Retrieval-Augmented LLMs improve clinical trial recruitment by localizing evidence in EHRs

Researchers explored retrieval-augmented large language models (LLMs) for identifying suitable patients for clinical trials from electronic health records. The study evaluated various LLMs, including general and medical-adapted versions, and tested strategies to handle long documents, such as default context windows, NER-based summarization, and dynamic evidence retrieval (RAG). The MedGemma model combined with RAG achieved the highest performance, demonstrating LLMs' potential to improve trial recruitment efficiency, particularly for criteria requiring long-term reasoning. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Demonstrates LLMs' potential to streamline clinical trial recruitment by improving evidence localization in patient records.

RANK_REASON Academic paper detailing novel application of LLMs and retrieval augmentation for a specific problem domain.

Read on arXiv cs.CL →

• LLMs
• MedGemma
• RAG
• N2C2
• EHR
• Ziyi Chen

• paper
• product