Self-MedRAG Improves Medical QA Accuracy and Reliability

Researchers on Jan. 8, 2026 propose Self-MedRAG, a hybrid retrieval and self-reflective framework for medical question answering that combines BM25 and Contriever via Reciprocal Rank Fusion and adds NLI/LLM-based verification. Evaluated on MedQA and PubMedQA, it raised accuracy from 80.00% to 83.33% and from 69.10% to 79.82%, demonstrating reduced unsupported claims and improved clinical answer reliability.