Neuroanesthesia & neurocritical care
Evaluation of ChatGPT in Predicting 6-Month Outcomes After Traumatic Brain Injury*
Authors' conclusion
This study does not support the use of ChatGPT for prediction of outcomes after TBI.
Study context
- Population
- Consecutive moderate-to-severe traumatic brain injury (TBI) patients admitted to a neuro-ICU of a level 1 trauma center, September 2021–October 2022
- Participants
- 80 patients
- Comparison
- ChatGPT vs Neurointensivist vs Non-neurointensivist vs IMPACT
- Outcome
- 6-month functional status dichotomized: favorable (GOSE ≥ 5) vs poor (GOSE < 5)
- Follow-up
- 6 months
Methodological detailssingle-centre, retrospective, observational
- Single-centre
- Retrospective
- Observational
Performance
As reported in the abstract
6-month outcome prognosis
| Test or model | AUC | Sensitivity | Specificity |
|---|---|---|---|
| ChatGPT | 0.62 (0.50-0.74) | 100% | 26% |
| Neurointensivist | 0.70 (0.59-0.82) | — not stated in the abstract | — not stated in the abstract |
| Non-neurointensivist | 0.71 (0.59-0.82) | — not stated in the abstract | — not stated in the abstract |
| IMPACT | 0.81 (0.72-0.91) | — not stated in the abstract | — not stated in the abstract |
AUC
ChatGPT0.62 (0.50-0.74)
Neurointensivist0.70 (0.59-0.82)
Non-neurointensivist0.71 (0.59-0.82)
IMPACT0.81 (0.72-0.91)
Sensitivity
ChatGPT100%
Neurointensivist— not stated in the abstract
Non-neurointensivist— not stated in the abstract
IMPACT— not stated in the abstract
Specificity
ChatGPT26%
Neurointensivist— not stated in the abstract
Non-neurointensivist— not stated in the abstract
IMPACT— not stated in the abstract
— not stated in the abstract
Secondary outcomes
Prediction of intracranial hypertension management, pulmonary infectious risk, removal of life-sustaining therapies
- ChatGPT generally less accurate than clinicians, with lower AUC for most outcomes
Original publication
Evaluation of ChatGPT in Predicting 6-Month Outcomes After Traumatic Brain Injury*. Crit Care Med. Published 6 March 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →