Third‐wave cognitive behaviour therapies for weight management: A systematic review and network meta‐analysis

Summary This systematic review and network meta‐analysis synthesized evidence on the effects of third‐wave cognitive behaviour therapies (3wCBT) on body weight, and psychological and physical health outcomes in adults with overweight or obesity. Studies that included a 3wCBT for the purposes of weight management and measured weight or body mass index (BMI) pre‐intervention and ≥ 3 months post‐baseline were identified through database searches (MEDLINE, CINAHL, Embase, Cochrane database [CENTRAL], PsycINFO, AMED, ASSIA, and Web of Science). Thirty‐seven studies were eligible; 21 were randomized controlled trials (RCT) and included in the network meta‐analyses. Risk of bias was assessed using RoB2, and evidence quality was assessed using GRADE. Random‐effects pairwise meta‐analysis found moderate‐ to high‐quality evidence suggesting that 3wCBT had greater weight loss than standard behavioural treatment (SBT) at post‐intervention (standardized mean difference [SMD]: −0.09, 95% confidence interval [CI]: −0.22, 0.04; N = 19; I2 = 32%), 12 months (SMD: −0.17, 95% CI: −0.36, 0.02; N = 5; I2 = 33%), and 24 months (SMD: −0.21, 95% CI: −0.42, 0.00; N = 2; I2 = 0%). Network meta‐analysis compared the relative effectiveness of different types of 3wCBT that were not tested in head‐to‐head trials up to 18 months. Acceptance and commitment therapy (ACT)‐based interventions had the most consistent evidence of effectiveness. Only ACT had RCT evidence of effectiveness beyond 18 months. Meta‐regression did not identify any specific intervention characteristics (dose, duration, delivery) that were associated with greater weight loss. Evidence supports the use of 3wCBT for weight management, specifically ACT. Larger trials with long‐term follow‐up are needed to identify who these interventions work for, their most effective components, and the most cost‐effective method of delivery.


| BACKGROUND
Although behavioural interventions are effective at helping people to lose weight, many people struggle to sustain effective weight management behaviours over extended periods due to a combination of biological, psychological, social, and environmental factors that drive weight gain. 1,2 Standard behavioural programmes can be effective in the short term, but less so in the longer term. [3][4][5][6] These usually combine diet and physical activity advice with core behavioural change techniques including goal setting, self-monitoring, problem solving, and planned social support. 7 It has been proposed that third-wave cognitive behaviour therapies (3wCBT), including acceptance and commitment therapy (ACT), dialectical behaviour therapy (DBT), mindfulness-based cognitive behavioural therapy (MBCT), and compassion-focused therapy (CFT), [8][9][10] may have better short-and long-term outcomes. 2 The theoretical case for 3wCBT for weight management has been well articulated. 2 In brief, these therapies encourage people to accept aversive internal experiences (eg, food cravings, physical discomfort) rather than avoid them. Increased present-moment, non-judgemental awareness and psychological flexibility may assist an individual in recognizing internal and external cues to overeat and alter behavioural responses to be more in line with their values. Fostering a compassionate attitude towards the self could also help prevent discouragement following minor lapses. 2,8,9 However, the evidence of their superior effectiveness is less clear. Previous systematic reviews and meta-analyses primarily focused on mindfulness-and/or acceptance-based interventions. [11][12][13][14][15][16][17][18] Three reviews 13,15,16 have reported a quantitative synthesis of pre-intervention to post-intervention change without comparing the effect against a comparator. Two of these three reviews reported a "small" pre-intervention to post-intervention change in weight 15 or body mass index (BMI) 13 while the other study 16 reported a null effect on BMI. Critically, only one review 14 reported a meta-analytic synthesis that compared the effectiveness of mindfulness-and acceptance-based interventions with those in other active interventions and control arms using appropriate statistical methods. A small but significant difference in weight or BMI was reported at post-intervention, favouring mindfulness and acceptance-based interventions over comparator arms. Subgroup analysis suggested that the effect may only hold when the comparator is waitlist control. In that review, there was no restriction on the minimum follow-up duration and outcomes were analysed at 1-month post-intervention (or the closest measurement to this).
Thus, the pooled estimates reflected a mix of very short-term and longer term effects. Moreover, without a restriction on minimum BMI, these findings are less relevant from a policy perspective because behavioural weight management programmes are intended for adults with overweight/obesity. 19,20 This concern is compounded by the finding that a lower BMI was associated with a larger effect size.
To our knowledge, no head-to-head trial exists that has compared the effectiveness of different types of 3wCBT on weight management. In the absence of head-to-head trials, network meta-analysis can estimate the indirect evidence on the comparative effectiveness of different types of 3wCBT. The proposed mechanism for the superior effects of 3wCBT is through improvements in eating behaviour and psychological outcomes, so it is also important to synthesize evidence on the impact of 3wCBT on these outcomes. In addition, evidence synthesis of the effect of 3wCBT on eating behaviour and psychological outcomes has been limited to pre-intervention to postintervention change 13,15,16 and has not considered longer follow-up periods.
To address these knowledge gaps, we conducted the most comprehensive, inclusive, and relevant review and quantitative synthesis of available evidence to date. We included different types of 3wCBT beyond mindfulness and acceptance-based interventions.
Our main objectives were (a) to evaluate the effectiveness of 3wCBTs on weight management by pooling the pre-intervention to post-intervention change effect estimates across all study types, (b) to compare the effectiveness of 3wCBTs on weight management against no/minimal interventions and standard behavioural treatment (SBT) separately using random-effects pairwise metaanalysis of randomized control trials (RCTs), (c) to estimate the comparative effectiveness of different types of 3wCBTs on weight management using random-effects network meta-analysis of RCTs, (d) to evaluate the impact of 3wCBT on eating behaviour and psychological and physical health outcomes, and (e) to provide a detailed description of intervention characteristics and to identify whether any of these are associated with better weight change outcomes by using meta-regression.

| Protocol and registration
The protocol was registered on Prospero (CRD42018088255) prior to article screening. 21

| Eligibility criteria
Participants were community-dwelling adults (≥18 years) with overweight or obesity (BMI ≥25 kg/m 2 ) seeking assistance with weight management. Studies had to include a 3wCBT intervention for the purpose of weight management. Multi-component interventions (eg, including diet and physical activity advice) were acceptable, with no restriction placed on the proportion of the intervention using 3wCBT. Interventions could be of any duration. Comparisons were (a) no/minimal intervention, (b) SBT, or (c) no comparator (single-arm pre-intervention to post-intervention studies). We defined SBT as structured programmes providing diet and/or physical activity advice and standard behaviour change techniques (eg, goal setting, self-monitoring, problem solving, social support). The primary outcome was body weight or BMI. Studies needed to measure this pre-intervention and at least 3-months post-baseline. Secondary outcomes were stress, anxiety, depression, meta-cognition, eating attitudes, eating behaviours, body satisfaction, quality of life, blood pressure, lipids, glycaemia, and adherence to treatment. All outcomes reported at 3months from baseline and beyond were extracted. All settings apart from laboratories were eligible. We included research articles, theses, and dissertations reporting RCTs, non-RCTs, prospective cohort and case series studies.

| Information sources
Databases (MEDLINE, CINAHL, Embase, Cochrane database [CEN-TRAL], PsycINFO, AMED, ASSIA, and Web of Science) were searched by ERL from inception with no restrictions, using keywords and subject heading searches relating to the concepts: (a) third-wave CBTs and (b) overweight, obesity, or weight management (see Table S1).
The initial search was conducted on 16 January 2018, and an updated search was conducted on 25 September 2019. Reference lists of eligible studies and relevant reviews were searched, and authors of relevant abstracts were contacted to identify whether findings had been accepted for publication.

| Study selection
Titles and abstracts, then full texts, were screened independently by two of three researchers, with a third reviewer adjudicating uncertainty or disagreement. Study authors were contacted to resolve any questions about eligibility. Non-English language texts were translated into English by colleagues who were fluent in that language. Attempts were made to contact authors to retrieve missing data. If there was no response after two attempts, we used the data available in the published work.

| Risk of bias
Two researchers assessed studies independently using the Risk of Bias 2 tool (RoB 2) 25 or the Risk Of Bias In Non-randomized Studies of Interventions tool (ROBINS-I), 26 dependent upon study design. A third reviewer adjudicated inconsistency. The quality of evidence was assessed using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach, which classifies studies as "high", "moderate", "low", or "very low" quality. 27

| Synthesis of results
Stata/SE v.14.2 49 was used for all statistical analyses. Following guidance, 50 we focused on 95% CIs, rather than statistical significance. For example, unlike conventional interpretations, we did not outright interpret an effect estimate "non-significant" if the lower or upper bound of the 95% CI was slightly above/below the null value; we interpreted them as "suggestive" of an effect.

| Pooled estimates of intervention-specific effects from all study types
Intervention-specific effects (post-intervention minus pre-intervention) were estimated by pooling effect estimates from all study designs. Due to heterogeneity in outcome measurement, effect estimates were reported as standardized mean change from the randomeffects meta-analysis. 51 Effect estimates were reported at the earliest measurement post-intervention (≥3 months from baseline) and at 3, 6, 9, 12, 18, 24, and 36 months from baseline. Outcomes falling between these time points were included with the closest time point.

| Intervention comparisons: Direct evidence from pairwise meta-analysis of RCTs
The direct effect comparing 3wCBT against (a) no/minimal intervention and (b) SBT was estimated using random-effects 51 pairwise meta-analysis of RCTs. The standardized mean difference (SMD) calculated using Hedges' method and 95% CI were reported. 52

| Intervention comparisons: Indirect and mixed evidence from network meta-analysis of RCTs
To compare types of 3wCBT, random-effects network metaanalysis of RCTs was conducted to estimate the indirect and mixed (direct plus indirect) evidence. 53 Basic assumptions were checked conceptually and statistically. 53 For example, to avoid violating the transitivity assumption, which requires that the comparator arm (eg, the waitlist control) is comparable across the trials, the comparators (SBT and no/minimal intervention) were not pooled/used together. Similarly, the intervention arms were dropped (namely resistance exercise 35 and food environment modification 54 ) if they were not comparable with other intervention arms. The consistency assumption was checked statistically to see if the direct and indirect effect estimates were comparable enough to pool them together into the mixed evidence. 55,56 Effect estimates were reported as SMD and 95% CIs. The relative ranking probability of each intervention being the best treatment was estimated using rankograms. 57

| Sensitivity analysis
The influence of individual studies on weight change of 3wCBT compared with no/minimal intervention or SBT was examined using influence plots, where one study was removed at a time to see its effect on the overall estimate. 58

| Meta-regression on intervention and study characteristics
Where at least 10 studies provided relevant data, 59 meta-regression was used to identify potential sources of heterogeneity for pre-specified characteristics: number of sessions (continuous; <12 vs. ≥12 sessions), duration of intervention (<3 months vs. ≥3 months, <6 months vs. ≥6 months), method of delivery (face-to-face vs. remote; group vs. individual), and risk of bias (low, some concerns, high). For consistency, we defined "change" as post-intervention minus pre-intervention values, so a negative change estimate indicates that the outcome decreased after the intervention.

| RESULTS
After duplicate removal, 8755 titles and abstracts were screened and 215 full text articles were assessed. Two additional studies and four articles related to already included studies were identified from reference lists [60][61][62][63][64] and contacting an author. 65 Fifty articles reporting 37 studies met the inclusion criteria ( Figure 1). Two studies were excluded from the meta-analyses due to cointerventions (pharmacotherapy 66 and bariatric surgery 39 ). Thirty-five studies were used in the pooled-effects meta-analysis of pre-intervention to post-intervention changes, and the 21 RCT design studies were used in the network meta-analysis that compared different interventions.

| Study characteristics
Seventeen studies 30 pre-intervention to post-intervention one-group design, and one study was a non-randomized three-group study. 39 The majority of studies were conducted in the United States (n = 28). The other studies were conducted in New Zealand, 34 Italy, 39 United Kingdom, 41,42,77 the Netherlands, 43,64 Finland, 44 and Portugal 70 (Table 1).  (Table 1 and Table S2a,b).

| Intervention characteristics
Twenty-two studies evaluated MBCT, 28 and a website for pharmacology support. 66 One study 76 used individual face-to-face lifestyle counselling and telephone delivery.
Three delivered interventions on an individual, remote basis using email 74 and online website, 34,79 two of these included telephone support. 74,79 One study 44 had two intervention arms with the same content delivered face to face or through mobile telephone.
Most interventions include home-based skills practice between sessions.
Intervention duration varied, with two lasting less than a week, 28,30 twelve studies between 1 and 3 months, 29 All studies were delivered on a weekly or alternating weekly basis, apart from two: a one-off 1-day workshop and 5-day residential retreat. 28,30 Several interventions had an "active phase," then an extended period with less regular sessions or telephone followup. 38 29 and participants' place of employment 80 (Table 1 and Table S3a,b).

| Risk of bias
Of the RCTs, the risk of bias was rated as 'high' in four, 36,41,42,74 'some concern' in eleven, 30 (Table S4b). The quality of the evidence was different for different comparisons, dependent on studies included.
For the comparison between 3wCBT and no/minimal intervention at post-intervention, the quality of evidence was 'high' (three studies); 40,44,67 for the comparison between 3wCBT and SBT, the quality of evidence was 'moderate' at post-intervention (nineteen studies) 30

| Intervention effects on body weight or BMI
Twenty-five studies 28   Only ACT interventions provided data for the 24 month comparison.

| Intervention comparisons: Direct evidence
Estimates at 6 and 18 months also suggested greater weight loss for 3wCBT versus SBT, but there was no evidence of a difference between the two groups at 3 and 9 months.

| Sensitivity analysis
In the influence plot analysis, removal of one study at a time did not have any effect on the overall effects estimates from the pairwise meta-analysis of weight change. Comparisons between no/minimal intervention and MBCT did not provide evidence of a difference at any time point, and there was no consistent pattern of effects. Comparisons between MBCT and SBT suggested that SBT was more effective at 9 months, but estimates at 12 and 18 months suggested that MBCT was favoured.

| Intervention comparisons: Indirect and mixed evidence
When CFT was compared with the other interventions, CIs were wide with no comparisons favouring CFT.
When interventions were relatively ranked, ACT was the best intervention post-intervention and at 3, 6, and 9 months post-baseline. MBCT was the best ranking intervention at 12 and 18 months post-baseline; however, this was based on only five studies (two MBCT) and three studies (one MBCT), respectively ( Figure S1).
In terms of absolute weight change, for example, the SMD in weight between 3wCBT and SBT equates to a difference of 0.6 kg post-intervention and 1.4 kg at 24-month follow-up from baseline.

| Interventions effects on secondary outcomes
Pooled arm-specific estimates (standardized mean change) of the effect of 3wCBT (combined) on secondary outcomes are presented in Figure S2. Pairwise estimates (SMD) from RCTs comparing 3wCBT and no/minimal intervention are presented in Figure S3; those comparing 3wCBT and SBT are presented in Figures 5A, 5B, and 5C.
One study 68

| Meta-regression of intervention characteristics
A sufficient number of studies for meta-regression (N ≥ 10) were only available at post-intervention and at 3-and 6-months since baseline for 3wCBTs versus SBT. Prespecified study and intervention characteristics were examined in the meta-regression at these time points including number of sessions, duration of intervention, method of delivery, and risk of bias (Table S7), and none were found to have any impact on the effect estimates on weight or BMI reported in the pairwise meta-analysis. There were too few studies in each stratum to analyse the potential effects of comorbidities (eg, diabetes). Due to the small number of studies, subgroup analysis was not conducted.

| Intervention adherence
There was substantial heterogeneity and poor reporting of attendance and adherence outcomes, limiting our ability to conduct any meaningful quantitative analysis (Table S8). Only 22 studies reported any attendance information, but, for all these studies, attendance was at least 60% at group sessions overall, and eight 31,35,42,46,54,69,71,72 out of 11 RCTs reporting attendance information for each group had a 3wCBT group with higher attendance than the control arm. Information provided on adherence included minutes of home meditation practice, number of mindful meals per week, food and exercise diaries, and completion of online modules.
Generally, within each study, there seemed to be a spread of engagement in the home practice aspect of interventions. This also varied with interventions delivered via internet: one study 44 found a 91% median completion of all modules, and another 34 found a mean of 32%.

| DISCUSSION
This comprehensive systematic review and network meta-analysis found high-quality evidence suggesting that 3wCBT results in greater weight loss than no/minimal intervention. Importantly, it also found moderate-quality evidence that suggests that 3wCBT results in greater weight loss than SBT at post-intervention and high-quality evidence from a small number of studies indicating that 3wCBT  ACT was ranked as the best intervention up to 12 months and was the only 3wCBT to have outcomes at 24 and 36 months. Network estimates suggested that MBCT resulted in greater weight loss than SBT at 12 and 18 months, but favoured SBT at 9 months, and there was no evidence that MBCT was more effective than no/minimal intervention. This suggests that we should interpret the finding that MBCT was the highest ranking intervention at 12 and 18 months with some caution. Only four studies evaluated a 3wCBT approach other than acceptance or mindfulness and these were of low quality and short follow-up. Although we identified three studies using DBT, all used non-randomized pre-intervention to post-intervention design and one was combined with pharmacotherapy; therefore, they were not included in the pairwise or network meta-analysis, limiting conclusions on DBT effectiveness. CFT was found to have no evidence for weight loss; however, this finding is based upon one unpublished thesis, 41 which was deemed to be of high risk of bias and of very low quality. To date, the evidence provides strongest support for the superiority of acceptance-based interventions. It is possible that the superiority of the acceptance-based programmes in this context is due to its focus on values and willingness to reduce experiential avoidance. However, more research is needed to confirm these differences and identify the mechanisms of action.
Changes in secondary outcomes were generally in the desired direc- Meta-regression did not identify any specific intervention characteristics (eg, duration, mode of delivery, number of sessions) that were more effective than others. This may have been due to the small number of studies. Similar to traditional behavioural weight management programmes, the majority of interventions were delivered in a group face-to-face format. Such delivery has often been found to be effective in weight loss, 83 with the group providing peer support and regular encouragement, particularly for those experiencing social isolation.
However, closed-group programmes led by clinical psychologists can be logistically difficult and costly to deliver, particularly in the context of national health services providing free or discounted health care.
Increasingly, standard behavioural weight management interventions are moving to more scalable methods of delivery to increase reach and reduce cost. Only four interventions 34,44,74,79 in our review used remote delivery through internet or mobile phone. Similarly, a review of online mindful eating interventions 18 found only two studies focusing on weight loss, and none were delivered through mobile telephone, highlighting a dearth of investigation into this research area.
Greater consideration may need to be given to the use of digital technology to facilitate intervention delivery, with its 24-hour accessibility, improved scalability, and increased reach. 84

| STRENGTHS AND LIMITATIONS
The study of 3wCBT for weight management is in its relative infancy, with the earliest included RCTs from 2008 33  CIs, thereby limiting the power to detect a difference. Many of the studies included in this review had high or serious risk of bias. However, it should be noted that we used a stringent assessment tool, and heterogeneity for many outcomes was low. For some studies, this may also reflect the slowness of the obesity field to adopt standards for trial reporting (eg, CONSORT), 23 rather than the quality of the research itself. It is also important to note that the studies with longer term follow-up (ie, 12 and 24 months from baseline) were of high quality, based upon the GRADE assessment tool, 27 so we can have greater confidence in the findings at these time points.
There was heterogeneity in the content of 3wCBT programmes, with a combination of standardized, modified, and novel programmes that varied in length and practice time. Some studies evaluated interventions that used combinations of different third-wave therapies, which may obscure potential differences between types of therapy. However, this is a reflection of how these interventions are used. In attempting to collate adherence and attendance data, we found a low number of studies reporting intervention fidelity information and substantial variability in reporting. 3wCBTs seem to have comparable attendance and attrition rates to standard behavioural programmes, 85 suggesting that they are an acceptable and feasible option. Lack of information stopped us from conducting a meta-regression to try to identify sources of heterogeneity in attendance or adherence.
Studies in our review had a high proportion of female participants; this is typical in weight loss programmes and mindfulness interventions. 86 This limits generalizability of findings to males 83  There are several strengths to this review. RCTs and pre-intervention to post-intervention studies were included in our pooled estimates, but only RCTs were included in pairwise and network metaanalysis to provide the estimates of the comparative effectiveness against a comparator. By conducting a network analysis, we could estimate comparisons between different types of 3wCBT that have not been directly compared, incorporating direct, indirect, and mixed evidence in our evaluations of the evidence. To maximize on relevant research, we included unpublished theses and contacted authors regarding abstracts in conference proceedings. Unlike previous reviews, 14 we restricted our analytic population to those with a BMI ≥25 kg/m 2 to make our results more relevant to health care policies that recommend weight management interventions for people with overweight/obesity.

| CONCLUSIONS
This systematic review and network meta-analysis found moderatequality evidence suggesting that 3wCBT results in a small increase in weight loss compared with SBT at post-intervention. It found highquality evidence from a small number of studies suggesting that 3wCBT results in greater weight loss than SBT at 12 and 24 month follow-up from baseline. Evidence specifically appears to support the use of acceptance-based programmes. Larger, high-quality trials are needed in this area to better understand who these interventions work for and how they work, so that we can target these interventions appropriately and identify the most crucial components and "active ingredients." Future research should also consider how we deliver these interventions in a cost-effective way that maximizes scalability while maintaining effectiveness.

CONFLICTS OF INTEREST
AA is the chief investigator on two publically funded (MRC, NIHR) trials where the intervention is provided by WW (formerly Weight Watchers) at no cost outside the submitted work. AJH reports receiving personal fees from Slimming World, outside the submitted work.