Analysis of post-treatment outcomes in group therapy studies under open enrollment
*Susan Paddock, RAND Corporation 
Terrance Dean Savitsky, RAND Corporation 

Keywords: Bayesian methods, conditionally autoregressive prior, multilevel model, multiple membership model, substance abuse treatment

Group therapy is a central treatment modality for behavioral health disorders such as alcohol and other drug use and depression. Since group therapy is often delivered under an open admissions policy, where new clients are continuously enrolled into a therapy group as space permits, it is desirable to design group therapy intervention studies accordingly. However, client attendance patterns under open admissions policies result in a complex correlation structure among client outcomes. Despite the ubiquity of open admissions in practice, guidance is needed on how to properly analyze post-treatment outcomes data collected from clients when the goal is to obtain an estimate of the intervention effect following treatment. To account for the fact that each post-treatment client outcome summarizes the effect of a client’s attendance of multiple sessions, we fit multiple membership models to client outcomes data. However, to address the limitation of standard multiple membership models in this context, we fully model the interrelatedness of client depressive symptom scores by assuming a conditionally autoregressive prior for session-level random effects in the multiple membership model. We compare the performance of our approach with the standard multiple membership approaches as well the commonly-employed approach of ignoring the correlation structure and demonstrate improved performance with our approach in the context of a study that delivered group cognitive behavioral therapy for depression to clients in residential substance abuse treatment.