ON SUBGROUP IDENTIFICATION VIA CHANGE POINT DETECTION AND MODEL AVERAGE

Abstract

In the evolving landscape of medical studies, understanding the nuanced interplay of genetic, lifestyle, and environmental factors in an individual's response to treatment has spurred the rise of personalized medicine. This shift from the traditional "one size fits all" paradigm emphasizes tailoring healthcare interventions to specific subpopulations, a pursuit known as subgroup identification. This thesis introduces a novel subgroup identification approach, Change Point Model Average (CPointMA), which copes with segment regression models with multiple threshold variables and structural breaks. The method employs a two-stage change point detection and a frequentist model averaging process to identify intricate subgroups in heterogeneous study populations. Extending our methodology, we further propose the Change Plane Model Average (CPlaneMA) method, addressing limitations of CPointMA by characterizing subgroups using linear combinations of variables and multiple cutoff points. Comprehensive simulations and real data examples showcase the efficacy of our proposed methods, providing valuable insights for personalized healthcare and beyond.