AI Analysis of EMS Calls Aids Pediatric Trauma Decision-Making

Misclassification can delay lifesaving interventions or lead to overuse of critical resources, while pediatric cases add further complexity because of age-specific physiology and limited provider familiarity. Hospitals need tools that can translate unstructured prehospital information into reliable activation decisions. A new study shows that a large language model can support pediatric trauma triage by improving the signal extracted from EMS communications.

Surgeons and trainees at the Jacobs School of Medicine and Biomedical Sciences at the University at Buffalo assessed whether a large language model (LLM) could improve prehospital communication for pediatric trauma activations. The team focused on the first notification call from paramedics, when hospitals must choose an activation level with minimal data. They examined whether an AI assistant could help reduce undertriage and overtriage by clarifying key clinical details before the patient arrives.

The project implemented the LLM as a “communication-aware” triage assistant. It processed transcripts of EMS calls, removed nonessential content, and extracted clinically relevant elements such as mechanism of injury, vital signs, mental status, and bleeding indicators. It then produced a structured summary with a recommended activation level that the physician could accept, modify, or override.

The evaluation used 133 pediatric emergency department (ED) activations and was published online on June 12, 2026, in the Journal of the American College of Surgeons. The pediatric focus was intentional because children’s physiologic responses to injury differ by age and can mask decompensation, and because paramedics encounter fewer pediatric trauma cases. The study reported that more than 98% of words in EMS call transcripts were nonmedical, and the LLM compressed transcripts by about 80% while preserving accuracy.

Triage accuracy achieved by the LLM was comparable to that achieved by trauma staff. Importantly, when clinicians initially made an incorrect decision, exposure to the LLM output tripled the odds that they would correct it. The researchers emphasized that human oversight remains essential and that the current value of the system is as a real-time cognitive aid to improve communication and preparation.

“In most medical settings, the danger lies in making the wrong diagnosis. In prehospital trauma, the bigger danger is that critical information never reaches the right people in time,” said Ascharya K. Balaji, M.D., a 2025 Jacobs School alumnus who is now a surgery resident at Tripler Army Medical Center.

“Right now, LLMs are promising cognitive aids, not replacements for clinical judgment. What the results do show is that LLMs can match or slightly exceed human accuracy in interpreting EMS communications and—perhaps more importantly—that when human clinicians see an LLM recommendation alongside their own thinking, they make better decisions,” said Peter C. W. Kim, M.D., Ph.D., corresponding author, vice chair for research and innovation in the Department of Surgery at the Jacobs School and head of its Data Augmented Research Technology in Surgery laboratory.

Related Links
University at Buffalo