LLM-Based Multi-Agent System with Retrieval-Augmented Generation for Medical Care Planning Generation in Sickle Cell Disease

2026-04-12 · Source: Paper Index on ACL Anthology · Field: Science & Research — Health & Medical Research, Mathematics & Computational Sciences · Depth: Expert, quick

Summary

A novel LLM-based Multi-Agent System (MAS) integrated with Retrieval-Augmented Generation (RAG) has been developed to generate medical care plans for Sickle Cell Disease (SCD). This system addresses the critical safety challenges of LLMs in high-risk clinical applications, particularly concerning hallucinations and unsafe outputs for underrepresented conditions like SCD. The MAS architecture decomposes clinical reasoning into specialized agents handling diagnosis, investigation, and treatment planning. Three RAG strategies—LLM-Guided Tree Retrieval, Metadata-Filtered Retrieval, and Semantic Similarity Retrieval—were evaluated against a baseline. Physician assessments and LLM-as-a-Judge evaluations demonstrated high clinical quality, with safety scores above 4 out of 5. Although average performance was comparable between RAG and baseline, the Tree Retrieval strategy significantly reduced clinically unsafe outputs compared to conventional Semantic Retrieval, highlighting RAG's role as a safety control mechanism rather than just a performance optimizer.

Key takeaway

For AI Architects designing clinical decision support systems, this research indicates that integrating RAG as a safety layer is crucial, especially for high-risk conditions like SCD. You should consider multi-agent architectures to modularize clinical reasoning and prioritize advanced RAG strategies like Tree Retrieval over basic semantic retrieval to minimize clinically unsafe outputs, even if average performance metrics appear similar. Focus on safety-specific evaluations, not just overall accuracy.

Key insights

RAG-enhanced multi-agent LLMs can improve safety and clinical quality in high-risk medical care planning.

Principles

• Decompose clinical reasoning into specialized agents.
• Frame retrieval as a safety control, not just optimization.
• Average performance metrics are insufficient for clinical AI.

Method

The system uses an LLM-based MAS with agents for diagnosis, investigation, and treatment. It integrates RAG via LLM-Guided Tree, Metadata-Filtered, and Semantic Similarity Retrieval strategies, evaluated by physicians and LLM-as-a-Judge.

In practice

• Implement multi-agent systems for complex clinical tasks.
• Prioritize Tree Retrieval for critical safety improvements.
• Incorporate physician review for clinical AI validation.

Topics

• LLM-based Multi-Agent System
• Retrieval-Augmented Generation
• Sickle Cell Disease
• Medical Care Planning
• Clinical Safety

Best for: AI Architect, AI Engineer, NLP Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• SEMA-RAG: A Self-Evolving Multi-Agent Retrieval-Augmented Generation Framework for Medical Reasoning — Multi-agent RAG with self-evolving, sufficiency-driven retrieval significantly improves medical question answering accuracy.
• Evaluating Multi-Agent LLM Architectures for Rare Disease Diagnosis — Multi-agent LLM architectures do not guarantee improved diagnostic accuracy; topology selection is critical.
• Beyond the Blank Stare: Injecting Medical Expertise into LLMs Through RAG-Powered Knowledge — RAG systems enhance LLM reliability in medicine by grounding responses in verifiable, external knowledge bases.
• Agentic AI-based Framework for Mitigating Premature Diagnostic Handoff and Silent Hallucination in Healthcare Applications — Multi-agent systems with deterministic orchestration and uncertainty quantification can mitigate LLM diagnostic failures in healthcare.
• LLM-Guided Evolution for Medical Decision Pipelines — LLM-guided evolution optimizes medical decision pipelines at inference-time, outperforming manual baselines through interpretable program-level changes.
• ClinicalMC: A Benchmark for Multi-Course Clinical Decision-Making with Large Language Models — Evaluating LLMs in evolving multi-course clinical scenarios is crucial for real-world healthcare deployment.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.