A Multi-Agent Architecture for Privacy-Preserving Natural Language Interaction with FHIR-Based Electronic Health Records
Carmen De Maio, Giuseppe Fenza, Domenico Furno, Teodoro Grauso, Vincenzo Loia · 2024
Integrating Electronic Health Records (EHRs) into clinical workflows is essential for enhancing healthcare delivery but poses significant challenges, such as improving human-machine interaction through natural language queries. This paper addresses these challenges by leveraging Large Language Models (LLMs) within a multi-agent architecture. The aim is to develop a privacy-preserving method enabling clinicians to interact with FHIR-based EHRs using natural language. The novelty of the proposed technique lies in using publicly available LLMs to construct URIs for retrieving FHIR resources and a local LLM to interpret these resources, thus safeguarding patient privacy by preventing direct exposure of sensitive data. Evaluated with the SyntheticMass dataset, the multi-agent system demonstrated superior accuracy and detail over a single-model approach while maintaining competitive response times.