Improving Large Language Model Applications in the Medical and Nursing Domains With Retrieval-Augmented Generation:...

Area of Science:

• Artificial Intelligence in Healthcare
• Natural Language Processing for Medical Applications
• Clinical Decision Support Systems

Background:

• Retrieval-augmented generation (RAG) is increasingly utilized to enhance large language models in medical and nursing fields.
• A comprehensive understanding of RAG's architecture and applications in medical and nursing reasoning is currently limited.
• This review addresses the need for clarity on RAG's role in healthcare AI.

Purpose of the Study:

• To summarize the current state of RAG in medical and nursing domains.
• To identify existing limitations and challenges of RAG applications in healthcare.
• To outline future development directions for RAG in medical and nursing reasoning.

Main Methods:

• Systematic literature search across PubMed, Web of Science, IEEE Xplore, and arXiv databases (November 2022 - May 2025).
• Queries combined terms for RAG, medical, and nursing domains.
• Review conducted following PRISMA-ScR guidelines.

Main Results:

• 67 studies met inclusion criteria; 94% focused on the medical domain, 6% on nursing.
• Identified 5 RAG framework types: text-based (54%), knowledge graph-enhanced (25%), agentic (9%), multimodal (3%), and plug-and-play (9%).
• Only 26 studies included explicit reasoning support, few aligned with clinical workflows, and only 12 addressed ethical considerations.

Conclusions:

• Key RAG development shifts include enhanced intent recognition, logic-driven retrieval, active knowledge retrieval, and coherent context construction.
• Most current RAG systems in healthcare lack reasoning methods or rely on data-driven associations without causal modeling.
• Integration of causal mechanisms is crucial for effective, domain-relevant reasoning in healthcare RAG systems.