Machine learning enabled subgroup analysis with real-world data to inform better clinical trial design

Restrictive eligibility criteria for clinical trials may limit the generalizability of treatment effectiveness and safety to real-world patients. In this paper, we propose a machine learning approach to derive patient subgroups from real-world data (RWD), such that the patients within the same subgroup share similar clinical characteristics and safety outcomes. The effectiveness of our approach was validated on two existing clinical trials with the electronic health records (EHRs) from a large clinical research network. One is the donepezil trial for Alzheimer’s disease (AD), and the other is the Bevacizumab trial on colon cancer (CRC). The results show that our proposed algorithm can identify patient subgroups with coherent clinical manifestations and similar risk levels of encountering severe adverse events (SAEs). We further exemplify that potential rules for describing the patient subgroups with less SAEs can be derived to inform the design of clinical trial eligibility criteria.