Group-based treatments for problem alcohol use are designed to reflect the pragmatic realities of treatment settings. One such consideration is that participants are often enrolled into therapy groups on an open, or rolling, basis, as space becomes available in the therapy group. A barrier to testing such interventions has been the interrelatedness of group members' experiences. The ongoing entry and departure of different group participants at different times in open-enrollment groups (OEGs) induces complex correlations among group members' outcomes. Failure to account for this correlation could lead to incorrect statistical tests of treatment effects, undermining our ability to draw conclusions about the effectiveness of treatment. In our previous project, we addressed this correlation by innovatively conceptualizing OEG sessions as spatially related and drawing upon a wealth of existing statistical methodology for spatial data analysis using conditional autoregression (CAR). We demonstrated the versatility of CAR for analyzing data collected during the active treatment phase or post-treatment. Despite these advances, alcohol treatment researchers continue to need new statistical methods to model correlation among OEG participant outcomes. This renewal application responds to NIAAA's recognition of the need to further develop, refine, validate, and creatively implement novel statistical methods for the treatment of alcohol use disorders (PA-13-160). Further guidance is needed regarding how to examine multimodal outcomes that are ubiquitous in group-based alcohol treatment research. Relaxing aspects of the standard CAR approach may result in better-fitting models. Our current methods allow one to estimate a causal treatment effect of a primary outcome for participants who are experimentally assigned to an OEG-based intervention versus a comparison when participants do not interact across study arms. However, a growing body of alcohol treatment research focuses on testing the effects of non- randomized factors on outcomes. Causal inference is challenging when a factor of interest varies among individuals within a therapy group, resulting in interference among individuals.
Specific Aims : (1) Develop innovative approaches to analyze multimodal or semicontinuous outcomes from group therapy studies; (2) explore alternatives to conditional autoregression for flexibly modeling session-to-session correlation; and (3) employ a state-of-the-art causal inference framework that appropriately accounts for the interference among participants in group-based alcohol treatment. Fulfillment of these Specific Aims will provide the alcohol treatment research community innovative statistical methods that synergistically address modeling correlation among OEG participant outcomes while advancing the ability of the field to test the effectiveness of treatment factors of interest.
This proposed research project is relevant to public health because its ultimate goal is to improve group-based treatments of problem alcohol use. Motivated by the ubiquity of group therapy for treating problem alcohol use, this project develops a statistical approach that can be used to test the effectiveness of group-based alcohol treatment in realistic settings.
