Abstract
It is increasingly recommended that hypothesis-generating studies be conducted after initial RCTs in order to identify moderators of differential treatment efficacy on individual outcomes. Such analyses are important because they help clarify the best inclusion and exclusion criteria or choice of stratification for maximizing power in subsequent RCTs, reduce the chances of discarding interventions that may appear to lack efficacy when only average treatment effects are taken into consideration, and facilitate the matching of individual clients to treatment alternatives. We identified predictors and moderators of treatment-related change in psychological distress among custodial grandmothers (n = 343) assigned within a prior RCT to behavior parent training (BPT), cognitive behavior therapy (CBT), or information only control (IOC) conditions. Latent change scores in psychological distress were estimated for each grandmother across pre-test to post-test and pre-test to six months, as indicated by self-reported and clinical ratings of depression and anxiety symptoms. These estimates served as outcomes in classification and regression tree analyses conducted separately within the CBT and BPT conditions to identify predictors of treatment efficacy. Matched groups based upon identified predictors were then formed across all RCT conditions, and Predictor × RCT Condition interactions were computed to test for moderation of differential treatment efficacy. Grandmother age was the only predictor and moderator of BPT efficacy at both measurement points, whereas multifaceted predictors and moderators emerged for CBT which varied by time since treatment.