Yeon Hee Park Department of Statistics, Sungkyunkwan University

From Enrichment to Personalized Randomization: Data-Driven Strategies for Precision Medicine

Precision medicine seeks to tailor treatments to patient subgroups who are most likely to benefit. My research initially focused on enrichment designs, which identify and restrict enrollment to treatment-sensitive subpopulations in clinical trials. While such designs can increase statistical power, they often involve burdensome screening procedures to identify biomarker-defined subgroups and may lead to the exclusion of a large portion of the patient population, raising concerns about generalizability and feasibility. In response to these limitations, I have shifted toward personalized randomization strategies that adaptively assign treatments to individual patients based on their predicted benefit. In this talk, I will introduce MARGO (Machine learning-assisted Adaptive Randomization for Group sequential trials based on Overlap weights), a novel data-driven framework that combines machine learning with causal inference tools to guide real-time, individualized treatment allocation. MARGO utilizes predictive models to estimate the probability of treatment success for each patient and applies overlap weighting to correct for covariate imbalance introduced by adaptive randomization. This enables valid group sequential testing while preserving flexibility and ethical efficiency. Through simulation studies, we demonstrate that MARGO achieves higher clinical benefit and statistical power compared to conventional designs, while effectively controlling type I error. This work highlights the potential of integrating modern machine learning and causal inference techniques into clinical trial design, offering a more ethical and efficient pathway to personalized medicine.