IntroductionBackground

The current eating disorder (ED) treatment model is falling short for patients [], with a significant majority of people with EDs failing to get help []. This may be due to limited access to services [] and the stigma and shame associated with their condition []. EDs have the highest mortality of any psychiatric disorder [], and they may be long-lasting and may cause physical, emotional, and neurobiological damage if left untreated []. The COVID-19 pandemic has further compounded the problem, with a surge in urgent referrals and increased waiting time in an already underresourced system []. Action is urgently required to address this treatment gap []. A promising strategy that can improve access to evidence-based treatments is the development and implementation of digital interventions. Digital interventions refer to the use of digital technologies, such as mobile apps, websites, or virtual reality, to deliver health care or behavioral interventions.

Advantages of digital interventions include the ability to reach many people at minimal or no additional cost per person, and they can be used at an individual’s convenience, at home, anonymously, and at a self-suited pace []. Shame and stigma may make people with EDs more likely to engage in digital interventions to achieve improvements in their symptoms [,], and evidence demonstrates that the demand for self-guided digital interventions is growing among people with EDs []. While digital self-management interventions are not the only solution to address the existing service gap, they can broaden the dissemination of evidence-based treatments and help more people get support for their condition [].

Digital interventions for EDs have shown promising evidence in treating ED symptoms [-] with results sustained, or even improved, at follow-up []. However, our understanding of how these interventions work and what contributes to their effectiveness is limited [], restricting the potential effectiveness and impact of digital ED interventions. It is widely recognized that digital health interventions should incorporate evidence-based methods and behavior change theory into their development []. Theory represents the accumulated knowledge of the mechanisms of action (MOAs; mediators) and moderators of change as well as the a priori assumptions about what human behavior is and what the influences on it are []. Using behavior change theory in designing digital health interventions may help pinpoint the factors influencing the target behavior, referred to as MOAs in behavioral science. These MOAs, such as knowledge and beliefs, are pathways through which interventions can impact behaviors. Designers can then connect these MOAs to practical elements called “behavior change techniques” (BCTs), which play a crucial role in transforming disordered behaviors into healthier target behaviors. While there are some dissenters regarding such systematization of practice, arguing for the importance of variability, there is general agreement of the value of better descriptions of interventions for clarity and replication []. This systematic approach has been applied in the development of effective digital health interventions in areas such as the treatment of addictive disorders, physical activity, and weight loss [,], as well as in more clinically oriented interventions, such as diabetes management [-]. Specific BCTs have been linked to improved clinical outcomes [-] and are a useful means of describing active components within complex digital interventions []. The integration of specific BCTs may optimize digital ED treatment interventions, helping achieve significant symptom improvement by addressing those factors (eg, food avoidance, dietary restriction, and body image concerns) that influence common ED behaviors (eg, bingeing and purging).

Objectives

This review aimed to gain insights from previous randomized controlled trials (RCTs) as to which BCTs may contribute to the effectiveness of digital ED interventions []. It focused on RCTs as they have the highest possible level of evidence compared to other study designs and can be used to make causal inferences []. It also assessed whether the interventions were grounded in theory, given that theory is a “necessary precursor to the development of effective interventions” [].

We hypothesized that interventions that specifically targeted the behavioral and psychological aspects of ED via the use of relevant BCTs would be more likely to improve ED outcomes. We also hypothesized that the interventions informed by theory were more likely to be effective. Having multiple modes of delivery (eg, apps, video, and audio) may be associated with enhanced treatment outcomes [] based on the idea that the diversity offered by multimedia formats might facilitate effectiveness through an enhanced and more engaging user experience [].

Our specific research questions were as follows:

Which BCTs are most frequently included in digital interventions for the treatment of EDs that have been evaluated in RCTs? Which BCTs are most frequently associated with effective interventions?Are included BCTs informed by theory?Which modes of delivery have been adopted to deliver the BCTs?Was there evidence to suggest that specific BCTs, or related factors, moderated the intervention effect size?
MethodsSearch Strategy

The searches were completed across the following databases between April 1, 2023, and June 30, 2023: MEDLINE (Ovid), Embase, PsycINFO, CINAHL, Emcare (Ovid), CENTRAL, Web of Science, and Scopus. The protocol was registered in the PROSPERO database (CRD42023410060). These findings are reported in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines []. The search strategy was developed based upon previous similar systematic reviews of digital interventions and EDs [,] and in consultation with a specialist librarian at University College London. The search strategy included 2 main concepts based on EDs and types of digital intervention (web based or smartphone). It included a combination of Medical Subject Headings (MeSH) terms and free-text terms. The search was adapted for each database. A Cochrane RCT filter was applied to the search results within relevant databases []. Full details of the search strategy can be found in .

The first reviewer (PT) initially screened all titles and abstracts for the first phase of the review, and a second reviewer (PB) screened a random 9.98% (375/3758) of the results within Covidence (Veritas Health Innovation). Both reviewers independently screened 100% (79/79) of articles in the final full-text screening stage. Results were compared, and any discrepancy was resolved by discussion. There was a good to excellent degree of interrater agreement (initial screening: κ=0.92 and final screening: κ=0.720).

Study Selection

Eligible studies were selected by applying the inclusion and exclusion criteria ().

Textbox 1. Inclusion and exclusion criteria.

Inclusion criteria

Adults in general populationSelf-management interventions and guided self-help interventions for individualsIncluded study participants who meet subthreshold and threshold criteria for an eating disorderStand-alone digital intervention with minimal or some therapist supportOutcome measure using the Eating Disorder Examination Questionnaire (EDE-Q)Randomized controlled trials

Exclusion criteria

Interventions aimed at <16 years oldIntervention aimed at health care professionalsIntervention specific to relapse prevention and aftercareIntervention specific to eating disorder preventionIntervention aimed at obesity and weight managementTelemedicine or teleconferencingAugmentation therapy (app as an add-on)Digital intervention with intensive levels of supplementary therapist supportGroup cognitive behavioral therapy; group therapyTechnologies that have been superseded (ie, CD-ROM, vodcast, and SMS text messaging)Interventions that used mobile phones but did not involve apps (eg, were based solely on SMS text messaging or emails)No clear description of the intervention design (not possible to code for behavior change techniques)Qualitative studiesFeasibility and acceptability studies as well as pilot studiesNo clear outcome measures (using the EDE-Q)Data Extraction

The primary researcher (PT) extracted and coded the data for included studies, including author, year, country of origin, study and participant characteristics (number of participants, age, gender, ethnicity, diagnosis, inclusion and exclusion criteria, and dropout rates), and intervention characteristics (intervention description, therapist involvement, BCTs, modes of delivery, duration of treatment, follow-up, and key outcomes). Outcomes data for all the studies were independently extracted by 2 reviewers (PT and TR). Results were compared, and any disagreements were resolved by discussion. Where key data were missing, study authors were contacted for the missing information. A cutoff period of 4 weeks was provided.

Outcome Measures

The Eating Disorder Examination Questionnaire (EDE-Q) [] was used as the primary outcome measure of interest, given that it is the National Institute for Health and Care Excellence “gold standard” measure of ED psychopathology and was used as the primary outcome measure in most of the included RCTs. It includes frequency data on key behavioral features of EDs in terms of number of episodes of the behavior (including bingeing and purging), making it a suitable outcome measure for this review []. Where reported, changes in the number of objective binge episodes (OBEs) after treatment were examined for consistency, providing complementary data on intervention effectiveness.

BCT Coding, Modes of Delivery, and Theory Coding Scheme

Each study was assessed for the presence of each of the 93 BCTs using the BCT Taxonomy v1 [], assessing the number of BCTs in each digital intervention and the frequency of each BCT in the sample overall. The BCT Taxonomy is a hierarchically organized, common language tool for the classification of the active ingredients [] required to bring about change in an intervention. The validity of this approach has been well established, and its reliability and value have been consistently demonstrated across multiple areas since its inception [-].

The modes of delivery used within each of the interventions to deliver the BCTs was assessed using relevant components from the Model of Delivery Ontology v2 []. If the modes of delivery were changed during the course of the study, the modes of delivery included within the initial study design were coded, as these were appropriate for the outcome measures used.

An adapted version of the theory coding scheme (TCS) [] was used to evaluate the theoretical basis of the included studies. These adaptations were made in consultation with an experienced behavior change scientist (KC), on the basis that the coding scheme was originally developed for use in a different context and some of the criteria were not relevant. Hobbis and Sutton [] justified the case for cognitive behavioral therapy (CBT) as an addition to the Theory of Planned Behavior–based interventions; hence, it was considered a valid theoretical basis when used to inform intervention design. All studies were independently coded against these frameworks by 2 reviewers (PT and RC), with any discrepancies resolved by discussion involving a third reviewer (KC). This meant the BCTs were double-blind coded by 2 reviewers across all studies. These results were compared, with a third reviewer involved where necessary to resolve any discrepancies. A briefing document was provided to the second reviewer in advance of coding, which included definitions and examples of BCTs, to ensure reliability. The coding was completed in 2 stages, with the second reviewer coding approximately 30% (5/17) papers first. The coding was compared between the 2 reviewers to identify any inconsistencies in applying the BCT framework, aiming to maximize consistency when reviewing the remaining 70% (12/17) of the papers.

For interventions to be included in the follow-up, they had to be assessed at least 8 weeks after the postintervention period. This time frame allows for a reasonable evaluation of sustained treatment effects and avoids coinciding posttreatment and follow-up evaluations across different studies.

Data Synthesis

The associations between BCTs and intervention effectiveness were analyzed. A brief narrative synthesis was used to organize and present the data within the text, with a summary of the information extracted from each study, including outcomes reported, BCTs, and other items provided in tabular form.

Frequency counts of the most commonly used BCTs were conducted for both all interventions and effective interventions, and the results were compared. The effectiveness of an intervention was determined by a statistically significant effect (P<.05) on ED behavior change (as measured by the EDE-Q 6.0). In studies with an active comparator, the pre-post outcome data for the intervention arm were examined independently to assess efficacy. These results were then considered in the context of the study design and compared with similar waiting list (WL) control studies. BCTs were considered effective if they were identified in at least 75% (13/17) of effective interventions []. A further division of effective interventions was completed based on whether they were effective at postintervention or follow-up.

Meta-Analytic Procedure

The purpose of this meta-analysis was to pool data across RCT studies regarding the effectiveness of digital interventions compared to waitlist control or treatment-as-usual (TAU) controls at postintervention and follow-up time points to explore what might be contributing to the overall effect sizes, primarily the contribution of any particular BCT. Studies with an active control group, such as face-to-face (F2F) therapy, bibliotherapy, another digital intervention, or day patient programs, as well as studies with missing (EDE-Q total) outcome data were excluded from the meta-analysis.

As a first stage, the meta-analysis procedure calculated pooled estimates of effect sizes (differences in EDE-Q total scores) at postintervention and follow-up time points for waitlist and TAU RCTs and presented these results as forest plots (using RevMan v. 5.4, The Cochrane Collaboration). Effects were based on means, SDs, and sample sizes reported within the studies. The primary outcome was EDE-Q behaviors (dietary restraint, weight concern, shape concern, and eating concern). As the included studies were RCTs, baseline values were not adjusted for across studies, as they would be expected to be similar across treatment and control groups. Due to substantial heterogeneity among the studies, which varied in design (eg, duration of treatment and level of therapist involvement), a random-effects model was used to estimate the weighted pooled effect for each outcome. This approach accounts for the distribution of the true effect across individual studies []. The I2 statistic was used as a measure of heterogeneity, describing the percentage of variation across studies that was due to heterogeneity rather than chance []. Heterogeneity >60% was considered substantial [] and suitable for subgroup analyses. Given that the EDE-Q primary outcome measure was continuous, the mean difference (MD) was used to describe the pooled outcome effects and the overall effect size (z-statistic) alongside its P value. Sensitivity analysis was completed to check for consistency of the effect size, and publication bias was explored using funnel plots ( [-,-]).

It was then possible to complete subgroup analyses to identify whether there was evidence for any BCTs acting as moderators of effect size. A shortlist of BCTs were identified upfront according to the transdiagnostic theory of EDs by Fairburn et al [,]. This was to avoid post hoc analysis of multiple BCTs, which would increase the likelihood of finding significant results through chance. If any of these prespecified BCTs were identified in >75% of effective interventions, they were included in the subgroup analyses: 2.2. Feedback on behavior, 2.3 Self-monitoring of behavior, 2.4 Self-monitoring of outcome(s) of behavior, 4.2 Information about antecedents, 7.7 Exposure, and 11.2 Reduce negative emotions. Additional related concepts were also explored, including mode of delivery (<5 vs ≥5 out of 12 possible sessions), TCS (high vs low), degree of therapist support (none or minimal vs some), and duration of therapy (<8 weeks vs ≥8 weeks). These factors were considered as they could contribute to heterogeneity and impact effect size.

Risk-of-Bias Assessment

The revised Cochrane Risk-of-Bias tool for randomized trials was used for assessing risk of bias in RCTs with studies assessed against 6 domains [] ( []). Risk-of-bias analysis was completed for all articles by PT, with over 20% (4/17) of the articles also being independently assessed by a second reviewer (TR). Disagreements were resolved via discussion. There was a high level of interrater agreement (interrater reliability [IRR]=0.9).

ResultsIncluded Studies

A PRISMA flow diagram () represents the literature search. A total of 17 RCT studies were identified for inclusion in this review.

‎Figure 1. PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram.

Of the 17 RCT studies identified, 12 (71%) included a WL comparator (or TAU), with 5 (21%) having active controls.

General Study Characteristics

The 17 studies included 12 (71%) parallel arm trials, 4 (24%) multiple-arm studies [,,,], and 1 (6%) cluster RCT []. Of these, 12 (71%) studies included active treatment compared to a WL control, informational control, or TAU, while 5 (29%) studies compared active treatments to other interventions, including F2F treatment [], day patient care [], and other digital treatment interventions [,,].

A total of 9 (53%) studies included all or nearly all female participants (>95%); 5 (29%) studies included 5% to 10% male participants, and 2 (12%) studies included >10% male participants. Ethnicity was not mentioned in 12 (71%) of the 17 studies, with 2 (12%) mentioning nationality but not ethnicity and only 3 (18%) providing any ethnic breakdown. Mean age ranged from 22.1 years [] to 43.2 years [] across studies, with participants aged between 17.3 and 55.5 years. The total number of participants overall was 5254, with 1956 included in the meta-analysis (WL and TAU studies only). Inclusion and exclusion criteria were highly variable, with some studies having clear diagnostic criteria that had to be met, excluding participants with comorbidities or with previous experience of inference-based CBT, while others permitted individuals to participate without meeting any diagnostic criteria, provided they were aged >16 years and had access to the internet. One study allowed participants to receive other forms of psychological, medical, or other treatment for their ED, whether in the digital intervention treatment arm or control condition [].

The studies took place in North America (2/17, 12%) [,], Europe (11/17, 65%; Switzerland, Germany, Sweden, Austria, and the Netherlands) [,,,,,-,,], and Australia or New Zealand (4/17, 24%) [,,,]. The included studies are listed in [-,-,-].

Summary of Intervention Types and Outcomes

The ED diagnoses included 6 studies focusing on binge eating disorder and binge eating symptoms [,,,,,], 3 studies on bulimia or eating disorders not otherwise specified [,,], and 8 studies concerning individuals with any ED symptoms [,-,-,]. The studies included a number of different interventions ( [-,-,-]), with the most common being Salut BED or Salut BN (5/17, 29%) [,,,,], Break Binge Eating or Break the Diet Cycle (4/17, 24%) [,,,], and Featback (2/17, 12%) [,].

Studies included internet and mobile-based digital interventions, frequently including messaging or email feedback or prompts. A total of 2 studies focused specifically on an app [,], 4 studies included blended internet and smartphone interventions [,,,], and 11 studies were internet-only interventions. Interventions lasted between 4 weeks and 12 months, with 11 interventions lasting ≤8 weeks and 6 interventions lasting >8 weeks [,,-,]. Interventions varied in the number of modules, ranging from 4 to 11, which resulted in differences in the amount of content provided and allowed for varying timescales to complete these modules.

Only studies with digital interventions with no or relatively minimal levels of therapist support (eg, weekly emails) as well as interventions with some therapist support were included. This resulted in 4 studies with no therapist involvement [,,,], 7 studies with minimal therapist involvement [,,,,,,], and 6 studies with some therapist involvement [,,,,,].

Outcome measures were most commonly the EDE-Q, although other measures such as the number of OBEs were also frequently reported. Dropout rates at postintervention measurement were between 6.7% and 58% for the digital intervention. They tended to be higher in the interventions with minimal or no support conducted in a community setting, such as those in which participants signed up and participated via an internet service [,,,,]. However, design characteristics such as feedback on behavior or feedback on outcomes of behavior also seemed important [].

The details of the digital interventions within the 17 studies, including their constituent BCTs, are described in .

Study Outcomes at Postintervention and Follow-Up

A total of 11 (92%) of the 12 RCTs that compared a digital intervention to a WL or TAU control demonstrated a significant improvement in ED outcomes (as measured by the EDE-Q) for the digital intervention over the control condition at postintervention, except for the study by Aardoom et al []. The WL and TAU control studies that reported the number of binge eating episodes at the postintervention time point (11/12, 92%) [-,,,,-,] also reported a significant reduction in OBEs compared to the WL and TAU control condition. All WL and TAU studies that reported follow-up data (9/12, 75%) reported a significant reduction in ED outcomes (EDE-Q total and OBEs) compared to the control condition, including the study by Aardoom et al [].

When the control condition was an active comparator, of traditional F2F treatment [] or a day patient program [], participants in the active comparator arm performed considerably better than the digital intervention at the postintervention time point, but results were comparable at follow-up in both studies. Where the active comparator was a similar digital health intervention, either broader in terms of functionality [] or consisting of interactive versus static content [,], there were no significant differences observed in EDE-Q total outcomes or secondary outcome measures at the postintervention time point (and no follow-up data).

BCTs in Effective Interventions

A total of 38 (41%) out of 93 BCTs were identified across the clinical content of the interventions (). The mean number of BCTs per intervention was 14 (SD: 2.57, range 9-18). The following BCTs were reported in >75% (13/17) of effective interventions: 2.3 Self-monitoring of behavior, 1.2 Problem-solving, 4.2 Information about antecedents, 2.2 Feedback on behavior, 2.4 Self-monitoring of outcomes of behavior, and 1.4 Action Planning. 7.7 Exposure and 11.2 Reduce negative emotions, which had been predicted to be important, were identified 56% (9/16) and 38% (6/16) of effective interventions, respectively. 5.2 Behavioral practice/rehearsal (10/16, 63%), 13.2 Framing/Reframing (10/16, 63%), and 7.1 Prompts/Cues (9/16, 56%) were present in >50% of effective interventions, suggesting they may also be important in supporting ED behavior change. The IRR was high (IRR=0.84).

BCTs were not identified from the following BCT categories in the taxonomy: 6. Comparison of the behavior, 14. Schedules consequences, or 16. Covert learning. Only 3 studies included a component from the 10. Reward and Threat category, the 10.4 Social Reward component.

Table 1. Behavior change techniques included in the treatment interventions (by studya)Evaluation of eating disorder studies[][][][][][][][][][][][][][][][][]AIb (n=17), n (%)EIc (at post intervention; n=16), n (%)2.3 Self-monitoring of behavior✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓✓17 (100)16 (100)1.2 Problem solving✓✓✓✓✓✓✓✓✓✓✓✓✓✓
✓✓16 (94)15 (94)4.2 Information about antecedents✓✓✓✓✓✓✓✓✓✓✓✓✓✓ ✓✓16 (94)15 (94)2.2 Feedback on behavior✓✓✓✓✓  ✓✓✓✓✓✓✓✓ ✓14 (82)13 (81)2.4 Self-monitoring of outcomes of behavior✓   ✓✓✓✓✓✓✓✓✓✓✓✓✓14 (82)13 (81)1.4 Action planning✓✓  ✓✓✓✓✓✓✓ ✓  ✓✓12 (71)12 (75)8.1 Behavioral practice or rehearsal✓ ✓  ✓✓✓ ✓  ✓ ✓✓✓10 (59)10 (63)13.2 Framing or reframing✓✓   ✓✓✓✓✓  ✓ ✓✓ 10 (59)10 (63)7.1 Prompts or cues   ✓ ✓✓✓✓✓✓✓ ✓   9 (52)9 (56)7.7 Exposure✓✓✓✓  ✓ ✓   ✓ ✓ ✓9 (52)9 (56)3.2 Social support (practical)✓ ✓✓✓      ✓ ✓✓ ✓8 (47)7 (44)4.1 Instructions on how to perform the behavior   ✓✓ ✓✓✓✓  ✓✓   8 (47)7 (44)5.1 Information about health consequences    ✓ ✓ ✓ ✓✓ ✓✓✓ 8 (47)7 (44)2.7 Feedback on outcomes of behavior        ✓ ✓✓✓✓✓ ✓7 (41)6 (38)3.1 Social support (unspecified)✓ ✓        ✓✓✓ ✓✓7 (41)6 (38)8.2 Behavior substitution   ✓   ✓ ✓✓   ✓✓ 6 (35)6 (38)11.2 Reduce negative emotions   ✓   ✓ ✓  ✓ ✓ ✓6 (35)6 (38)8.4 Habit reversal   ✓ ✓ ✓ ✓  ✓    5 (29.4%)5 (31.3%)5.6 Information about emotional consequences    ✓   ✓  ✓✓✓   5 (29)4 (25)8.3 Habit formation   ✓  ✓     ✓  ✓ 4 (25)4 (25)9.3 Comparison of future outcomes✓ ✓  ✓  ✓        4 (25)4 (25)1.1 Goal setting (behavior)   ✓✓✓           1 (18)3 (19)1.3 Goal setting (outcome)   ✓✓✓           1 (18)3 (19)9.2 Pros and cons✓ ✓  ✓           1 (18)3 (19)12.4 Distraction       ✓ ✓  ✓    1 (18)3 (19)15.4 Self-talk✓✓          ✓    1 (18)3 (19)3.3 Social support (emotional)   ✓       ✓ ✓   1 (18)2 (13)5.3 Information about social and environmental consequences           ✓ ✓✓  1 (18)2 (13)10.4 Social reward     ✓     ✓ ✓   1 (18)2 (13)5.4 Monitoring of emotional consequences✓ ✓              2 (12)2 (13)15.3 Focus on past success   ✓ ✓           2 (12)2 (13)15.1 Verbal persuasion about capability           ✓ ✓   2 (12)1 (6)1.5 Review behavioral goal     ✓           1 (6)1 (6)1.7 Review outcome (goal)     ✓           1 (6)1 (6)1.9 Commitment   ✓             1 (6)1 (6)2.1 Monitoring of behavior by others without feedback          ✓      1 (6)1 (6)4.4 Behavioral experiments ✓               1 (6)1 (6)13.4 Valued self-identity   ✓             1 (6)1 (6)

a Carrard et al [] (2011), Ruwaard et al [] (2013), de Zwaan et al [] (2017), Strandskov et al [] (2017), Wyssen et al [] (2021), Tregarthen et al [] (2019), Linardon et al [] (2022a), Linardon et al [] (2022b), Linardon et al [] (2021b), Linardon et al [] (2020), Melisse et al [] (2023), Rohrbach et al [] (2022), Fitzsimmons-Craft et al [] (2020), Aardoom et al [] (2016), Jacobi et al [] (2012), Högdahl et al [] (2023), Wagner et al [] (2013).

b AI: All interventions.

c EI: Effective interventions.

Follow-up data (>8 weeks after postintervention) was available for 9 (53%) out of the 17 studies. In 2 of the studies, there was no data available for the control condition because participants received the intervention. However, since the outcome effects at postintervention were sustained at follow-up, these studies were still included in the analysis [,]. A total of 2 studies included an active comparator [,], showing improvements on the EDE-Q for the digital intervention arm at postintervention that were sustained or improved at follow-up; hence, they were included in the analysis. This analysis () resulted in the following BCTs being identified in effective interventions at follow-up (in >75% of interventions): 2.2 Feedback on behavior, 2.3 Self-monitoring of behavior, 2.4 Self-monitoring of outcomes of behavior, 4.2 Information about antecedents, and 1.2 Problem-solving (these all were the same at the postintervention time point). The BCTs of 3.2 Social support (practical), 3.1 Social support (unspecified), and 5.1 Information about health consequences were more evident in the interventions that were effective at follow-up compared with the postintervention time point. These may be important in sustaining positive outcome effects; however, these findings are based on a small number of studies.

Definitions of the most common BCTs (included in at least 9/17, >50% of interventions), with examples of how they were implemented within the interventions, are included in .

Table 2. Behavior change techniques included in effective treatment interventions at follow-up (by study).Evaluation of eating disorder StudiesCarrard et al [] (2011)Ruwaard et al [] (2013)de Zwaan et al [] (2017)Wyssen et al [] (2021)Rohrbach et al [] (2022)Fitzsimmons-Craft et al [] (2020)Aardoom et al [] (2016)Jacobi et al [] (2012)Högdahl et al [] (2023)Effective (at follow-up; 9 studies had follow-up data), n (%)2.3 Self-monitoring of behavior✓✓✓✓✓✓✓✓✓9 (100)1.2 Problem solving✓✓✓✓✓✓✓ ✓8 (89)2.2 Feedback on behavior✓✓✓✓✓✓✓✓ 8 (89)4.2 Information about antecedents✓✓✓✓✓✓✓ ✓8 (89)2.4 Self-monitoring of outcomes of behavior✓  ✓✓✓✓✓✓7 (78)3.1 Social support (unspecified)✓ ✓ ✓✓✓ ✓6 (67)3.2 Social support (practical)✓ ✓✓✓ ✓✓ 6 (67)1.4 Action planning✓✓