Agentic RAG Improves Clinical Information Extraction Accuracy.

Researchers have developed and deployed an agentic Retrieval-Augmented Generation (RAG) pipeline, named ACIE, at University Medicine Essen. This system is designed to extract complex clinical information from extensive patient records, which often consist of hundreds of diverse documents and thousands of data points. Traditional RAG methods struggle with this type of data due to issues like temporal reasoning, dependencies across multiple documents, and incomplete metadata. The ACIE pipeline specifically addresses these limitations by reasoning over complete patient contexts and ensuring that every extracted answer is grounded in verifiable source passages for clinicians. An evaluation involving nuclear-medicine physicians verifying 7,326 extractions against cited sources showed a high acceptance rate of 96.5%, with specific extraction types ranging from 80% to 99% accuracy. This demonstrates the system's robustness and reliability in a real-world clinical setting.