Foundation Model of Electronic Medical Records for Adaptive Risk Estimation

• DOI: 10.48550/arxiv.2502.06124
• Publication Date: 2025-02-09
• Authors: Paweł Renc; Michal K. Grzeszczyk; Nassim Oufattole; Deirdre Goode; Yugang Jia; Szymon Bieganski; Matthew B. A. McD...; Jarosław Wa̧s; Anthony E. Samir; Jonathan W. Cunni...; David O. Bates; Arkadiusz Sitek
• Keywords: Foundation (evidence); Medical record

We developed the Enhanced Transformer for Health Outcome Simulation (ETHOS), an AI model that tokenizes patient health timelines (PHTs) from EHRs. ETHOS predicts future PHTs using transformer-based architectures. The Adaptive Risk Estimation System (ARES) employs to compute dynamic and personalized risk probabilities clinician-defined critical events. ARES incorporates a explainability module identifies key clinical factors influencing estimates individual patients. was evaluated on MIMIC-IV v2.2 dataset in emergency department (ED) settings, benchmarking its performance against traditional early warning systems machine learning models. processed 299,721 unique patients into 285,622 PHTs, with 60% including hospital admissions. contained over 357 million tokens. outperformed benchmark models predicting admissions, ICU prolonged stays, achieving superior AUC scores. ETHOS-based demonstrated robustness across demographic subgroups strong reliability, confirmed via calibration curves. provides insights patient-specific contributing risk. ARES, powered by ETHOS, advances predictive healthcare providing dynamic, real-time, estimation enhance clinician trust. Its adaptability accuracy position it as transformative tool decision-making, potentially improving outcomes resource allocation inpatient settings. release full code at github.com/ipolharvard/ethos-ares facilitate research.