JMIR Publications

ABSTRACT

Background:

The release of ChatGPT in November 2022 drastically reduced the barrier to using artificial intelligence (AI) by allowing a simple text based online interface to a Large Language Model (LLM). One use case where ChatGPT could be useful in triaging patients at the site of a disaster using the Simple Triage and Rapid Treatment (START) protocol. However, LLMs suffer from several common errors including hallucinations (also called confabulations) and prompt dependency.

Objective:

This study addresses the research problem: "Can ChatGPT adequately triage simulated disaster patients using the START protocol?" by measuring three outcomes: repeatability, reproducibility, and accuracy.

Methods:

Nine prompts were developed by five disaster medicine physicians. A Python script queried ChatGPT Version 4 for each prompt combined with 391 validated patient vignettes. Ten repetitions of each combination were performed: 35 190 simulated triages. Results were evaluated using a Gauge Repeatability and Reproducibility study (Gauge R and R). Repeatability was defined as variation due to repeated use of the same prompt. Reproducibility was defined as variation due to use of different prompts on the same patient vignette. Accuracy was defined as agreement with a pre-determined reference standard.

Results:

Although 35 102 (99.7%) queries returned a valid START score, there was considerable variability. Repeatability (use of the same prompt repeatedly) was 14.0% of overall variation. Reproducibility (use of different prompts) was 4.1% of overall variation. Accuracy of ChatGPT for START was 63.9% with a 32.9% over-triage rate and a 3.1% under-triage rate. Accuracy varied by prompt with a maximum of 71.8% and a minimum of 46.7%.

Conclusions:

This study indicates that ChatGPT version 4 is insufficient to triage simulated disaster patients via the Simple Triage and Rapid Treatment (START) protocol. It demonstrated suboptimal consistency, with its overall precision falling below 50%. Overall accuracy of triage was only 63.9%. Healthcare professionals are advised to exercise caution while employing commercial Large Language Models (LLMs) for vital medical determinations, given that these tools may commonly produce inaccurate data, colloquially referred to as hallucinations or confabulations. Artificial intelligence guided tools should undergo rigorous statistical evaluation — using methods such as Gauge R and R — before implementation into clinical settings. Clinical Trial: NONE