A strategy to identify differential treatment effects based on recursive partitioning methods.
Keywords: treatment effect, moderation, recursive partitioning
Treatment effects for an entire study population can be described as mixtures of the effects among subgroups. Given this fact, for example, at least three groups may be exhaustively explored with respect to the statistical significance and the direction of treatment effects: significantly harmful, significantly beneficial, or non-significant. It is relatively easy to evaluate the direction and degree of the treatment effects across the subgroups that had been determined a priori. However, as dimensions of covariates that describe the differential subgroups increase and as the knowledge of them remains still at an early stage of a clinical investigation, the statistical identification of those groups becomes imperative. For this purpose, we employ Recursive Partitioning techniques that have been immensely developed in various scientific communities during the past decades. In this talk, we introduce a Recursive Partitioning strategy to identify the subgroups whose treatment effects are moderated with respect to their characteristics which are fairly complex yet clinically important. We discuss simulation study results and analyses based on a cardiovascular data set of young adults.