EHRSummarizer: A Privacy-Aware, FHIR-Native Architecture for Structured Clinical Summarization of Electronic Health Records

Details

The proposed EHRSummarizer addresses the challenge of fragmented electronic health record (EHR) interfaces by providing a coherent view of patient data. The system is designed to be privacy-aware and FHIR-native, retrieving a targeted set of high-yield FHIR R4 resources. These resources are then normalized into a consistent clinical context package, which serves as the basis for generating structured summaries intended to support structured chart review. Key features include configurability for data minimization and stateless processing, allowing flexible deployment, including local inference within an organization’s trust boundary. The architecture mitigates risk by constraining the summarization stage solely to evidence present in the retrieved context package, avoiding diagnostic or treatment recommendations. The system is designed to indicate missing or unavailable domains where feasible. While prototype demonstrations were conducted using synthetic and test FHIR environments, this manuscript does not report clinical outcomes or controlled workflow studies. Future evaluation must focus on faithfulness, omission risk, temporal correctness, usability, and operational monitoring.