Investigating predictors of short-term self-help treatment outcomes is particularly important in light of recent guidelines suggesting these represent potential first-line treatment for BED (see
Wilson et al., 2007). Our findings suggest that negative affect, as measured by the BDI, was the most salient predictor of guided self-help treatment outcome for BED. Negative affect predicted attrition and post-treatment levels of negative affect and eating disorder psychopathology. Our finding that negative affect predicted attrition is at odds with findings from a smaller study of self-help for BED (
Loeb et al., 2000), while our finding that negative affect did not predict post-treatment binge eating corroborated a previous report (
Peterson et al., 2000).
In contrast to the robust predictive significance of negative affect, the presence of psychiatric comorbidity, and more specifically MDD, did not predict or moderate attrition or outcomes. These findings are consistent with previous research in which subtyping BED patients using the BDI (a broad measure of negative affect) had much greater concurrent and predictive utility than subtyping by MDD co-morbidity (
Stice et al., 2001).
Wilfley and colleagues’ (2000) previously reported that psychiatric comorbidity in BED patients was not predictive of group CBT or group interpersonal psychotherapy outcomes. On the other hand, we did find that personality disorders, particularly Cluster C disorders, were predictive of post-treatment levels of eating disorder psychopathology and negative affect.
Key dimensional outcomes, including binge frequency, eating psychopathology, and negative affect were predicted by their respective pretreatment levels. It was not the case, however, that binge remission (a categorical outcome) was predicted by pretreatment binge frequency. This latter finding replicates two studies of CBT, delivered in group (
Peterson et al., 2000) and self-help (
Loeb et al., 2000) formats.
We were unable to identify any significant moderators of treatment. Moderators refer to whom and under what conditions treatments have different outcomes (
Kraemer et al., 2002). It appears that the guided self-help versions of these two structured and behaviorally oriented treatments performed comparably in patients with BED. Thus, our findings provide no guidance for matching patients to either CBTgsh or BWLgsh treatments. The only other study to date that investigated moderators of BED treatments also failed to find any significant moderators (
Hilbert et al., 2007).
The identification of predictors and moderators of treatment in the present study was limited to some extent by the sample size. Power analysis for continuous outcome measures (assuming a two-tailed alpha of .05) revealed that, after controlling for treatment and baseline measurement, the current sample size provided adequate power (.80) to identify a predictor accounting for an additional 7–9% depending upon the specific outcome and predictor variables (
Hintze, 2001). Power analysis for tests of moderation revealed that depending upon the strength of the relationship between predictor and outcome, the current size provided adequate power (.80) to identify a moderator accounting for an additional 6–9% of the criterion variance over-and-above the main effects model (
Hintze, 2001). Power analyses for logistic regression analysis indicated that after controlling for treatment, the current sample size provided adequate power (.80) to identify a predictor with an adjusted odds ratio of 2.08 and a moderator with an adjusted odds ratio ranging from 2.08–2.11 depending upon the specific variables (
Hintze, 2001).
Although our study was powered to detect effect sizes as described above, the ability to detect smaller effects would require larger sample sizes. Large sample sizes are often not feasible in conducting behavioral studies although there are ways to partly remedy the problem (
Jaccard et al., 2006;
Maxwell 2000,
2004) which were considered in the present study. First, with the exception of negative affect and self-esteem (
r=−.69), the selected variables were not highly correlated (
r’s ranged from .01–.35). Second, significance tests were supplemented with effect size measures and confidence intervals. Finally, multiple regression, using product terms and the full continuum of the measures, was used to test moderating effects, a more powerful approach than the use of categorical groupings.
We note other potential limitations to consider. Findings in the present study pertain to overweight individuals with BED who participated in a study of guided self-help treatments at a university medical center, and may not generalize to different clinical settings, treatment methods, or longer-term approaches or outcomes. For example, unlike the two Stanford studies (
Agras et al., 1995,
1997) with more intensive forms of CBT, we did not find that either age at presentation for treatment (
Agras et al., 1997) or earlier onset of BED (
Agras et al., 1995) was related to outcome. It could be that younger patients fare worse than older patients in group CBT, but not guided self-help interventions. The majority (73%) of participants were Caucasian; although the racial and ethnic diversity was representative of New Haven County and the state of Connecticut. Lastly, failure to observe moderating effects may have been limited by a narrow range of scores on dependent measures that are typical of clinical samples.
Overall, negative affect, but not MDD comorbidity, was a robust negative prognostic indicator for most dimensional treatment outcomes for BED as well as for attrition. Key dimensional treatment outcomes were predominately predicted, but not moderated, by their respective pretreatment levels and by the presence of personality disorders, particularly Cluster C disorders. Despite the prognostic significance of these findings for dimensional outcomes, none of the variables tested were predictive of binge remission (i.e., a categorical outcome). The absence of moderating effects for any of the variables tested suggests future research should focus on the differential effects of other, perhaps more intensive, types of treatments for BED. In sum, the present study found poorer prognosis for patients with negative affect and personality disorders suggesting that treatments may be enhanced by attending to the cognitive and personality styles of these patients.
The publisher's final edited version of this article is available at 