Menu
Back to AI and Diagnosis
02/23/26

ChatGPT Health performance in a structured test of triage recommendations.

Ramaswamy, Ashwin; Tyagi, Alvira; Hugo, Hannah; Jiang, Joy; Jayaraman, Pushkala; Jangda, Mateen; Te, Alexis E; Kaplan, Steven A; Lampert, Joshua; Freeman, Robert; Gavin, Nicholas; Tewari, Ashutosh K; Sakhuja, Ankit; Naved, Bilal; Charney, Alexander W; Omar, Mahmud; Gorin, Michael A; Klang, Eyal; Nadkarni, Girish N. Nat Med.

Collection Topics
AI and Diagnosis
Share

Most recent

Emergency department revisit quality review screening: exploring human decision-making and artificial intelligence support.

Handler, Robles-Granda, Mefford, et al., arXiv
09/09/26 AI and Diagnosis Measurement, Monitoring, and Event Analysis

A five-phase evaluation framework for diagnostic and predictive medical artificial intelligence.

Ye, Chen, Huang, et al., NPJ Digit Med.
09/03/26 AI and Diagnosis

A detection, diagnostic, and triage AI buyer’s guide: questions that matter.

Baird, Bernstein, Chambers, et al., J Digit Imaging Inform Med.
09/03/26 AI and Diagnosis Culture, Work System, and Implementation

Summary

Patients using LLMs to diagnose symptoms receive inconsistent results, posing safety risks. This study assesses ChatGPT Health’s triage accuracy using a wide range of clinician-authored scenarios. The results find that ChatGPT Health under-triaged emergencies in over 50% of cases, frequently recommending non-urgent care when immediate action was needed and failed to provide crisis resources in a suicide scenario. Across all scenarios, ChatGPT Health’s triage advice was more accurate when objective data (vital signs or lab results) were included in the vignettes. Dive into more insights from UCSF CODEX Director Sumant Ranji, MD, SFHM.

Read the paper
Ramaswamy A, Tyagi A, Hugo H. et al. ChatGPT Health performance in a structured test of triage recommendations. Nat Med. 2026;32(5):1671-1675.