MedCoT-RAG: Causal Chain-of-Thought RAG for Medical Question Answering

This project develops MedCoT-RAG, a domain-specific framework that enhances medical question answering by combining causal-aware retrieval with structured chain-of-thought prompting. Unlike standard RAG methods that rely on surface-level similarity, MedCoT-RAG retrieves evidence aligned with diagnostic logic and guides models to generate step-by-step reasoning reflective of clinical workflows. Evaluated across three medical QA benchmarks, the framework improves accuracy, interpretability, and consistency — outperforming both vanilla RAG and advanced domain-adapted baselines. MedCoT-RAG demonstrates how structured causal reasoning can transform LLMs into more trustworthy tools for clinical decision support.

Publications:

Full paper

Project information

  • Category: Future Healthcare, Large Language Models, and Translation and Practice
  • Contact Person: Amir M. Rahmani
  • MedCoT-RAG: Causal Chain-of-Thought RAG for Medical Question Answering

info@futurehealth.uci.edu

© Copyright 2021 UCI Institute for Future Health - All Rights Reserved