Identifying cohorts at elevated risk of cancers using generative modeling of patient health states

While large language models are powerful generators of new text, forecasting disease progression from longitudinal health histories remains a challenging problem. We introduce GenEHR, an autoregressive generative model trained on electronic health records (EHRs) from millions of patients in diverse health systems. The model learns from real-world multimodal patient histories, including diagnoses, medications, lab values, and procedures while explicitly representing irregular intervals between visits to forecast future clinical events. We combine the general-purpose patient representation learned during foundational training with parameter-efficient supervised adaptation for the task of risk stratification for 17 cancer types. In five large EHR cohorts, supervised adaptation substantially improved prediction performance of a first cancer diagnosis within a five year horizon window. Our retrospective results support the prospective evaluation of GenEHR-CancerRisk as a clinical decision-support tool for prioritizing high-risk patients for screening and prevention of aggressive cancer types, such as pancreatic and ovarian cancer.