Configurable Clinical Information Extraction with Agentic RAG: What Works, What Breaks, and Why

The article presents the Agentic Clinical Information Extraction (ACIE) system, an on-premise retrieval-augmented generation (RAG) pipeline designed to handle complex patient contexts by integrating complete document-level metadata for enhanced extraction accuracy. It addresses shortcomings in standard RAG approaches, particularly in temporal reasoning and cross-document dependencies. In a study involving 7,326 clinician evaluations, ACIE achieved a 96.5% acceptance rate for extracted information, demonstrating its potential for reliable clinical decision support and verification in medical contexts.