Healink multi-agent framework beats doctors in post-discharge follow-ups
AI framework outperforms human physicians in clinical safety and authoritativeness.
Post-discharge clinical follow-up is critical for continuity of care but suffers from workforce shortages and fragmented records. Researchers from multiple institutions present Healink, a memory-enhanced multi-agent framework that uses large language models to generate prescription-grounded, traceable responses for post-discharge communication. The architecture combines a triage routing mechanism, a unified memory enhancement module with a relational database for low latency, and a constraint-based retrieval-augmented generation (RAG) engine that vectorizes historical records and uses weighted similarity functions to prevent cross-departmental drug conflicts.
Healink was evaluated on a dataset of 400 continuous and 85 highly complex real-world follow-up cases, plus the webMedQA benchmark. In a single-blind evaluation by clinical experts, it outperformed human physician baselines in both authoritativeness and clinical safety. By generating white-box evidence chains, Healink provides a scalable, safe paradigm for intelligent patient management that could significantly improve societal healthcare outcomes.
- Healink uses a multi-agent architecture with triage routing and a unified memory module for low-latency retrieval.
- Outperformed human physician baselines in authoritativeness and clinical safety in single-blind evaluation.
- Generates traceable white-box evidence chains, preventing cross-departmental drug conflicts.
Why It Matters
Healink could reduce hospital readmissions and improve patient outcomes through scalable AI-powered post-discharge care.