Unlocking Patient History: AI-Driven Language-Based Interface for Accessing HL7-CDA Episode Data

Abstract

[Work in progress] Healthcare professionals face significant challenges when navigating Electronic Health Records (EHRs), particularly in high-complexity clinical scenarios involving structured formats such as the HL7 Clinical Document Architecture (CDA). This systematic review examines the state of the art in patient history information retrieval from EHR documents, ranging from rule-based extraction to Retrieval-Augmented Generation (RAG) and agentic Large Language Model (LLM) architectures. Through an analysis of 25 selected papers, a critical gap is identified: while RAG-based systems offer superior context-awareness compared to traditional methods, agentic architectures remain underexplored for clinical retrieval. Furthermore, most existing research relies on synthetic data and Fast Healthcare Interoperability Resources (FHIR) standards, leaving a significant gap in the handling of HL7-CDA documents with real clinical data that evolve over time. This review provides a foundation for the development of intelligent interfaces that transform complex EHR navigation into intuitive, conversational patient history access. We conclude that future research should prioritize agentic architectures validated on real clinical data to bridge identified gaps.