RESCUE: An end-to-end multi-agent LLM system for proactive rare-disease patient screening in the EHR

RESCUE, a multi-agent LLM workflow in a pediatric hospital EHR, achieved AUC 0.808 on a holdout set of n=12,591 and flagged higher genomic-assessment follow-up.

• Why it matters: Rare-disease screening is hindered by long diagnostic delays and by tools that require pre-identified patients or laborious, institution-specific rules, limiting clinical utility and feasibility.
• What they did: RESCUE uses four sequential agents to map data into a four-tier genetic-evidence taxonomy, train a positive-unlabeled XGBoost classifier, apply rule-based screening across the EHR, and review notes for medical necessity and operations.
• The result: Among 175,842 eligible patients, those flagged were 7.4-fold more likely to receive genomic assessments, and blinded chart review found 80% precision for previously missed, appropriate ES/GS candidates.

AI summary of the abstract · may contain errors · not medical advice · read the paper.