FedEHR-Agents Boosts Automated EHR Modeling with Federated AI

The rise of large language models (LLMs) is enabling autonomous clinical agents to automate complex Electronic Health Record (EHR) modeling workflows. However, individual hospitals face limitations due to institution-specific data and environments, while direct data sharing across hospitals is restricted by patient privacy concerns. Federated Learning (FL) offers a privacy-preserving solution, but existing FL approaches are typically "model-centric," only federating prediction models or their updates. This research proposes FedEHR-Agents, an "experience-centric" federated agentic optimization framework. Each hospital deploys an autonomous clinical EHR agent that refines its local modeling experience through historical memory, task-specific evaluation, and prompt refinement. A central federated server then aggregates this "experience" across hospitals, distilling it into global meta-prompts for further local refinement. Experiments on multi-hospital EHR benchmarks demonstrate that FedEHR-Agents consistently outperforms both local and traditional federated baselines across various clinical prediction tasks, proving robust across different federation scales and LLM backbones. This highlights clinical modeling experience as a valuable collaborative object beyond conventional parameter-centric FL.