Graduate Student Seminar: Subgroup Identification Based on Mixed Model for Repeated Measures for Alzheimer's Disease Trial

In precision medicine, the identification of subgroups is pivotal for designing personalized treatments. While current methods for subgroup identification primarily concentrate on either the total treatment effect in cross-sectional studies or the mean effect in longitudinal studies, the FDA typically evaluates treatment effects at the end of longitudinal studies through the Mixed Model for Repeated Measures (MMRM). Therefore, we introduce the Interaction Tree with Mixed Model for Repeated Measures (IT MMRM) as a novel approach for identifying subgroups. The mixed effect model with a nonlinear time trend allows us to model the dependence of longitudinal measures flexibly and single out a time dependent treatment interaction to build up the interaction tree. Our IT-MMRM demonstrates superior performance over existing subgroup identification techniques in simulations, especially when the time dependent treatment effect is the focal point. We also explore different tuning parameter options and use bootstrap methods to prune the trees, thereby mitigating the risk of overoptimism. The IT-MMRM model is applied to an Alzheimer's disease study, aiming to uncover subgroups with varying long-term treatment responses.