Mediators of change in cognitive behavior therapy and interpersonal psychotherapy for eating disorders: A secondary analysis of a transdiagnostic randomized controlled trial

Abstract Objective Understanding the mechanisms of action of psychological treatments is a key first step in refining and developing more effective treatments. The present study examined hypothesized mediators of change of enhanced cognitive behavior therapy (CBT‐E) and interpersonal psychotherapy for eating disorders (IPT‐ED). Method A series of mediation studies were embedded in a randomized controlled trial (RCT) comparing 20 weeks of CBT‐E and IPT‐ED in a transdiagnostic, non‐underweight sample of patients with eating disorders (N = 130) consecutively referred to the service. Three hypothesized mediators of change in CBT‐E (regular eating, weighing frequency, and shape checking) and the key hypothesized mediator of IPT‐ED (interpersonal problem severity) were studied. Results The data supported regular eating as being a mediator of the effect of CBT‐E on binge‐eating frequency. The findings were inconclusive regarding the role of the other putative mediators of the effects of CBT‐E; and were similarly inconclusive for interpersonal problem severity as a mediator of the effect of IPT‐ED. Discussion This research highlights the potential benefits of embedding mediation studies within RCTs to better understand how treatments work. The findings supported the role of regular eating in reducing patients' binge‐eating frequency. Other key hypothesized mediators of CBT‐E and IPT‐ED were not supported, although the data were not inconsistent with them. Key methodological issues to address in future work include the need to capture both behavioral and cognitive processes of change in CBT‐E, and identifying key time points for change in IPT‐ED.


| INTRODUCTION
Understanding how psychological treatments work provides one of the strongest foundations for enhancing their potency (Kazdin & Nock, 2003). Without such understanding, it is unclear whether they work as hypothesized, or whether only some components are key to helping patients recover, while others are redundant. This problem is particularly acute in treatments that have several components and are implemented in a personalized manner.

Research into hypothesized mediators of IPT-ED has been limited.
This article describes findings from four mediation studies embedded within a transdiagnostic randomized controlled trial (RCT) of nonunderweight patients with eating disorders that compared enhanced cognitive behavior therapy (CBT-E) for eating disorders, a leading form of CBT-ED, and IPT-ED (Fairburn et al., 2015). These mediation studies drew on previous conceptual work outlining hypothesized mediators of both treatments and methodological considerations relevant to their investigation (Murphy et al., 2009). The present research limited its focus to hypothesized mediators of three core CBT-E procedures (regular eating, weekly weighing, shape checking) and their effects on the specific key eating disorder behavior and psychopathology they target (binge eating, weight concern, shape concern). It also examines what is arguably the key hypothesized mediator of IPT-ED; interpersonal problem severity. It was hypothesized that; CBT-E: 1. decreases binge-eating frequency through increasing regular eating; 2. decreases weight concern by decreasing weighing frequency; 3. decreases shape concern by decreasing shape checking;

| Design
Four mediation studies were embedded within an RCT comparing CBT-E (N = 65) and IPT-ED (N = 65) (Current controlled trials: ISRCTN 15562271). There was a closed 60-week follow-up during which patients received no other treatment unless clinically essential (see Fairburn et al., 2015).

| Sample
One hundred and thirty adult patients (98% female; 95% white; mean age 26 years) were recruited through consecutive referrals to an eating disorder clinic. Patients were assessed for DSM-IV eating disorder diagnosis (American Psychiatric Association, 1994), consented and randomized. Participants had a body mass index between 17.5 and 39.9 (inclusive), and had not previously received CBT-E or IPT-ED.
Fifty-three patients (41%) met diagnostic criteria for bulimia nervosa and 77 (59%) met criteria for eating disorder not otherwise specified, of whom eight (10%) had binge-eating disorder.

| Treatments
In both treatments patients attended one preparatory session (90 min), followed by 20 50-min individual sessions over 20 weeks and a review session 20 weeks after treatment had ended.

| Enhanced cognitive behavior therapy-Focused version
CBT-E (Fairburn, Cooper, Shafran, et al., 2008) has multiple procedures that are implemented sequentially in a flexible manner. Each procedure directly targets one or more specific features hypothesized to maintain the eating disorder. Key procedures in the first two stages of treatment are "regular eating" and "weekly weighing." "Regular eating" addresses one form of dietary restraint; delayed eating, which is hypothesized to maintain binge eating. It involves establishing a regular pattern of eating of three meals and two snacks a day (without changing the quantity or variety of food). "Weekly weighing" targets frequent weighing, which is hypothesized to maintain preoccupation and overevaluation of weight.
It involves in-session weekly weighing that is jointly interpreted with the therapist. In stage three, treatment procedures are individualized in accordance with the patient's personal formulation of the factors maintaining their psychopathology. For most patients, "shape checking" is a key procedure that addresses frequent shape checking, which is hypothesized to maintain preoccupation and overevaluation of shape.
It involves self-monitoring shape checking behaviors, evaluating their utility, and reducing their frequency.

| Interpersonal psychotherapy for eating disorders
IPT-ED (Fairburn, 1992(Fairburn, , 1997Murphy et al., 2012) is derived from IPT for depression (Klerman, Weissman, Rounsaville, & Chevron, 1984) and closely resembles it. Treatment targets the key interpersonal problem(s) thought to be maintaining the patient's eating disorder, focusing on one or two of the following key problem areas; interpersonal role disputes, role transitions, grief, interpersonal deficits, or life goals. Identified problem areas are examined in detail and ways to resolve them are discussed. Treatment is conceptualized as a unitary intervention to reduce eating disorder psychopathology, rather than a collection of separate procedures (Murphy et al., 2009).

| Therapists
The same therapists delivered both treatments. All received 6 months training in CBT-E and IPT-ED before starting the trial and were supervised weekly thereafter. Treatment sessions were recorded and audited to ensure treatment fidelity. A random audit confirmed high levels of fidelity (Fairburn et al., 2015).

| Procedure and measures
Each hypothesis was examined separately, with careful attention to the timing of measurement so as to detect change following implementation of treatment and to avoid confounding with other treatment procedures (Murphy et al., 2009). In CBT-E this focused on the specific timing of the implementation of each treatment procedure of interest while in IPT-ED treatment was assessed in a unitary fashion over a longer period (see Figure 1) because existing evidence shows that IPT-ED is slower to achieve its effects (Agras et al., 2000;Fairburn et al., 1991Fairburn et al., , 1993Fairburn et al., , 2015. Measures of all hypothesized mediators and outcomes were assessed in both treatments (see below). Weekly measures of eating disorder symptoms and behaviors were based on the Eating Disorder Examination (EDE; 16.0) clinical interview (Fairburn, Cooper, & O'Connor, 2008), which was administered at baseline, end of treatment, and 20-week, 40-week, and 60-week follow-up in the main trial (Fairburn et al., 2015).
2.6.1 | Hypothesis 1: Regular eating and bingeeating frequency Regular eating and binge-eating frequency were assessed during the first 4 weeks of treatment, when "regular eating" is a key focus of CBT-E and most changes attributable to this procedure are thought to occur (Fairburn, Cooper, Shafran, et al., 2008). Independent blind raters used patients' self-monitoring records to assess the following for each patient each week: i. Regular eating (hypothesized mediator), patient's adherence to a daily eating pattern of three meals plus two or three snacks, rated on a scale of "0" (absence of regular eating) to "6" (marked adherence to regular eating).
F I G U R E 1 Treatment and measurement timings in enhanced cognitive behavior therapy (CBT-E) and interpersonal psychotherapy for eating disorders (IPT-ED) ii. Binge-eating frequency (outcome), number of objective bingeeating episodes, defined as eating a substantially larger amount of food than most people would eat during a 2-hr period, with sense of control not assessed as the monitoring record captured only whether food intake was considered "excessive." Records were first rated separately by two raters, who then consulted to agree the final rating. Since it was hypothesized that the effect of the "regular eating" procedure might differ between patients who did and did not binge eat at baseline, this was assessed as a moderator (based on baseline EDE).
2.6.2 | Hypothesis 2: Weighing frequency and concern about weight Weighing frequency and concern about weight were assessed for the first 6 weeks of treatment, when "weekly weighing" is a key focus of treatment and before other treatment procedures are introduced that might impact on these variables (e.g., "shape checking", or other Stage 3 procedures that focus on increasing variety of food intake).
Patients self-reported each week: i. Weighing frequency (hypothesized mediator), the number of times they had weighed themselves.
Since it was hypothesized that the effect of the weekly weighing procedure might differ between patients who at baseline did and did not frequently weigh themselves, this was assessed as a moderator (using a once per week cut-off, the frequency of weighing in-session in CBT-E).
2.6.3 | Hypothesis 3: Shape checking and concern about shape Shape checking and concern about shape were assessed throughout treatment as the use and timing of this procedure was personalized to the patient. Only patients who received this procedure were included in the "intervention group," with the control group comprising both CBT-E patients who did not receive the procedure and IPT-ED patients.
Patients self-reported each week: i. Shape checking (hypothesized mediator), how often they had actively checked their body shape or size in; (a) mirrors, and (b) by pinching or measuring their body, rated on a scale of "1" (not at all) to "5" (many times a day). These were averaged into a single score.
ii. Concern about shape (outcome) rated on a scale of "0" (not at all) to "6" (markedly).

| Hypothesis 4: Interpersonal problem severity and eating disorder psychopathology
Interpersonal problem severity and eating disorder psychopathology were assessed at the beginning and end of treatment, and 20 and 40 weeks after treatment. This timeframe was chosen because it was hypothesized that a reduction in problem severity would take considerable time and that in turn this would lead to a progressive improvement in psychopathology (Murphy et al., 2009).
Three independent, blind raters conducted semistructured interviews to assess: i. Interpersonal problem severity (hypothesized mediator), the severity of the patient's key interpersonal problem (in IPT-ED this was the problem targeted in treatment), rated on a scale of "0" (no problem-no or minor difficulties, no impairment) to "6" (marked problem-difficulties occurring most of the time, substantial impairment).
ii. Eating disorder psychopathology (outcome), assessed using the global score from the EDE.
Assessors were trained and supervised by MO'C, an expert in the EDE.

| Data analysis
The hypothesized relationships between treatment, the hypothesized mediator and outcome were assessed using statistical mediation. A two-stage analytic strategy was used as described in Sivyer et al. (2020).
Stage 1 explored the effect that treatment had on change in the hypothesized mediator and outcome over time using multilevel modeling (Krull & MacKinnon, 1999, 2001. Random effects were fitted at the patient level and on time. Treatment, time, relevant predictors and moderators, and the interactions between treatment and time, treatment and the moderator, and treatment and the hypothesized mediator were included as independent variables. For personalized treatment procedures the week of implementation of the personalized procedure was also included as an independent variable . Stage 2 explored the relationships between treatment, the hypothesized mediator, and the outcome over time using autoregressive structural equation modeling (Cole & Maxwell, 2003), with relationships specified for treatment predicting the hypothesized mediator and outcome, and the hypothesized mediator predicting the outcome. The following effects were assessed at each time point: • The "indirect effect" of treatment on the outcome (i.e., the effect of treatment achieved via the hypothesized mediator and, at later time points, through change achieved earlier in treatment).
• The "total effect" of treatment on the outcome (i.e., the overall effect of treatment on the outcome, through both its direct effect on the outcome and its indirect effect through the hypothesized mediator and change achieved earlier in treatment).
For personalized treatment procedures, the model examined the first 6 weeks of implementation of the personalized treatment procedure compared to a similar time period in the control group (based on the average week of implementation in the intervention group) . Model fit was considered acceptable where; chisquare p > .05, comparative fit index (CFI) > .95, and root mean squared error of approximation (RMSEA) <0.08 with pclose <.05 (Kline, 2011). Standardized betas were used as a measure of effect size for both indirect and total effects. Where the results were consistent with the hypothesized model, a second, reversed model was run in which the outcome predicted the hypothesized mediator to verify the direction of these relationships.
Statistical mediation was concluded only if the results across both stages and the models within them were consistent regarding the hypothesized mediator mediating the effect of treatment on the outcome. As the analyses were investigating different, clearly specified hypotheses, correction for multiple testing was not applied (Perneger, 1998;Rothman, 1990

| RESULTS
Baseline descriptives and measure validity are reported in Table 1, with change over time shown in smoothed mean line graphs for each hypothesis (Figures 2-5). Key model coefficients for each hypothesis are reported in Table 2 The effect of CBT-E and IPT-ED on regular eating and bingeeating frequency did not differ between patients who were binge eating at baseline and those who were not, suggesting that binge-eating status was not a moderator.

| Autoregressive structural equation model: Indirect effects
There was a negative relationship between regular eating and bingeeating frequency in all treatment weeks, including baseline; however, the relationship over time was complex. Within the same treatment week greater regular eating was associated with lower binge-eating frequency as hypothesized, but between treatment weeks a positive relationship was observed, with higher levels of regular eating associated with increased binge eating the following week. Overall, there was evidence of an indirect effect of CBT-E decreasing binge-eating frequency via regular eating. This effect was found immediately following the implementation of the "regular eating" procedure in Week 2 of treatment, and was still present at Week 4. The reversed model assessing whether change in regular eating was better explained by the direct effect that CBT-E had on binge-eating frequency, rather than the other way around, supported regular eating as a mediator.
3.2 | Hypothesis 2: Weighing frequency and concern about weight 3.2.1 | Multilevel models: Change during treatment This rate of change slowed by Week 6. Change in the hypothesized mediator appeared to have little observable relationship to change in the outcome in either treatment (Figure 3).

| Autoregressive structural equation model: Indirect effects
There was a positive relationship between weighing frequency and concern about weight at baseline in the direction hypothesized, with more frequent weighing associated with increased concern about weight. However, the relationship between these variables was inconsistent during treatment. Although there was a trend toward CBT-E decreasing concern about weight via decreased weighing frequency in frequent weighers, this did not reach statistical significance in either frequent or nonfrequent weighers at any week, despite CBT-E having a greater effect in decreasing weighing frequency. 3.3 | Hypothesis 3: Shape checking and concern about shape 3.3.1 | Multilevel models: Change during treatment The shape checking procedure was used with 53 patients in CBT-E (82%). On average, it was implemented in Treatment Week 8 (mean = 7.66, SD = 2.01).
Prior to its implementation, shape checking initially remained stable in both CBT-E and IPT-ED. As treatment progressed patients in CBT-E gradually started to increase their shape checking. However, following implementation of the shape checking procedure, those patients in CBT-E who received the intervention began to decrease their shape checking over time.
Concern about shape decreased in both IPT-ED and CBT-E. However, for patients in the shape checking intervention group, concern about weight decreased even further following implementation of the shape checking procedure. Change in concern about shape slowed toward the end of treatment across all groups. Shape checking and concern about shape had a similar trajectory in those who received the shape checking procedure (Figure 4).

| Autoregressive structural equation model: Indirect effects
There was a positive relationship between shape checking and concern about shape at baseline as hypothesized, with higher shape checking scores associated with increased concern about shape. However, the relationship between these variables was inconsistent during treatment.
There was a trend toward an indirect effect of the shape checking procedure decreasing concern about shape via decreased shape checking; however, this did not reach statistical significance. Instead, there was a statistically significant indirect effect for the shape checking procedure temporarily increasing concern about shape via increased shape checking immediately following implementation of the shape checking procedure (Week 1 of the intervention). However, the effect size was negligible.

| DISCUSSION
This study investigated hypothesized mechanisms of CBT-E and IPT-ED using data from an RCT. Three hypothesized mediators of CBT-E (regular eating, weighing frequency, and shape checking) and the key hypothesized mediator of IPT-ED (interpersonal problem severity) were examined. Of the three hypothesized mediators of CBT-E examined, only regular eating was consistent with it being a mediator of the effect of CBT-E on binge-eating frequency. The findings were inconclusive for the other two hypothesized mediators of CBT-E, and for the hypothesized mediator of IPT-ED.
Regular eating has consistently been associated with decreased frequency of binge eating (Ellison et al., 2016;Shah, Passi, Bryson, & Agras, 2005;Waller, Evans, & Pugh, 2013;Wilson, Fairburn, Agras, Walsh, & Kraemer, 2002;Zendegui, West, & Zandberg, 2014). However, this is the first study to assess change in both variables on a weekly basis. It is unlikely that these findings are due to patients increasing food quantity or variety as patients were advised to change only their eating pattern (see Fairburn, Cooper, Shafran, et al., 2008).
Nonetheless, some patients may have made further dietary changes.
Overall the results supported the hypothesized negative relationship between regular eating and binge-eating frequency during the first 4 weeks of treatment and within the same treatment week; however, there was also a positive relationship between regular eating and binge-eating frequency between treatment weeks. This may reflect lapses in regular eating and binge eating between treatment weeks, falsely making it look like high levels of regular eating the previous week were associated with higher levels of binge eating the following week. Further research should explore potential lapses during treatment in binge eating, possibly using an ecological momentary assessment design with more frequent measurement of these behaviors.
Binge-eating frequency decreased at different rates in the two treatments. In IPT-ED change in binge-eating frequency appeared to occur independently of change in regular eating, which remained stable during the period studied. This suggests that the effect of IPT-ED on binge-eating frequency may be mediated via a different process, as would be hypothesized by the different models underpinning these treatments.
Only one study has investigated the role of frequent weight and shape checking in CBT-E, which, consistent with the CBT-E model, found that decreased body checking was associated with decreased concern about shape in inpatients with anorexia nervosa (Calugi, El Ghoch, & Dalle Grave, 2017). The current research did not replicate these findings. This may be due to sample differences (inpatients vs. outpatients, underweight vs. non-underweight), or issues with statistical power.
A temporary increase in concern about shape has been suggested as a potential side-effect of monitoring shape checking (Fairburn, Cooper, Shafran, et al., 2008). Many patients are unaware of how frequently they check their shape and bringing these behaviors into conscious awareness can be distressing. The temporary increase in concern about shape found in this study is consistent with this. Raised awareness of these behaviors may also explain the increased selfreporting in shape checking. Although differences might be expected between patients who shape check and those who do not, in general patients self-reported high levels of shape checking at baseline, with 63% (N = 82) checking their shape at least daily. Very few did not check their shape (N = 9; 7%). However, more research is needed to examine other relevant behaviors such as shape avoidance. T A B L E 2 Key coefficients of multilevel (MLM) and structural equation (SEM) models examining the impact of treatment on the hypothesized mediator and outcome, and the indirect and total effects of treatment on the outcome for all hypotheses in the first week following implementation and the last time point studied Hypothesis 3: Shape checking frequency as a mediator of the effect of CBT-E on concern about shape in the shape checking intervention group vs. control group (including patients in IPT-ED and those in CBT-E who did not receive the shape checking intervention) Both "weekly weighing" and "shape checking" are complex procedures addressing behaviors and cognitions. This study measured only the behavioral aspect of these procedures (i.e., frequency of weighing or shape checking). As such, it may have missed important cognitive changes affecting concern about weight and shape. For example, more frequent weighing would not necessarily increase concern about weight in the presence of more benign interpretations of weight fluctuations. Future work should consider assessing both cognitive and behavioral processes associated with these procedures, and other hypothesized mediators (e.g., managing moods/events).

MLM
As there was little change in interpersonal problem severity and eating disorder psychopathology in both treatments during followup, it was difficult to draw conclusions regarding the mediational relationship between these variables. There has been relatively little research into the mechanisms of IPT-ED, however, previous research has identified that interpersonal problems improve in both CBT-ED and IPT-ED (Fairburn et al., 1993;Wilfley et al., 2002). It is unclear whether this is the result of improving interpersonal problem severity, improving eating disorder psychopathology, or both. The fact that both IPT-ED and CBT-E improve interpersonal problem severity, whether directly or indirectly, may explain why examining IPT-ED as a secondary treatment for nonresponders to CBT-ED has not had much success (Mitchell et al., 2002). Further work should examine change in interpersonal problem severity and eating disorder psychopathology during treatment, rather than at follow-up to gain a better understanding of these processes and use more frequent measurement.

| Strengths and limitations
This research used some unvalidated measures. As such little is known their psychometric properties, although preliminary analyses suggest moderate-strong correlations with equivalent items on the EDE at All models controlling for relevant predictors and moderators. SEM models also controlling for shared measurement variance. Confidence intervals are bias-corrected based on 1,000 bootstrap resamples. p values based on uncorrected 95% confidence intervals. Indirect and total effects of treatment incorporate transmission of treatment effects via autoregressive paths. b Indicates the presence of a nonlinear trend suggesting that effects plateau toward the end of the treatment period studied. c Indicates that this was compared to nonfrequent weighers in IPT-ED.
baseline and interrater reliability for the interpersonal interview was good. Binge-eating frequency did not assess sense of loss of control, and hence "objective overeating" was used as a proxy for objective binge eating.
The sample size (N = 130) was relatively small for the structural equation models used. It is likely that tests of the indirect effect were underpowered, which is a common problem in such analyses (MacKinnon, Lockwood, Hoffman, West, & Sheets, 2002). However, there is a lack of definitive research into the sample size requirements for longitudinal statistical mediation (Cole & Maxwell, 2003;Little, 2013). Although personalization is a clinical strength of these treatments, there has been little research regarding how to examine mediational processes under such conditions. Further work is needed to identify the best methods.
Although research suggests that early change during treatment consistently predicts better outcomes (Linardon, de la Piedad Garcia, & Brennan, 2017), this research did not examine whether change in the hypothesized mediators was maintained or was related to longer-term outcomes.
A key strength was that the mediation studies were embedded within a carefully designed RCT, a design that has been underutilized in assessing how treatments work (Dunn et al., 2015;. Treatment fidelity was high (Fairburn et al., 2015). Randomization ensured that the analyses were less likely to be influenced by confounders, at least in the case of the relationship between treatment and mediator, and treatment and outcome for the nonpersonalized elements of the treatments.
The hypothesized mediator and outcome variables were both measured throughout the period during which change is likely to occur.
This was planned a priori to deal with the challenge of ensuring that changes in the hypothesized mediators occurred prior to changes in the relevant outcomes in a treatment where rapid mediational effects are likely. This enabled in-depth, longitudinal examination of specific treatment procedures and the processes they were hypothesized to affect. Case-by-case extraction of data in order to deal with the challenge of treatment personalization was a further strength.

| CONCLUSIONS
The findings add further support for the clinical importance of the