LLM Framework for Detecting Cutaneous Immune-Related Adverse Events
July 24, 2026
A retrieval-augmented, multi-agent LLM framework for identifying cutaneous immune-related adverse events (cirAEs) from clinical notes achieved an F1 score of 0.88. The human-in-the-loop system improved inter-rater agreement to 0.82 kappa and reduced manual review time by 50% compared to unassisted review.
HOW THIS AFFECTS YOU
●
builderThe high F1 score and reduced latency suggest viable patterns for clinical-grade agentic workflows.
●
healthThis demonstrates a validated path for using LLMs to scale clinical toxicity monitoring.