Screening for missed opportunities for diagnosis in the ED using etriggers and large language models.

Emergency department (ED) electronic alerts aim to flag missed diagnoses but have limited accuracy. This retrospective study tested six commercial LLMs on 288 ED cases from two high-risk cohorts, and the models collectively found missed diagnoses in 13.5% with physician agreement on 81.9% of those triggers. Model performance varied by LLM and case type, suggesting that while LLMs may help screen for missed diagnoses, organizations should require careful model selection and build human review into the process to enhance reliability.