This research introduces MAP (Medical Agent Pathway), a novel multi-agent framework that enhances large language models for complex inpatient clinical decision-making.

• Developed comprehensive system for medical triage, diagnosis, and treatment planning
• Created new benchmark for inpatient pathway evaluation with real medical cases
• Demonstrated significant performance improvements through multi-agent collaboration compared to single-agent approaches
• Identified key limitations in current LLMs for clinical reasoning and outlined future research directions

Significant for healthcare as it addresses the critical gap between AI capabilities and the complex, multifaceted nature of inpatient care, potentially reducing physician cognitive load and improving patient outcomes.

MAP: Evaluation and Multi-Agent Enhancement of Large Language Models for Inpatient Pathways