Medical Retrieval-Augmentation Generation Framework for Healthcare Prediction.

Electronic Health Records (EHRs) are pivotal for healthcare prediction tasks, offering rich patient data such as symptoms, diagnoses, and treatments. Recent advances in Retrieval-Augmented Generation (RAG) have gained attention due to the ability to retrieve relevant information from medical sources to improve EHR-based predictions. However, existing RAG approaches for medical applications often struggle with flat data representations, which fail to capture the complex inter-dependencies among medical entities, leading to fragmented and verbose responses. In this work, we propose MedGR, a novel framework for healthcare prediction that incorporates graph-based clinical text indexing with a dual-level medical retrieval architecture. By leveraging graph-structured knowledge, we synthesize information from multiple sources into coherent and contextually enriched responses in an efficient manner. The experiment results showed that our medical RAG framework achieved high precision performance on both diagnosis and medical code prediction.