Confidence-Calibrated LLM Pipeline for Adverse Drug Reaction Detection from Clinical Instant Messaging: Development and Temporal Validation

This manuscript develops and temporally validates a confidence-calibrated large language model (LLM) pipeline for adverse drug reaction (ADR) detection, entity extraction, and causality assessment from clinical instant messaging. It extends prior proof-of-concept work toward a more deployment-oriented validation setting.

Working paper (under review at JAMIA)