Llms struggle to update decisions during emergency room triage conversations
LLMs Anchor on Chief Complaint and Fail to Integrate Evidence in Sequential Clinical Triage
Computation and Language
Summary
This paper studies how large language models (LLMs) make decisions during emergency room triage, where nurses and patients talk turn by turn. The authors found that while models do well when seeing the whole conversation, they struggle to improve their decisions as the conversation unfolds. Models mainly focus on the initial complaint and fail to properly consider new information shared later. Human experts do much better at updating decisions, showing current models miss key parts of the process.
What this means in practice
- •For emergency department managers: Evaluate triage software beyond offline tests by examining model performance at each step of nurse-patient talks to ensure real-time reliability.
- •For clinical ai developers: Design and test triage models that better integrate new information in sequential conversations instead of relying mainly on initial complaints.
Authors
Dipankar Srirag, Haokai Zhao, Ashutosh Kumar, Eleanor Hopper, Michael Dalton, Quoc Dung Nguyen, Aditya Joshi, Salil S. Kanhere, Padmanesan Narasimhan
Abstract
Triage in the emergency department (ED) is a sequential decision process that unfolds turn by turn. Existing evaluations of large language models (LLMs) for triage use completed retrospective records and report performance close to that of physicians. We implement a methodology for evaluating LLMs on sequential triage, the task of predicting a triage acuity label from a growing prefix of a nurse-patient conversation. We evaluate six LLMs at five sequential checkpoints on two corpora: 425 LLM-generated (SIMULATED) and 50 physician-authored (CLINICIAN) conversations, both labelled under the Emergency Severity Index (ESI). Every model, measured by quadratic weighted kappa (QWK), degrades from moderate-to-substantial agreement on completed records to fair-to-moderate agreement at every sequential checkpoint. Controlled perturbations show that the label at every checkpoint is anchored on the chief complaint exchanges, and prompting interventions fail to lift this plateau. Models extract clinically relevant content from later turns, yet the surprisal of the true label rises across the checkpoints. So the model fails to integrate the evidence. Three expert clinicians on the same conversations reach a QWK of 0.887-0.929, while the best model reaches 0.295. Predictions concentrate at ESI-2 and ESI-3, and models agree with each other more than with the ground truth, so ensembling worsens the failure. Deploying LLMs for ED triage based on offline benchmarks alone misses this sequential failure.