Online Program

Identifying Target Subgroup with CART for Pharmacogenetic Approach of Reducing Heavy Alcohol Drinking

Bankole Johnson, University of Virginia 
*Lei Liu, Northwestern University 
Baqun Zhang, Northwestern University 

Keywords: substance abuse, personalized medicine

Heavy alcohol drinking can cause serious health and social problems. However, pharmacotherapy for alcohol use disorder (AUD) is limited as there exist only four medications approved by FDA to treat AUD, and their effect sizes are in general small. Developing new and more effective medications to treat alcoholism remains a high priority for researchers (Willenbring 2007). Furthermore, finding the target subgroup for treatments in which the severity of drinking in alcohol-dependent individuals can be substantially decreased is an important scientific and health goal. The common way for such a problem is the regression-based analysis which fits models for the outcomes on covariates and treatment (including interactions between treatment and the covariates), and subgroups are identified based on significance of specific interaction terms. A novel and general framework for the identification of subgroup from classification perspective was proposed in Zhang et al. We applied this method with CART (classification and regression tree) to data from a phase II, 11-week randomized controlled trial of ondansetron in 283 alcohol-dependent individuals (Johnson et al. 2011). While ondansetron has no overall benefit effect on heavy drinking in the whole population, we identified a subgroup of 1/3 of the population for which the benefit of ondansetron is substantial. An analysis with mixed-effects linear regression model confirmed our findings.