Retrieval Augmented MedLM

This paper presents a novel approach to leverage large language models (LLMs) for medical question answering (QA) by integrating them with external knowledge sources. We utilize de-identified clinical discharge notes from MIMIC-IV and Apollo Hospitals as our data source. We propose a novel summarization technique that extracts and condenses the core medical information from the discharge notes, eliminating unnecessary verbosity. This results in concise "medical summaries" that effectively inform the LLM while reducing context overload. We evaluate our approach using RAGAS, a novel framework for label-free evaluation of Retrieval-Augmented Generation (RAG) pipelines. Clinician validation further confirms the effectiveness of our approach, highlighting its potential to enhance medical QA systems.