eating disorder outcomes following behavioral weight management in adults, and no review of eating disorder risk in weight management has included a meta-analysis.
Eating disorder risk may be measured using a variety of assessments examining different outcomes. This ranges from risk scores for global eating disorder risk or binge eating severity, for example, using the Eating Disorder Examination Questionnaire (EDE-Q) 17 or the Binge Eating Scale (BES), 18 to the number of episodes of disordered eating behaviors (e.g., binge eating, loss of control, or purging) and/or the prevalence of various disordered eating behaviors or symptoms.
To comprehensively investigate the impact of behavioral weight management on eating disorder risk, this systematic review aimed to examine the change in a broad range of eating disorder outcomes, including prevalence, global eating disorder scores, binge eating scores, and episodes.

| METHODS
This systematic review was prospectively registered on PROSPERO (CRD42021265340) and follows the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA). 19

| Eligibility criteria
This systematic review included randomized controlled trials of behavioral weight management interventions conducted in adults (aged ≥ 18 years at baseline) with overweight or obesity, defined as body mass index (BMI) ≥ 25 kg/m 2 . Interventions were targeted to individuals or groups. Trials aimed at obesity prevention in a community sample and including individuals in a healthy weight range were excluded, but obesity prevention trials in a population classified as having overweight were included. Trials were excluded if they included bariatric surgery or pharmacotherapy, or targeted secondary or syndromic causes of obesity (e.g., Prader Willi syndrome) or an alternate medical condition (e.g., type 2 diabetes and sleep apnea), or if they aimed to treat eating disorders. Trials comparing two or more active interventions, for example, a novel intervention compared to standard care, as well as those comparing a weight management intervention to a no-treatment control were eligible. Intervention arms were defined as those providing any advice or information relating to nutrition, physical activity, sedentary behavior, sleep health, or behavior change outcomes, for the purposes of weight management, including print information, online programs, or individual or group consultations. The intervention duration was defined as any period with ongoing contact with the study team, and follow-up was defined as the period with no contact or intervention provided. Support provided during a weight maintenance period was considered part of the intervention. Control arms were defined as those providing no treatment or support during the study period, for example, waitlist control groups. There was no limit on intervention duration, setting (e.g., community, inpatient, and outpatient), date, or language. Articles in a language other than English were translated using Google Translate to assess eligibility.

| Outcomes
Trials were required to report one or more measures of eating disorder risk, symptoms, or behaviors at baseline and postintervention or follow-up using a validated self-report questionnaire or diagnostic interview, for example, clinical diagnosis (e.g., using the Eating Disorder Examination [EDE] interview), global risk score or binge eating episodes (e.g., using the EDE-Q), and binge eating severity (e.g., using BES).

| Information sources and search strategy
Electronic databases (MEDLINE, Embase, PsycINFO, and Scopus) were searched from inception to March 2022 (see Table S1 for search strategies). Records were imported into Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia) to remove duplicates and screen against eligibility criteria. Two reviewers independently screened records by title and abstract and then full text, with reasons for full text exclusion recorded ( Figure 1). The clinical trials registries 20 ClinicalTrial.gov and WHO International Clinical Trials Registry Platform were searched from inception to May 2022 using the key words weight management OR obesity treatment.
Records were exported to Microsoft Excel (Microsoft Corporation, WA, USA) and first screened by title and outcomes by one reviewer, and then the complete online records were screened by two reviewers. Conflicts were resolved through discussion.

| Data extraction
Data were extracted independently by one reviewer and checked for accuracy by a second. Extracted data included participant characteristics, study and intervention design, setting, frequency of contact, eating disorder assessment tool, and weight-related (e.g., weight F I G U R E 1 Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) flow diagram. *Two articles each were identified for 11 trials. and/or BMI) and eating disorder outcomes at all measured timepoints.
Where the required data for meta-analysis were not reported, authors were contacted to request this. One study was excluded from metaanalysis because the required data could not be obtained.

| Risk of bias assessment
Risk of bias was assessed independently by two reviewers with discrepancies resolved through discussion. Each intervention arm within a trial was assessed independently, using the US Academy of Nutrition and Dietetics Quality Criteria Checklist: Primary Research. 21 The checklist allows a rating of positive, neutral, or negative to be given to each intervention arm within the study.

| Data synthesis
First, a narrative synthesis of results was conducted according to the Synthesis Without Meta-analysis (SWIM) guidelines. 22 Trials were synthesized as those comparing a weight management intervention with a no-treatment or weight-neutral comparator and trials comparing two or more weight management intervention arms. As the focus of this review was to examine eating disorder risk from a safety perspective, we primarily examined the pre-post intervention change within each intervention arm and the difference between weight management interventions and controls when the comparator did not provide a weight management intervention, that is, waitlist or weight neutral control groups.
Where the required data were available, meta-analyses were conducted to determine the difference in means between baseline and post-intervention, post-intervention and follow-up, and baseline and follow-up for global eating disorder and binge eating scores. Postintervention was defined as the outcome measure taken immediately following the end of the intervention. Follow-up was defined as the latest data collection timepoint, following a period of no intervention.
Meta-analyses were conducted to examine the within group change within individual intervention arms and the difference between intervention arms and weight-list controls where more than two studies were available. Analyses were conducted using the metafor package 23 For analyses that required the combination of scores from different questionnaires, Hedges' g and the respective 95% confidence interval (95% CI) were calculated by applying the method recommended by Goulet-Pelletier (2018) 25 for estimating the effect size between two repeated measurements. Hedges' g was chosen as the measure of effect size as it carries the same interpretation as Cohen's d (0.2 = small effect, 0.5 medium effect and 0.8 large effect), and it is not biased when applied to small samples. We used the pooled standard deviation as the standardizer for Hedges' g, as has been recommended. 25,26 When the correlations between timepoints were not available, we estimated Hedges' g through using a conservative estimate of r = 0.7 according to the recommendation of Rosenthal (1993). 27 In the small number of trials where only the standard deviation of the difference was reported, we estimated the pooled standard deviation assuming the same correlation value. We probed how robust our conclusions were to our assumptions in sensitivity analyses which varied the assumed correlation values [r = 0.3, 0.5, 0.9]. To display any differences between trials that required the use of a correlation assumption to calculate the pooled standard deviation, we reported the subgroups for trials that had reported scores at each time point (no correlation assumption) against the small number of trials that only "reported difference scores." When scores were synthesized across the same questionnaire (EDE-Q only or BES only), the effect measures used were mean difference scores and the respective 95% confidence interval.
We used a random effects meta-analysis model to synthesize all estimates as the trials being examined employed varying interventions and measurement tools. This was deemed appropriate as random effects models assume there are true effect estimates that vary between studies and that this is because of heterogeneous factors.
Heterogeneity was examined through τ 2 which captures the between study variance and prediction intervals which captures the likely effect sizes of a new study if that study was selected at random from the same population of studies. Inconsistency between observed variance and heterogeneity was captured with I 2 . Heterogeneity was further explored by examining whether the time of data collection was related to the effect estimate using meta-regression in each respective analysis. Where sufficient data were available, sub-group analyses were conducted to compare active intervention arms with standard care/minimal interventions (e.g., information booklet).
Publication bias was examined by funnel plots and Egger's test.

| Intervention characteristics
Most trials compared two or more weight management intervention groups,  and five compared a weight management to a weightneutral intervention group using a non-diet approach to health focusing on self-acceptance and improving emotional well-being. [70][71][72][73] Four trials included a no-treatment or waitlist comparator group. 14,71,73,74 Interventions were primarily conducted in community or hospital outpatient settings, with one trial conducted as part of care provision within the US military. Intervention duration ranged from 4 weeks to 18 months with follow-up of 10 weeks to 36 months from the end of intervention. Interventions were delivered by multidisciplinary teams or a single health professional. Two trials 34,36 did not report who delivered the intervention.
Dietary components within interventions included nutrition education with or without an energy prescription (ranging 1200 kcal/day to 2000 kcal/day) with some trials including additional approaches such as mindful eating. 35

| Outcomes
Studies measured eating disorder risk using nine different assessment tools. The EDE-Q was most frequently used to assess global eating disorder risk and/or objective or subjective binge eating episodes, and the BES was most frequently used to measure binge eating severity.
Across all included assessments, a higher score indicates greater symptom severity. Outcome data for all included trials can be found in Tables S2 and S3.

| Risk of bias assessment
No differences in quality ratings were assigned for intervention arms within a trial; an overall rating per trial is reported in Table S2 76 in the weight-neutral group 73 or in both the intervention and weight-neutral groups. A greater reduction was reported in the weight-neutral group on the drive for thinness and maturity fears subscales. 70 One trial 14 reported no difference between groups on the Anorectic Cognitions Scale, and another 72 found a greater reduction on the EDE-Q post-intervention in the weight-neutral group compared to the weight management intervention group, with no difference between groups at 24 months followup. On the subscales for the EDE-Q, there was a reduction in both groups on the shape and weight concern subscales, with no difference between groups. There was a significant difference in the dietary restraint subscale with an increase in the weight management group and no change in the weight-neutral group.

| Change in global eating disorder risk score
Nine trials reported change in global eating disorder risk scores. Two trials reported no change in eating disorder risk within the included intervention groups. 37,72 Four trials reported a reduction in ED risk within both intervention groups with no difference between groups. 30,49,50,62 Three trials reported no difference between intervention groups but did not report within group change. 32,40,46 No trials reported increased scores within or between groups.
Five trials reported on individual subscales within the EDE-Q.
Studies reported reduced eating concern, 43,54,62,72 weight concern, and shape concern 50,62,72 in one or more intervention arms. One trial reported increased dietary restraint, 72 one trial reported an increase post-intervention, no change from baseline at 1 month follow-up and a reduction from baseline at 6 months follow-up, 50 and another reported no change. 54

| Change in binge eating score
Twenty-eight trials reported on the change in binge eating scores.
Five trials reported no change in binge eating scores within at least one intervention group, 41 One trial 57

| Meta-analyses
Change in global eating disorder risk score  Figure S16). However, it is noteworthy that meta-regression for both post-measurement to follow-up, and baseline to follow-up are comprised of less than 10 data points and fewer individual trials, and consequently, we suggest that further evidence is needed to confirm the stability of these findings. Furthermore, the sensitivity analyses we conducted demonstrate that these conclusions are robust to our correlation assumptions (Figures S17-S19). We did not find a difference between the minimal or usual care interventions and active intervention arms for change in binge eating (Q M = 2.31; p = 0.13) ( Figure S20). However, as only four minimal intervention arms were included in this analysis, more evidence is needed to support these findings.
We did not find any indication of publication bias, with funnel plot symmetry (Figures S21-23) and Egger's test p-value Results are consistent with previous reviews examining the impacts of weight management on the mental and psychosocial health of adults with overweight or obesity, 13,15 and reviews examining eating disorder risk in weight management interventions in adults 15,16 and children and adolescents. 11 As with these earlier reviews, reporting of long-term eating disorder outcomes was limited, with only nine of 49 included trials reporting follow-up from the end of intervention ranging from 10 weeks to 36 months. Further research is required to understand the implications of weight management on eating disorder symptoms and behaviors over much longer periods.
F I G U R E 5 Forest plot of the change in binge eating between baseline and post-intervention following behavioral weight management in adults with overweight and obesity. Each estimate was standardized using Hedges' g. A correlation of 0.7 was assumed between time points when necessary for the calculation of Hedges' g. A random effects model was used to combine estimates from each trial. Abbreviations: BES, Binge Eating Scale, BITE, Bulimic Investigatory Test of Edinburgh; IG, intervention group; MAEDS, The Multifactorial Assessment of Eating Disorders Symptoms.
Overall, we identified few trials with a no-treatment control arm.
As such, we were unable to conduct a meta-analysis comparing Investigations should also consider whether the time after supervised interventions have ended represents a significant risk for eating disorder emergence.
Most included trials measured binge eating severity or episodes; fewer trials considered global eating disorder risk or compensatory behaviors as symptoms of bulimia nervosa. This is problematic as the F I G U R E 6 Forest plot of the change in binge eating between post-intervention and follow-up following behavioral weight management in adults with overweight and obesity. Each estimate was standardized using Hedges' g. A correlation of 0.7 was assumed between time points when necessary for the calculation of Hedges' g. A random effects model was used to combine estimates from each trial. Abbreviations: BES, Binge Eating Scale, IG, intervention group.
F I G U R E 7 Forest plot of the change in binge eating between baseline and follow-up following behavioral weight management in adults with overweight and obesity. Each estimate was standardized using Hedges' g. A correlation of 0.7 was assumed between time points when necessary for the calculation of Hedges' g. A random effects model was used to combine estimates from each trial. Abbreviations: BES, Binge Eating Scale, IG, intervention group.
focus on binge eating may result in missed symptoms of restrictive eating disorders, for example, atypical anorexia nervosa or other disordered eating behaviors such as laxative misuse. It is possible that symptoms of binge eating are more likely to be measured because of the known associations between dietary restriction and the onset of binge eating, 8,9 or because binge eating is more common among adults with overweight or obesity compared to those with lower weight. 4

| Strengths and limitations
This review included a comprehensive search of published literature and clinical trial registries and is the first review examining eating disorder risk in adults undergoing weight management interventions to include meta-analyses. We combined both narrative synthesis and meta-analyses across a broad range of eating disorder outcomes to ensure a comprehensive assessment of weight management interventions. Consideration of eating disorder risk as a safety outcome allowed us to examine risk across a range of interventions. This contributed to a higher heterogeneity between intervention arms; however, prediction intervals indicated that results were robust.
Meta-analysis of aggregate data, conducted as part of this review, represents the likely change in eating disorder risk for most adults.
Considering that eating disorders are likely to affect a small proportion of adults undergoing weight management, this review was limited in its ability to report on change within smaller subgroups as sufficient data were not available from individual trials to permit examining indi-