Mejora de la precisión del triaje de traumatismos con modelos de lenguaje grandes: una comparación con las decisiones...

Accurate prehospital trauma triage and communication determine morbidity, mortality, and system efficiency. Advancements in large language models (LLMs) offer an opportunity to improve triage and yet to be implemented in prehospital trauma triage.

This retrospective cohort study evaluates LLM performance in trauma triage and accuracy of prehospital tele-communication. Of 410 pediatric activations at a Level I center (January 2023-May 2025, IRB #00009569), 133 activations with EMS recordings, human-generated trauma pages, and injury severity scores (ISS) were analyzed. Audio was transcribed with OpenAI Whisper. Structured "Essential Transcripts" were generated with Named Entity Recognition (NER). Entity ablation tested redaction of triage parameters on accuracy. In a prospective arm, trauma surgeons reviewed EMS transcripts and triaged activations pre and post LLM exposure. Cribari criteria defined over and under-triage. McNemar's test and 95% confidence intervals assessed paired differences in accuracy.

The primary endpoint: LLM under-triage demonstrated modest improvement; 4.8% (3-8.2) vs. 5.1% (3.1-9.3, p 0.73, Bonferroni p =1). For secondary endpoints, LLM triage outperformed human clinicians: 83.5% accuracy (80.6-90.6) vs. 78.9% (73.9-82.6, p<0.01, Bonferroni p <0.01), with over-triage 58.6% (51.4-73.7) vs. 71.8% (p <0.05, Bonferroni p <0.09). "Essential Transcripts" reduced transcript length by 80.8% (p<0.001) while preserving accuracy (81.9%, 76.6-87.5, p <0.001, Bonferroni p <0.05). Entity ablation had marginal effect on triage. In prospective evaluation, human triage accuracy improved following LLM exposure (73.7% (69.8-77.2)) to 75.8% ((71.9-79.2), p=0.04, Bonferroni p = 0.12), significantly improving the odds of a correct triage decision (OR 2.57, 95% CI 1.39-6.83, p<0.01).

LLMs achieve triage accuracy comparable to trauma staff in retrospective review of pediatric trauma. Further validation is needed to assess clinical outcomes, generalizability, and user acceptance before widespread deployment.