Project Summary The goal of this application, submitted in response to PAR-19-368, “Accelerating the Pace of Drug Abuse Research Using Existing Data” is to leverage data from the NIDA Clinical Trials Network to enhance our understanding of treatment effect heterogeneity in psychosocial treatments for substance use disorders. Treatment effect heterogeneity is particularly a concern in research of substance use disorder treatments, in part due to heterogeneous sub-phenotypes of patients in symptom profile, disease course, and recovery trajectory. Nevertheless, analysis of treatment effect heterogeneity in substance use disorder research has been often conducted in a suboptimal manner using subgroup analysis (i.e., estimating impacts separately within groups defined by a single covariate), which could result in finding spurious differences in treatment effects by subgroup due to the performance of multiple statistical tests and random variability across patients. Another challenge of single covariate-based subgroup analysis is that most covariates have small moderating effects and their individual contribution to treatment effect heterogeneity is not meaningfully informative to treatment decisions. Our objective is to apply a novel statistical method, causal forest approach, to systematically examine treatment effect heterogeneity of psychosocial treatments for substance use disorders. This study uses data from 12 randomized controlled trials in the NIDA Clinical Trials Network which examined effectiveness of nine distinct psychosocial treatments against treatment-as-usual condition (Motivational Incentives, Motivational Enhancement Therapy, Screening Motivational Assessment, Therapeutic Education System, Brief Strategic Family Therapy, Twelve-Step Facilitation, Motivational Interviewing, Seeking Safety, and Exercise Program). For each type of psychosocial treatment, we propose to implement the causal forest approach to estimate the expected effect of a treatment for each individual while taking into account multiple covariates simultaneously. The estimated treatment effects will be used to test the presence and degree of treatment effect heterogeneity in each type of psychosocial treatment. Using the variable importance measure obtained from the causal forest, we also plan to identify the most important covariates contributing to treatment effect heterogeneity in each type of psychosocial treatment. These analyses will be repeated for multiple outcomes (e.g., abstinence, reduction in frequency of target substance use) to examine whether and how the degree of treatment effect heterogeneity as well as common effect moderators differ across outcomes. Overall, these analyses will grow the evidence base that can be used by treatment providers to guide treatment decisions for individual patients with substance use disorders.