An Electronic Health Record–Integrated, Large Language Model–Powered Tool to Triage Surgical Patients

Abstract

Key Points Question Can surgical patient triage be automated using a large language model (LLM) agentic workflow? Findings In this quality improvement study, the LLM tool recommended hospitalist consultation for nearly a quarter of the 6193 triaged cases. The tool achieved 94% sensitivity and 74% specificity, and post hoc medical record review suggested that most discrepancies reflected modifiable gaps in clinical criteria, institutional workflow, or physician practice variability, rather than LLM misclassification. Meaning The findings of this study suggest that an LLM-powered human-in-the-loop agentic workflow could accurately triage surgical patients for a surgical comanagement service.