LAUSR

The effects of treatment often vary over subpopulations characterized by baseline patient features. Detection of such treatment-subgroup interaction is of central importance to precision medicine and personalized care. In this paper, we propose an analytical framework for treatment-subgroup interactions detection and treatment effectiveness heterogeneity evaluation in a real-world data setting. Model-based recursive partitioning analysis (MOB) is used for subgroup identification, filter-based confounder selection and multivariate logistic regression are used for confounding reduction and treatment effectiveness assessment. We illustrate this approach by a real-world case study that analyzes the effects of 15 drugs among patients with myocardial infarction (MI) using China Acute Myocardial Infarction (CAMI) registry data. The results show that our approach effectively identifies meaningful patient subgroups involved in treatment-subgroup interactions and thus can potentially aid decision making in personalized medicine.