Paramedic PapersHome Operations & systems Trained ChatGPT matched paramedics on triage scenarios
Operations & systems
Trained ChatGPT matched paramedics on triage scenarios
Utilization of Large Language Models in Prehospital Triage: A Focus on ChatGPT’s Performance According to Pre-KTAS
Fire Science and Engineering · Added 3 November 2026
A Korean simulation study of 100 prehospital triage scenarios found a Pre-KTAS-trained ChatGPT matched 10 active-duty paramedics on accuracy (0.81 vs 0.80) with the lowest under-triage rate (6.1%), while an untrained LLM lagged well behind at 0.69.
- Study type
- Simulation
- Design
- Simulation-based comparative study, 100 Pre-KTAS scenarios × 30 independent repetitions
- Participants
- 100 scenarios; trained and untrained LLM vs 10 active-duty 119 paramedics
- Main finding
- Accuracy: trained LLM 0.81 vs paramedics 0.80 vs untrained LLM 0.69 (trained LLM and paramedics > untrained, p<.001); trained LLM lowest under-triage (6.1%) and highest sensitivity (0.95); paramedics highest specificity (0.96) and PPV (0.93)
- Caveat
- Simulated text scenarios, Korean triage scale, repetition inflates apparent precision, no field validation
- Bottom line
- Hypothesis-generating
Why it matters in Australia (Medium relevance). Text scenarios, a Korean triage scale and no field testing — but LLM triage support is coming, and this is a tidy early look at how much the training material matters.
Learning goal
Weigh the role of artificial intelligence tools in supporting prehospital triage accuracy and under-triage risk.
More
- Topic: Operations & systems
- Daily scan: 3 November 2026
Related papers on Operations & systems
-
US Medicaid cuts threaten EMS funding
A US Health Affairs commentary warns that H.R. 1 and subsequent Centers for Medicare and Medicaid Services rules will deepen financial strain on local EMS systems already struggling with inadequate reimbursement.
-
Turning HEMS debriefs into shared learning
Focus groups with 30 HEMS clinicians found a quality improvement project capturing routine debriefs ('Coffee and Cases') strengthened organisational learning through three themes: education, accessibility and calibration of learning points.
-
Remote reads of paramedic ultrasound held up in pilot
A usability pilot had four physicians interpret paramedic-acquired lung and IVC ultrasound remotely via a cloud platform across laptop, tablet and smartphone, with substantial diagnostic agreement (0.68–0.80) and the laptop version rated most usable.
-
Predicting ground vs helicopter EMS travel times
A modelling study validated open-source routing and historical-data methods for predicting ground and helicopter EMS travel times, achieving R² of 0.79 for ground drive time to scene but only 0.38 for helicopter time to hospital.