aceAgreement and classification performance of ChatGPT-5-generated and nurse-led emergency department triage using an expert reference decision: A retrospective study
This study evaluated the agreement and classification performance of ChatGPT-5-generated triage compared with nurse triage using an expert reference decision in adult emergency department patients. In this retrospective cross-sectional study, 1564 adult patients with documented yellow or green triage classifications were included. Standardized triage-level patient information was entered into ChatGPT-5 using an identical prompt for all patients. Agreement was assessed using Cohen's κ, and classification performa…
For adult emergency patients classed yellow or green, ChatGPT-5 triage reached higher accuracy and sensitivity against an expert reference than nurse triage, so it undertriaged fewer yellow patients.
In expert-reference green patients, ChatGPT-5 triage assigned yellow more often than nurses did (15.0% vs. 2.3%), which suggests more over-triage of low-acuity patients.
The findings apply only to the conditions evaluated: retrospective data on adult patients with documented yellow or green classifications, so performance on other acuity levels or in live use is not established.
Evidence
- Peer-reviewedInternational Emergency Nursing2026-10-12
How should this claim be treated?
Truvace Impact Record TRV-2026-1421, v1: “Agreement and classification performance of ChatGPT-5-generated and nurse-led emergency department triage using an expert reference decision: A retrospective study.” Truvace, 2026-10-15. /record/TRV-2026-1421 (accessed at citation time). sha256 48c5ade12d3053eb…
Calibration history
Every change to this record since certification, in the open. None yet — the reading has held since it entered the record.
Certified into the record
How to verify without trusting this page
Fetch the canonical text of any version from /api/record/TRV-2026-1421 and hash it yourself — for example shasum -a 256 on the saved canonical field. The result must equal content_hash, and each version’s text ends with prev:followed by the prior version’s hash (version 1 chains to 64 zeros). If a single character of any version had been altered since certification, the chain would not reproduce.