LLM ensemble PREVIA predicts future PTSD risk from emergency department trauma narratives with expert-level accuracy.

PREVIA, a clinically informed large language model ensemble, predicts 1-month post-traumatic stress disorder risk from emergency department trauma narratives with AUROC 78.49% ± 5.75% in 145 New York City patients.

• Why it matters: Emergency clinicians need early PTSD risk stratification after discharge, but time constraints make standard screening tools impractical, leaving a gap for scalable, timely identification.
• What they did: In a longitudinal study, 145 patients from two New York City level-1 trauma centers provided unstructured trauma narratives in response to a routine-care question, and PREVIA extracted early PTSD symptoms, peri-traumatic symptoms, risk, and protective factors to predict PTSD 1-month post-trauma.
• The result: PREVIA achieved sensitivity 95.4% ± 3.99% and precision 46.76% ± 8.66%, matched an expert PTSD clinician on 30 randomly selected patients, and produced interpretable explanations without added clinical burden.

AI summary of the abstract · may contain errors · not medical advice · read the paper.