Nat CancerJClub
An agentic AI system enhances clinical detection of immune-related adverse events in a multi-phase validation study.
Nature Cancer · · Journal Article · Open access
Gallifant, Chen + more
Abstract ↗AI summary
The abstract is read at the publisher; the summary is JClub's.
Agentic AI with human verification extracts immune-related adverse events from clinical notes, reaching macro-averaged F1 of 0.92 for current-event detection.
- Why it matters: Immune-related adverse events occur in up to 40% of patients on immune checkpoint inhibitors, but manual chart review is laborious and inconsistent.
- What they did: The team built an agentic large language model to extract presence, temporality, severity grade, attribution, and certainty for six irAE types, then validated it retrospectively and prospectively.
- The result: Retrospective performance included 0.66 F1 for multi-class severity grading and a 0.14 F1 gain from self-consistency; prospective silent deployment over 3 months yielded binary F1 of 0.72–0.79.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.
- Open access