AI Tool Predicts PTSD Risk From ER Patient Accounts

NYU Langone researchers built PREVIA, a tool that flags high-risk trauma patients one month after injury using emergency room narratives.

Updated on Nov. 9, 2026 in Artificial Intelligence

AI Tool Predicts PTSD Risk From ER Patient Accounts

Live Poll

Would you trust an AI tool to screen you for PTSD risk in the ER?

Researchers at NYU Langone Health developed PREVIA, an artificial intelligence tool that predicts post-traumatic stress disorder risk by analyzing emergency room patient descriptions of how their injuries occurred. The findings were published in the journal Nature Communications.

Why it matters

Existing PTSD questionnaires are hard to administer in emergency rooms because of limited clinician time and overcrowding, leaving many at-risk patients unscreened. A tool that works from routine intake conversations could allow earlier identification during the first month after trauma, when intervention may be most effective.

In a 30-patient clinical evaluation, PREVIA matched high-risk PTSD patients with 100% accuracy, compared with 89% for PTSD-specialized emergency clinicians working from the same cases. The 95.4% accuracy figure comes from narratives drawn from 145 emergency room patients, but the study does not yet show how the tool performs in everyday emergency care.

The players

NYU Langone Health

A New York-based academic medical center whose research team developed PREVIA and published the findings.

Katharina Schultebraucks

The NYU Langone researcher who led the team that developed PREVIA.

The details

PREVIA was built on the open-source Llama and Mistral models and searches patient statements for early PTSD-related signs such as loss of control, memory loss, and depersonalization. It analyzes routine emergency room questions and answers and generates summaries that integrate underlying conditions, socioeconomic circumstances, and social support, while explaining the basis for each risk classification.

The first month after trauma is identified as the window for flagging high-risk PTSD patients.
Findings were published in Nature Communications on November 9, 2026.

Roadmap

PREVIA is built on the open-source Llama and Mistral models, reflecting a broader move in medical AI toward adapting general-purpose language models to specific clinical screening tasks. It positions emergency screening as a place where such tools could replace time-intensive questionnaires that clinicians rarely have the capacity to administer.

If validated in wider use, screening built into routine emergency room intake could identify patients at elevated PTSD risk without adding questionnaires that strain staff time. Patients flagged early could be offered follow-up support during the first month after an injury, when the source identifies intervention as most useful.

The takeaway

Early, low-burden screening could shift PTSD care from reactive treatment toward timely prevention, provided tools like PREVIA are tested beyond initial study samples. Clinicians and health systems should watch for replication results before relying on automated risk flags.