Repeatability, Reproducibility, and Diagnostic Accuracy of a Commercial Large Language Model (ChatGPT) to Perform Disaster Triage Using the Simple Triage and Rapid Treatment (START) Protocol
Bibliographic record
Abstract
Abstract Objective The release of ChatGPT in November 2022 drastically lowered the barrier to artificial intelligence with an intuitive web-based interface to a large language model. This study addressed the research problem: “Can ChatGPT adequately triage simulated disaster patients using the Simple Triage and Rapid Treatment (START) tool?” Methods Five trained disaster medicine physicians developed nine prompts. A Python script queried ChatGPT Version 4 with each prompt combined with 391 validated patient vignettes. Ten repetitions of each combination were performed: 35190 simulated triages. Results A valid START score was returned In 35102 queries (99.7%). There was considerable variability in the results. Repeatability (use of the same prompt repeatedly) was responsible for 14.0% of overall variation. Reproducibility (use of different prompts) was responsible for 4.1% of overall variation. Accuracy of ChatGPT for START was 61.4% with a 5.0% under-triage rate and a 33.6% over-triage rate. Accuracy varied by prompt between 45.8% and 68.6%. Conclusions This study suggests that the current ChatGPT large language model is not sufficient for triage of simulated patients using START due to poor repeatability and accuracy. Medical practitioners should be aware that while ChatGPT can be a valuable tool, it may lack consistency and may provide false information.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.137 | 0.362 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".