A Bayesian Approach to Subgroup Identification

This article discusses subgroup identification, the goal of which is to determine the heterogeneity of treatment effects across subpopulations. Searching for differences among subgroups is challenging because it is inherently a multiple testing problem with the complication that test statistics for subgroups are typically highly dependent, making simple multiplicity corrections such as the Bonferroni correction too conservative. In this article, a Bayesian approach to identify subgroup effects is proposed, with a scheme for assigning prior probabilities to possible subgroup effects that accounts for multiplicity and yet allows for (preexperimental) preference to specific subgroups. The analysis utilizes a new Bayesian model selection methodology and, as a by-product, produces individual probabilities of treatment effect that could be of use in personalized medicine. The analysis is illustrated on an example involving subgroup analysis of biomarker effects on treatments.