Predictors and Moderators of Response to Cognitive Behavioral Therapy and Medication for the Treatment of Binge Eating Disorder

Abstract

Objective

To examine predictors and moderators of response to cognitive-behavioral therapy (CBT) and medication treatments for binge-eating disorder (BED).

Method

108 BED patients in a randomized double-blind placebo-controlled trial testing CBT and fluoxetine treatments were assessed prior, throughout-, and post-treatment. Demographic factors, psychiatric and personality-disorder co-morbidity, eating-disorder psychopathology, psychological features, and two sub-typing methods (negative-affect, overvaluation of shape/weight) were tested as predictors and moderators for the primary outcome of remission from binge-eating and four secondary dimensional outcomes (binge-eating frequency, eating-disorder psychopathology, depression, and body mass index). Mixed-effects-models analyzed all available data for each outcome variable. In each model, effects for baseline value and treatment were included with tests of both prediction and moderator effects.

Results

Several demographic and clinical variables significantly predicted and/or moderated outcomes. One demographic variable signaled a statistical advantage for medication-only (younger participants had greater binge-eating reductions) whereas several demographic and clinical variables (lower self-esteem, negative-affect, and overvaluation of shape/weight) signaled better improvements if receiving CBT. Overvaluation was the most salient predictor/moderator of outcomes. Overvaluation significantly predicted binge-eating remission (29% of participants with versus 57% of participants without overvaluation remitted). Overvaluation was especially associated with lower remission rates if receiving medication-only (10% versus 42% for participants without overvaluation). Overvaluation moderated dimensional outcomes: participants with overvaluation had significantly greater reductions in eating-disorder psychopathology and depression levels if receiving CBT. Overvaluation predictor/moderator findings persisted after controlling for negative-affect.

Conclusions

Our findings have clinical utility for prescription of CBT and medication and implications for refinement of the BED diagnosis.

Binge-eating disorder (BED), a research category in the DSM-IV (American Psychiatric Association, 1994), is characterized by recurrent binge-eating accompanied by feelings of loss of control and marked distress in the absence of inappropriate weight compensatory behaviors. BED is prevalent (Hudson, Hiripi, Pope, & Kessler, 2007) and strongly associated with obesity, elevated psychiatric and medical co-morbidity, and psychosocial impairment (Grilo, White, & Masheb, 2009; Hudson et al., 2007). BED differs from obesity and other eating disorders on numerous clinical markers (Allison, Grilo, Masheb, & Stunkard, 2005; Grilo et al., 2008; Grilo, Masheb, & White, 2010).

Although some effective medication (Reas & Grilo, 2008) and psychological (Wilson, Grilo, & Vitousek, 2007) treatments have been identified for BED, even in studies with the best outcomes, a substantial proportion of patients do not achieve abstinence from binge-eating, and most clinical trials have reported little to no weight loss (Wilfley et al., 2002; Wilson, Wilfley, Agras, & Bryson, 2010). Thus, it is important to find ways to predict BED patients’ response to treatments as this may lead to more effective decision-making about treatment prescriptions (Kraemer, Wilson, Fairburn, & Agras, 2002).

Relatively few data have been published on predictors of outcome for BED and critical reviews have highlighted the significant methodological shortcomings of most of this literature (Berkman, Lohr, & Bulik, 2007). Early studies identified several factors that may be associated with poor outcome. In some studies, severity of binge-eating (Agras et al., 1995; Agras, Telch, Arnow, Eldredge, & Marnell, 1997; Loeb, Wilson, Gilbert, & Labouvie, 2000; Peterson et al., 2000), earlier onset of BED (Agras et al., 1995), and younger age at time of presentation for treatment (Agras et al., 1997) predicted worse binge-eating outcomes, whereas psychiatric co-morbidity was generally unrelated to treatment outcomes (Wilfley et al., 2000). Recent studies, which performed more sophisticated and systematic analyses, have identified some additional potential patient predictors of treatment outcome for BED although they have generally failed to replicate many of the potential predictors identified in the first-generation studies. Hilbert and colleagues (2007) reported that greater interpersonal problems and greater shape/weight concerns predicted poorer treatment outcomes. Masheb and Grilo (2008a) reported that binge-eating frequency, eating-disorder psychopathology, and depressive levels significantly predicted their respective post-treatment levels and that personality disorders predicted higher levels of eating-disorder psychopathology and depressive levels at post-treatment. Wilson et al (2010) reported that education, depression levels, and self-esteem predicted binge-eating remission.

Thus, relatively little is known about predictors of treatment outcome for BED and overall the best conclusion is that reliable predictors have yet to be identified. Even less is known about treatment moderators for BED. We are aware of six published reports of analyses of moderators (Hilbert et al., 2007; Masheb & Grilo, 2008a, Masheb & Grilo, 2008b; Robinson & Safer, 2011; Sysko, Hildebrandt, Wilson, Wilfley, & Agras, 2010; Wilson et al., 2010). Hilbert and colleagues (2007) explored numerous clinical variables but failed to find any significant moderators for CBT and interpersonal psychotherapy (IPT) delivered via intensive group formats. Similarly, Masheb and Grilo (2008a) tested numerous variables (age, psychiatric and personality-disorder co-morbidity, binge-eating, eating-disorder psychopathology, self-esteem, and depressive levels) and failed to find any significant moderator effects in a controlled trial comparing CBT and behavioral weight loss (BWL) treatments delivered via guided-self-help (i.e., CBTgsh and BWLgsh). Masheb and Grilo (2008b), in a further re-analysis of moderation effects found that two different sub-typing methods – grouping by negative-affect (Grilo, Masheb, & Wilson, 2001a; Stice & Agras, 1999; Stice & Fairburn, 2003) or by overvaluation of shape/weight (Grilo et al., 2008; Hrabosky, Masheb, White, & Grilo, 2007) – predicted worse outcomes but did not significantly moderate BED treatment outcomes. Robinson and Safer (2011) tested a broad range of demographic, psychiatric, and clinical variables and reported that the only factors that moderated better response to dialectical behavior therapy versus an active comparison group therapy were the presence of personality-disorder pathology and earlier onset of overweight and dieting. Wilson and colleagues (2010) tested numerous clinical variables including negative-affect sub-typing and reported that two variables - self-esteem and global eating-disorder psychopathology - significantly moderated treatment outcomes; patients with lower self-esteem and higher eating-disorder psychopathology benefiting more from IPT than from BWL or CBTgsh. Lastly, Sysko and colleagues (2010), in a further analysis of moderation effects in the Wilson et al (2010) trial, found that heterogeneity significantly moderated outcomes. Specifically, latent-class-analysis identified two classes of patients characterized by greater severity of binge-eating, shape/weight concerns, and depression levels, that were significantly more likely to remit from binge-eating if receiving IPT or CBTgsh.

In the present study, we examined predictors and moderators of response to individual CBT and antidepressant medication for BED. This study is important for several reasons. CBT is considered the best-established treatment for BED (NICE, 2004; Wilson et al., 2007). Findings from Masheb and Grilo (2008) and Wilson et al (2010) pertain to CBTgsh (not traditional CBT) and are mixed. Antidepressant medications are considered an acceptable alternative treatment for BED (NICE, 2004). Although some studies have questioned the clinical-effects of antidepressants for BED (Ricca et al., 2001; see: Reas & Grilo, 2008), these medications are frequently and widely prescribed. If moderators of treatment response to CBT versus antidepressant medication could be identified, this would inform rationale treatment prescription.

Method

Participants

Participants were 108 consecutively evaluated adult patients who met DSM-IV (APA, 1994) research criteria for BED and participated in a randomized double-blind placebo-controlled study of CBT and fluoxetine treatments alone and in combination (balanced two-by-two factorial design). A detailed description of the study and primary outcomes has been reported elsewhere (Grilo, Masheb, & Wilson, 2005) but will be briefly described here.

Participants were required to be aged 18 to 60 years and at least 100% of ideal weight. Exclusion criteria were: any concurrent treatment for eating, weight, or psychiatric problems; medical conditions (diabetes, thyroid problems) that influence eating or weight; severe current psychiatric conditions requiring different treatments (psychosis, bipolar disorder, current substance dependence); and pregnancy or lactation. The study was approved by the Yale institutional review board. All participants provided written informed consent.

The 108 participants were aged 21 to 59 years (mean = 44.0, SD = 8.6), 78% (N = 84) were female, and 87% (N = 95) had at least completed high school. The participant group was 89% (N = 96) Caucasian, 8% (N = 9) African American, and 3% (N = 3) Hispanic American. Mean body mass index (BMI; weight (kg) divided height (m²)) was 36.3 (SD = 7.9).

Diagnostic Assessment and Baseline Measures

Diagnostic procedures were conducted by trained and monitored doctoral-level research-clinicians. DSM-IV (APA, 1994) Axis I psychiatric and Axis II personality disorder diagnoses were based on the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I/P; First, Spitzer, Gibbon, & Williams, 1996) and the Diagnostic Interview for DSM-IV Personality Disorders (DIPD-IV; Zanarini et al., 1996), respectively. Inter-rater reliability for the SCID-I/P and DIPD-IV was good, with kappa coefficients (Cohen, 1960) ranging from 0.57 to 1.0.

The BED diagnosis assessed with the SCID-I/P was confirmed on the Eating Disorder Examination Interview -12th Edition (EDE; Fairburn & Cooper, 1993). The EDE, an investigator-based interview, was administered at baseline to assess the features of eating disorders. The EDE focuses on the previous 28 days except for diagnostic items, which are rated for the required six-month duration. The EDE assesses the frequency of different forms of overeating, including objective bulimic episodes (OBEs; i.e., binge eating defined as unusually large quantities of food with a subjective sense of loss of control). The EDE is also comprised of four subscales and generates a total score which reflects global severity. Items are rated on 7-point forced-choice scales (0–6), with higher scores reflecting greater severity or frequency. The EDE has established inter-rater and test-retest reliability (Grilo, Masheb, Lozano-Blanco, & Barry, 2004). In the present study, inter-rater reliability coefficient for OBEs was .98.

Baseline and Repeated Measures

Several measures were administered at baseline, monthly during treatment, and at the end of treatment. The Eating Disorder Examination-Questionnaire (EDE-Q; Fairburn & Beglin, 1994), the self-report version of the EDE, has received support for use with BED (Grilo, Masheb, & Wilson, 2001b; 2001c). Studies have found the EDE-Q to have good test-retest reliability (Reas, Grilo, & Masheb, 2006) and convergent validity with the EDE interview in BED (Grilo et al., 2001b, 2001c); it also performs well as a measure of change in treatment trials (Sysko, Walsh, & Fairburn, 2005). The Beck Depression Inventory (BDI; Beck & Steer, 1987) 21-item version is a well-established inventory of depression levels (Beck, Steer, & Garbin, 1998) and negative affect (Grilo, Masheb, & Wilson, 2001c).

Weight and height

Weight and height were measured at baseline evaluation and again immediately prior to beginning treatment using a medical balance-beam scale. Weight was then measured monthly during the course and at the end of treatment. BMI was calculated from these measurements.

Randomization and Treatment Conditions

Participants were randomized to one of four treatments (balanced 2-by-2 factorial design) for 16 weeks: (1) fluoxetine (60 mg/day); (2) placebo; (3) CBT and fluoxetine (60 mg/day); or (4) CBT and placebo. The randomization schedule was computer-generated without restriction or stratification created in blocks of eight to ensure approximately equal allocation to treatments. Each randomization was made after completion of all assessments and acceptance of the participant into study. The randomization schedule was maintained by the research pharmacist and conducted separately from (blinded) investigators.

Medication Treatment

Medication treatment conditions were administered in double-blind placebo-controlled fashion and consisted of minimal clinical management focused on the medication regimen without any additional psychotherapeutic intervention. Participants were asked to take three pills each morning for 16 weeks. The blinded medications were fixed-dose throughout the study consisting of either fluoxetine (60 mg/day) or placebo, prepared in identical-appearing capsules.

Cognitive-Behavioral Therapy (CBT)

CBT was administered in weekly individual 60-minute sessions for 16 weeks by doctoral-level research clinicians following a widely-used (see NICE, 2004) published manual (Fairburn et al., 1993). The research-clinicians were experienced with CBT and BED patients, received extensive training, and were carefully monitored (via audiotapes of sessions and regular supervision) throughout the study by the investigators.

Overview of Analyses

Predictors and moderators of treatment outcome were evaluated in the context of mixed-effects models conducted using SPSS Version 19. Outcome variables included one primary categorical outcome (remission from binge-eating, defined as zero OBEs for the previous 28 days as measured by the EDE-Q) and changes in four dimensional outcomes: frequency of binge-eating (number of OBEs on the EDE-Q), eating-disorder psychopathology (total global score on the EDEQ), depression level (BDI total score), and BMI. The mixed-effects model for remission from binge-eating was based on a generalized linear model with a binary logit response function, the model for binge-eating frequency was based upon a generalized linear model with negative binomial with log link response function, and the models for eating-disorder psychopathology, depression level, and BMI were based on the general linear model.

Predictors/ moderators that were evaluated included demographic characteristics (age, gender, education), eating-disorder characteristics (age of onset, OBE frequency, EDE total score), associated psychopathology (axis I psychiatric disorders, axis II personality-disorders), self-esteem (RSE total), and depression level (BDI total). In addition, two sub-typing methods (grouping by negative affect subtype, overvaluation of shape/weight) used in one previous study of predictors and moderators (Masheb & Grilo, 2008b) were also tested. Both of these sub-typing methods have been found to identify meaningful categorizations of BED patients, with patients sub-typed as negative-affect type (versus dietary type) and with overvaluation (versus without overvaluation) representing more disturbed variants of BED; such subtyping may have greater prognostic significance than traditional approaches using individual variables. For the first sub-typing method, participants were categorized as either negative affect or dietary subtype using cluster-analysis of EDE restraint, BDI, and RSE scores following the approach used in several studies (Grilo, Masheb, & Wilson, 2001a; Stice & Agras, 1999; Stice & Fairburn, 2003). SPSS Quick Cluster algorithm was used to group participants based on similarity in levels of the four selected variables as in previous studies. Quick Cluster selects k clusters, with well-separated non-missing values as initial centers, and then iteratively clustered participants into one of the two groups on the basis of squared Euclidean distances. For the second sub-typing method, participants were also categorized as having overvaluation of shape/weight based on scores of four or greater on relevant EDE interview items following the approach used in several studies (Grilo et al., 2008, Grilo et al., 2010; Hrabosky et al., 2007). The items assess how shape and/or weight influence how someone feels (i.e., judges, thinks, evaluates) themselves as a person (Fairburn & Cooper, 1993). To be categorized with overvaluation, shape/weight must be considered to be more important than virtually anything else (e.g., work performance, being a parent, marriage, friendships, etc) in the person’s scheme for self-evaluation (Fairburn & Cooper, 1993; Grilo et al., 2008)¹.

Baseline values of outcome measures were evaluated as predictors/moderators of other outcomes. All predictors were centered around the grand mean prior to analyses. All models included an autocorrelation term to account for serial dependency among within-person observations based on a first-order autoregressive (AR1) component. Mixed-effects models used all available data after baseline. All models included a random intercept, and fixed effects for baseline observation, treatment group (medication-only (active drug or placebo without CBT; N=54) versus CBT (with active drug or placebo; N=54)), assessment time-point, and the predictor variable. Models were constructed to evaluate simultaneously whether a given variable predicted or moderated either the overall amount of change or the rate of change in the outcome. Prediction of the amount of change was based on the main effect of the predictor variable, while prediction of the rate of change was based on the predictor-by-assessment month interaction. Moderation of the amount of change was based on the group-by-predictor, while the moderation of the rate of change was based on the group-by-predictor-assessment time-point interaction.

Results

Table 1 summarizes the predictor and moderator analyses for the primary categorical outcome of binge-eating remission. Results for the predictor and moderator analyses for the dimensional outcomes are summarized in Table 2 (for frequency of binge-eating), Table 3 (for eating-disorder psychopathology reflected by EDE-Q global score), Table 4 (for depression level reflected by BDI score), and Table 5 (for BMI). A summary of key significant findings follows.

Table 1.

Patient Characteristics and Pre-treatment Clinical Characteristics Predicting and Moderating Remission from Binge-Eating.

Predictor	Remission (Abstinence from Binge-Eating)
	Predicts				Moderates
	Amount		Rate		Amount		Rate
	Est.	95% CI	Est.	95% CI	Est.	95% CI	Est.	95% CI
Age	.042	−.016, .101	−.014	−.040, .013	.022	−.066, .110	.015	−.026, .056
Female	.125	−.994, 1.245	.114	−.343, .571	−.642	−2.42, 1.13	.243	−.538, 1.03
Education > H.S.	1.31*	.145, 2.47	−.130	−.659, .399	.414	−1.93, 2.76	−.112	−.710, .486
Age of BED onset	.050**	.012, .088	.003	−.021, .027	.020	−.045, .084	−.014	−.047, .019
Body Mass Index	.009	−.056, .073	−.011	−.036, .014	−.080	−.177, .018	.013	−.019, .045
Psychiatric Disorder	.017	−.988, 1.02	.137	−.336, .610	.859	−.643, 2.36	.173	−.518, .865
Personality Disorder	−.682	−1.72, .354	−.088	−.598, .422	.474	−1.12, 2.07	.293	−.401, .987
Self-Esteem	−.034	−.114, .045	.025	−.012, .061	.071	−.053, .195	−.010	−.062, .041
EDE Global	−.015	−.037, .007	.005	−.005, .015	−.011	−.042, .019	−.010	−.025, .004
Depression (BDI)	−.002	−.054, .050	−.004	−.023, .016	−.037	−.131, .057	−.005	−.036, .025
Negative-Affect	−.638	−1.60, .325	−.050	−.559, .459	1.83	−.527, 4.18	.290	−.289, .869
Overvaluation	.641**	.118, 1.16	.182	−.268, .632	1.22	−.348, 2.78	.121	−.579, .822

Note. BED = binge eating disorder; EDE = Eating Disorder Examination; BDI = Beck Depression Inventory

^*p < .05,

^**p < .01,

^***p < .001

Table 2.

Patient Characteristics and Pre-treatment Clinical Characteristics Predicting and Moderating Changes in Binge-Eating Frequency.

Predictor	Binge-Eating Frequency
	Predicts				Moderates
	Amount		Rate		Amount		Rate
	Est.	95% CI	Est.	95% CI	Est.	95% CI	Est.	95% CI
Age	.013	−.017, .043	.018**	.005, .030	−.021	−.065, .023	−.020**	−.035, −.005
Female	.341	−.405, 1.086	.259	−.026, .544	−.750	−1.70, .197	−.448**	−.759, −.137
Education > H.S.	−.374	−1.34, .587	−.028	−.550, .494	−1.31	−2.76, .141	.218	−.340, .777
Age of BED onset	−.041**	−.070, −.011	−.014*	−.027, −.002	.017	−.020, .055	.018**	.004, .031
Body Mass Index	−.005	−.055, .044	.001	−.023, .025	.027	−.029, .084	−.001	−.026, .025
Psychiatric Disorder	−.090	−.774, .594	−.174	−.422, .073	.079	−.868, 1.026	.197	−.120, .514
Personality Disorder	.016	−.708, .741	.001	−.407, .409	−.050	−.982, .882	−.100	−.538, .338
Self-Esteem	−.002	−.050, .046	−.008	−.025, .009	−.009	−.074, .057	.011	−.013, .034
EDE Global	.006	−.010, .021	.001	−.007, .009	.002	−.016, .021	−.002	−.011, .006
Depression (BDI)	.011	−.021, .042	.007	−.004, .018	.006	−.040, .051	−.008	−.023, .008
Negative-Affect	.048	−.676, .771	−.103	−.470, .264	−.444	−1.38, .493	.078	−.319, .476
Overvaluation	−.097	−.748, .554	−.036	−.331, .260	−.321	−1.13, .492	.053	−.294, .400

Note. BED = binge eating disorder; EDE = Eating Disorder Examination; BDI = Beck Depression Inventory

^*p < .05,

^**p < .01,

^***p < .001

Table 3.

Patient Characteristics and Pre-treatment Clinical Characteristics Predicting and Moderating Changes in Eating-Disorder Psychopathology (EDEQ Global Score).

Predictor	Eating-Disorder Psychopathology (EDEQ Global Score)
	Predicts				Moderates
	Amount		Rate		Amount		Rate
	Est.	95% CI	Est.	95% CI	Est.	95% CI	Est.	95% CI
Age	.048	−.634, .729	−.131	−.308, .046	−.222	−1.15, .709	.043	−.200, .286
Female	−3.08	−16.4, 10.3	−.301	−3.81, 3.20	7.06	−12.0, 26.1	.825	−4.12, 5.77
Education > H.S.	6.39	−7.92, 20.7	−1.67	−5.40, 2.06	−4.75	−30.3, 20.8	4.64	−1.98, 11.2
Age of BED onset	−.288	−.780, .205	−.200**	−.325, −.074	.032	−.675, .740	.188*	.004, .372
Binge-eating frequency	−25.7*	−47.6, −3.74	−3.85	−9.85, 2.14	31.7*	2.32, 61.1	4.28	−3.58, 12.1
Body Mass Index	.345	−.429, 1.11	.110	−.090, .311	−.467	−1.49, .553	.019	−.249, .288
Psychiatric Disorder	5.46	−7.71, 18.6	.557	−2.98, 4.10	−17.2	−34.9, .505	−1.74	−6.50, 3.02
Personality Disorder	3.09	−9.25, 15.4	2.93	−.297, 6.16	−12.0	−28.9, 4.91	−2.39	−6.88, 2.10
Self-Esteem	−.071	−1.00, .860	−.003	−.239, .234	−.687	−1.94, .563	.019	−.312, .351
Depression (BDI)	−.184	−.756, .388	−.028	−.169, .114	.675	−.145, 1.50	.113	−.105, .331
Negative Affect	9.58	−3.43, 22.6	1.46	−1.76, 4.69	−17.4*	−34.7, −.020	−4.57	−9.24, .097
Overvaluation	−.827	−12.2, 10.5	2.09	−.886, 5.06	−18.1*	−33.3, −2.81	−3.75	−7.95, .456

Note. BED = binge eating disorder; EDEQ = Eating Disorder Examination-Questionnaire; BDI = Beck Depression Inventory

^*p < .05,

^**p < .01,

^***p < .001

Table 4.

Patient Characteristics and Pre-treatment Clinical Characteristics Predicting and Moderating Changes in Depression Levels as reflected by Beck Depression Inventory (BDI) Scores.

Predictor	Depression Levels (BDI Total Score)
	Predicts				Moderates
	Amount		Rate		Amount		Rate
	Est.	95% CI	Est.	95% CI	Est.	95% CI	Est.	95% CI
Age	.134	−.066, .334	−0.68*	−.123, −.013	−.279*	−.554, −.003	.049	−.027, .125
Female	2.01	−2.07, 6.10	.066	−1.03, 1.16	−2.70	−8.47, 3.08	.558	−.992, 2.11
Education > H.S.	−.746	−5.05, 3.56	.024	−1.15, 1.20	−1.29	−9.15, 6.58	.326	−1.76, 2.41
Age of BED onset	−.041	−.190, .108	−.024	−.064, .016	−.075	−.287, .137	.009	−.050, .068
Binge-eating frequency	−1.89	−8.61, 4.84	.855	−1.03, 2.74	3.15	−5.86, 12.2	−1.60	−4.06, .871
Body mass Index	−.034	−.270, .202	.025	−.039, .088	.116	−.194, .426	−.028	−.113, .057
Psychiatric Disorder	.569	−3.51, 4.65	−.216	−1.32, .891	−3.67	−9.00, 1.66	−.689	−2.18, .798
Personality Disorder	−.543	−4.44, 3.36	−.493	−1.51, .526	−1.58	−6.75, 3.58	.614	−.803, 2.03
Self-Esteem	.166	−.174, .506	.004	−.070, .078	−.437*	−.809, −.065	−.037	−.141, .067
EDE Global	−.054	−.149, .041	−.018	−.042, .005	.080	−.028, .188	.024	−.006, .054
Negative-Affect	4.11	−1.07, 9.28	.264	−.750, 1.28	−7.13**	−12.3, −1.96	−1.32	−2.79, .144
Overvaluation	1.05	−2.32, 4.42	.200	−.733, 1.13	−6.88**	−11.4, −2.36	−1.08	−2.33, .312

Note. BED = binge eating disorder; EDE = Eating Disorder Examination; BDI = Beck Depression Inventory

^*p < .05,

^**p < .01,

^***p < .001

Table 5.

Patient Characteristics and Pre-treatment Clinical Characteristics Predicting and Moderating Changes in Body Mass Index.

Predictor	Body Mass Index
	Predicts				Moderates
	Amount		Rate		Amount		Rate
	Est.	95% CI	Est.	95% CI	Est.	95% CI	Est.	95% CI
Age	.020	−.025, .065	−.007	−.023, .010	−.051	−.113, .011	−.006	−.029, .017
Female	−1.09*	−1.99, −.191	−.178	−.512, .156	.633	−.616, 1.88	.056	−.409, .521
Education > H.S.	−.417	−1.41, .575	−.670***	−1.02, −.321	.389	−1.33, 2.11	1.01***	.402, 1.61
Age of BED onset	−.007	−.040, .027	.006	−.006, .018	−.014	−.063, .035	−.010	−.027, .008
Binge-eating frequency	−.862	−2.20, .472	−.015	−.580, .550	2.28	−.006, 4.57	.390	−.562, 1.34
Psychiatric Disorder	.537	−.359, 1.43	.239	−.089, .567	−.551	−1.75, .648	−.195	−.637, .247
Personality Disorder	.704	−.138, 1,55	−.033	−.340, .274	−.659	−1.82, .505	.206	−.224. .636
Self-Esteem	.066*	.006, .125	.012	−.010, .034	−.064	−.148, .021	−.015	−.047, .016
EDE Global	−.007	−.026, .012	.005	−.002, .011	−.001	−.025, .023	−.005	−.014, .004
Depression (BDI)	−.054**	−.089, −.019	.001	−.012, .014	.037	−.016, .091	−.003	−.024, .017
Negative-Affect	1.29**	.478, 2.09	.102	−.206, .410	−.798	−1.98, .381	−.020	−.472, .433
Overvaluation	.289	−.479, 1.06	−.163	−.444, .117	−.363	−1.45, .719	.065	−.331, .461

Note. BED = binge eating disorder; EDE = Eating Disorder Examination; BDI = Beck Depression Inventory

^*p < .05,

^**p < .01,

^***p < .001

What Variables Predict or Moderate Binge-eating Remission?

Mixed models analyses revealed three significant predictors of binge-eating remission: patients with less-than-college education (Odds Ratio = 3.71, 95% C.I. = 1.16 – 11.82), older age at BED onset (Odds Ratio = 1.05, 95% C.I. = 1.01 – 1.09), and without overvaluation of shape/weight (Odds Ratio = 1.90, 95% C.I. = 1.13 – 3.19) were more likely to remit. No significant moderators of binge-eating remission were found.

Given that negative-affect subtype was associated with overvaluation (phi = .325, p < .001) and may signal a more disturbed variant of BED, we tested whether controlling for negative-affect would attenuate the findings for overvaluation. Controlling for negative-affect in the full model did not alter the finding that overvaluation significantly predicted binge-eating remission (wald chi-square (df=1, N = 108) = 7.632, p = .006); in contrast, negative-affect remained non-significant (wald chi-square (df=1, N = 108) = 1.248, p = .26).

Figure 1 shows the proportion of participants categorized with versus without overvaluation of shape/weight who achieved remission overall and separately for medication-only and CBT treatments. Overall, participants with overvaluation of shape/weight were significantly less likely to remit than those without overvaluation (N=18/62 (29%) vs N=26/46 (57%); chi-square (df=1, N=108) = 8.266, p = 0.004, phi = .27). Among participants receiving medication-only, those with overvaluation of shape/weight were significantly less likely to remit than those without overvaluation (N=3/30 (10%) vs N=10/24 (42%); chi-square (df=1, N=54) = 7.315, p = 0.007, phi = .37). Among participants receiving CBT, those with overvaluation of shape/weight had a non-significant trend to remit less than those without overvaluation (N=15/32 (47%) vs N=16/22 (73%); chi-square (df=1, N=54) = 3.563, p = 0.059, phi = .26).

Figure 1. Remission rates for patients by overvaluation status.

Percentage of participants categorized at baseline with overvaluation of shape/weight versus without overvaluation who achieved remission from binge-eating at post-treatment shown separately for overall (N=108), for medication-only (N=54), and for cognitive-behavioral therapy (CBT; N=54) study groups.

Do Demographic Variables Predict or Moderate Dimensional Outcomes?

Several demographic variables were found to be significant predictors. Younger age and older age at BED onset predicted significantly greater reduction and/or faster reduction in binge-eating frequency (Table 2). Older age at BED onset also predicted significantly faster reduction in EDEQ global scores (Table 3). Older age predicted significantly faster reduction in BDI scores (Table 4). Male gender predicted significantly greater BMI loss and less-than-college education predicted significantly faster BMI loss.

Several demographic variables moderated treatment outcomes. Younger participants had significantly greater reductions in binge-eating frequency if receiving medication-only treatments, while female participants receiving medication-only had the least reductions in binge-eating frequency. Participants with an older age at onset of BED had significantly faster reductions in binge-eating if receiving CBT (Table 2). Participants with older age at onset of BED had significantly faster rate of reduction in EDE-Q global scores if receiving CBT (Table 3). Older participants had significantly greater BDI reductions if receiving CBT (Table 4).

Does Co-morbidity Predict or Moderate Dimensional Outcomes?

Neither axis I psychiatric disorder nor axis II personality-disorder co-morbidity significantly predicted or moderated any of the treatment outcomes (Tables 1–5).

Do Pretreatment Clinical Characteristics Predict or Moderate Dimensional Outcomes?

Higher baseline binge-eating frequency predicted significantly greater reductions in EDE-Q global scores (Table 3). Lower self-esteem, higher depression (BDI) scores, and participants sub-typed as negative-affect predicted significantly greater BMI loss (Table 5).

Several pretreatment clinical variables also moderated treatment outcomes. Participants with higher baseline binge-eating frequency had significantly greater reductions in EDE-Q global scores if receiving CBT (Table 3). Participants with lower self-esteem had significantly greater reductions in BDI scores if receiving CBT (Table 4). Patients categorized with negative-affect subtype and patients with overvaluation of shape/weight had significantly greater reductions in EDE-Q global scores (Table 3) and in BDI scores (Table 4) if receiving CBT.

We performed a further series of analyses using the same models in which we tested the effects of overvaluation controlling for negative-affect and vice versa (tested the effects of negative-affect controlling for overvaluation) for the significant findings observed for the EDE-Q global and BDI outcomes. Overvaluation remained a significant moderator of EDE-Q global outcome scores (F (df=1, N = 100.34) = 5.565, p = .02) when controlling for negative-affect; in contrast, negative-affect was no longer significant when controlling for overvaluation (F (df=1, N = 101.14) = 3.482, p = .07). Overvaluation remained a significant moderator of BDI outcome scores (F (df=1, N = 98.87) = 9.144, p = .003) when controlling for negative-affect and negative-affect remained a significant moderator of BDI outcome scores when controlling for overvaluation (F (df=1, N = 100.21) = 6.58, p = .02).

Discussion

This study examined predictors and moderators of response to CBT and antidepressant medication for BED using data from a published RCT (Grilo et al., 2005). Mixed-effects models analyses were applied to baseline variables and repeated-measures (monthly) through post-treatment constructed to evaluate simultaneously whether each variable predicted or moderated either the overall amount of change or the rate of change in the outcome variable. This statistically rigorous approach, which used all available data from repeated monthly assessments, extends previous studies relying on fewer outcome assessment points. We selected potential predictor/moderator variables based on clinical or empirical grounds from the literature and considered five different treatment outcomes following rigorous conventions (e.g., Grilo et al., 2011; Wilson et al., 2010) - i.e., binge-eating remission considered as a primary outcome and improvements in dimensional measures of binge-eating, eating-disorder psychopathology, depression, and BMI considered as secondary outcomes.

Overall, our findings suggest that overvaluation of shape/weight was the most salient predictor and moderator of treatment outcomes. Overvaluation of shape/weight significantly predicted the primary treatment outcome of binge-eating remission. Overall, 29% of participants categorized as having overvaluation of shape/weight (determined by scores of 4 or greater on relevant EDE items, per previous studies (Hrabosky et al., 2007; Grilo et al., 2008; Grilo et al., 2010; Masheb & Grilo, 2008b)) achieved remission from binge-eating whereas 57% of those who did not overvalue their shape/weight achieved remission. The presence of overvaluation of shape/weight was associated with significantly lower remission rates if receiving medication-only treatment (10% vs. 42% for without overvaluation) and with lower remission rates at trend level (p=0.059) if receiving CBT (47% vs. 73%). Although overvaluation of shape/weight was not a statistically significant moderator of binge-eating remission, it was found to significantly moderate dimensional treatment outcomes; specifically, patients with overvaluation of shape/weight had significantly greater reductions in eating-disorder psychopathology (EDE-Q global) and depression (BDI) levels if receiving CBT. The significant predictor and moderator findings for overvaluation of shape/weight findings existed even after controlling for negative-affect. These findings suggest that, among patients with BED, overvaluation accounts prospectively for unique and meaningful variation in binge-eating, eating-disorder psychopathology, and depression treatment outcomes.

Our findings here extend a previous report by Masheb and Grilo (2008b) that overvaluation has negative prognostic significance for guided self-help CBT and BWL interventions. They are also consistent with Wilson et al.’s (2010) finding that eating-disorder psychopathology moderated treatment outcome, and Sysko et al.’s (2010) finding, based on latent transition analysis, that the class characterized by the most severe eating-disorder psychopathology (i.e., greater shape/weight concerns and more frequent binge-eating) responded more to IPT than to either CBTgsh or BWL. The findings suggest that clinicians assess for overvaluation and, if present, be aware that it may signal a more disturbed variant of BED that may require greater clinical attention. In particular, the presence of overvaluation may signal the need or indication for intensive specialist treatment such as CBT (per our findings here) or IPT (per Sysko et al., 2010) rather than medication-only.

The findings regarding overvaluation of shape/weight also have potential implications for the assessment and classification of BED (Masheb & Grilo, 2000). Overvaluation of shape/weight (i.e., the undue influence of weight or shape on self-evaluation” (American Psychiatric Association, 1994, p. 545) is a related - but distinct - clinical construct from the general concept of body dissatisfaction, as demonstrated by factor-analytic (Grilo, Crosby et al., 2010), longitudinal (Masheb & Grilo, 2003), and latent genetic and environmental risk factor analyses in twin studies (Wade, Zhu, & Martin, 2011). Although many persons may be dissatisfied with their appearance, many fewer define their self-worth primarily based on their shape/weight (Hrabosky et al., 2007; Masheb & Grilo, 2003), making it critical for clinicians to keep this distinct in mind. In terms of classification, the nosological system (DSM-IV and the proposed DSM-5) includes the presence of overvaluation as a required criterion for bulimia nervosa but does not include this, or any other cognitive criterion, for the research diagnosis of BED. The current DSM-5 proposal will include BED as a formal diagnosis but also does not include any cognitive criterion. A series of studies with clinical and community samples has suggested that overvaluation of shape/weight be considered as a diagnostic specifier (i.e., to specify or signal a meaningful clinical subtype, but not be a required criterion as it would exclude many persons with a clinically meaningful problem who differ significantly from overweight comparison groups (Hrabosky et al., 2007; Grilo et al., 2008; Grilo et al., 2010). The present findings that overvaluation significantly predicted binge-eating remission and was a significant moderator of other important dimensional treatment outcomes, even after controlling for negative-affect, represents further empirical support for its consideration in DSM-5 as a diagnostic specifier.

In addition, we found several other statistically significant predictors of treatment outcomes. In terms of binge-eating remission, lower educational attainment and older age at BED onset predicted higher remission rates. In terms of predicting any of the four secondary dimensional treatment outcomes, several other significant findings emerged. Younger age at treatment presentation and older age at BED onset predicted improvements in binge-eating frequency, older age at BED onset predicted improvements in eating-disorder psychopathology, older age at treatment presentation predicted improvements in depression levels, and male gender and lower educational attainment predicted improvements in BMI. Higher baseline binge-eating frequency predicted greater reduction in eating-disorder psychopathology, and patients with lower self-esteem, higher BDI depression scores, and those sub-typed as having negative-affect type BED lost significantly more weight. These findings - except for the negative-affect group losing more weight which requires further examination in future research - are broadly consistent with some findings that greater baseline severity variables predict some dimensional outcomes (Masheb & Grilo, 2008a, 2008b; Wilson et al., 2010). Noteworthy is that neither psychiatric disorder nor personality disorder co-morbidity predicted (or moderated) any treatment outcomes, a finding that is partly at odds with isolated findings from two studies (Masheb & Grilo, 2008a; Robinson & Safer, in press) but broadly consistent with other studies (Grilo et al., 2003; Wilfley et al., 2000).

In terms of additional significant moderators, we found a few statistically significant findings for demographic variables. Only one demographic variable signaled a statistical advantage for receiving medication-only: younger participants had greater reductions in binge-eating frequency. In contrast, several demographic variables signaled a statistical advantage for receiving CBT. Female participants receiving medication-only had the least reduction in binge-eating frequency and participants with older age at BED onset had significantly greater reductions in eating-disorder psychopathology and older participants had significantly greater reduction in depression if receiving CBT. In addition, several baseline clinical variables reflecting greater severity signaled a statistical advantage for receiving CBT. Specifically, participants with lower self-esteem, those categorized with negative-affect, and those with overvaluation of shape/weight had significantly better improvements in eating-disorder psychopathology and depression levels if receiving CBT (rather than medication-only). These specific significant moderator findings are consistent with those recently by Wilson et al (2010) and Sysko et al (2010) indicating an advantage for specialist IPT over less-intensive CBTgsh or BWL in patients. Importantly, rather than relying on the weight/shape concern scales (a broader admixture of items as demonstrated in various studies (Grilo, Crosby et al., 2010; Wade et al., 2011), our findings here highlight the specific prognostic value of overvaluation of shape/weight.

The primary analyses from this RCT indicated that CBT was significantly more effective than anti-depressant (i.e., fluoxetine) medication across multiple outcome measures including a robust two-fold advantage for producing binge-eating remission (Grilo et al., 2005). The NICE (2004) guidelines note that antidepressant medications are an acceptable alternative treatment for BED (a recommendation graded with less support than the grade “A” assigned to CBT as the best-established treatment). However, other controlled studies that have tested fluoxetine directly with CBT alone and in combination (Ricca et al., 2001), or compared to CBT as adjunct methods to enhance behavioral weight loss treatments (Devlin et al., 2005) have reported minimal (Devlin et al. 2005) to no support (Ricca et al., 2001) for antidepressant treatment for BED. Such findings from comparative RCTs are consistent with the modest effect size (relative-risk of 0.19 for medication versus placebo) reported in a meta-analysis of anti-depressant treatments for BED (Reas & Grilo, 2008). Moreover, to date, no placebo-controlled trials have reported any follow-up data for antidepressant-only or any medication-only for BED (Reas & Grilo, 2008). Ricca et al (2001) reported that CBT was significantly more effective than either fluoxetine or fluvoxamine at both post-treatment and at 12-month follow-up, in an open-label comparative trial. Nonetheless, treatment guidelines such as NICE (2004) do suggest that antidepressant medications represent a possible treatment option and the clinical realty is that they are frequently and widely prescribed for BED and binge-eating problems, perhaps reflecting, in part, their ease of delivery. For these reasons, understanding the predictors and moderators of treatment outcomes for these two distinct interventions is critical as they might inform treatment prescriptions across clinical settings. Perhaps our most salient or overall parsimonious finding is that the presence of overvaluation of shape/weight – which signals a more disturbed subgroup of BED – appears to represent a strong negative prognostic indicator for medication-only and suggests consideration of specialist (CBT or IPT) treatment.

We note several potential limitations to consider as further context. Findings in the present study pertain to overweight individuals with BED who participated in a RCT testing CBT and medication treatments at a university medical center, and may not generalize to different clinical settings, treatment methods, or to persons not willing to take medications or receive CBT. Our analyses pertain to acute post-treatment outcomes and may not generalize to longer-term outcomes. Since most (89%) participants were white, meaningful analyses of ethnic/racial differences were not possible, and the findings may not generalize to more diverse patient groups or to member of different ethnic/racial groups. Larger studies with more diverse patients (see Franko et al., in press) and clinical settings and with longer-term outcomes are needed to replicate and extend our findings.