A patient’s first account of why they came to an emergency department may contain signals that help forecast their risk of developing post-traumatic stress disorder, according to a new study led by researchers at NYU Langone Health.
The finding points to a possible new use for generative AI in a setting where clinicians must make fast decisions and where conventional mental-health screening can be difficult to add to already crowded intake workflows. The researchers say their system, called PREVIA, analyzes a patient’s spontaneous description of the injury or event that brought them to the emergency department.
A routine question with predictive value
The study examined trauma narratives from 145 emergency-department patients. The research team reported that its set of AI models correctly identified more than 95.4% of patients considered at risk. In a head-to-head test involving 30 people from the same group who later developed PTSD, PREVIA flagged all 30 cases, while an emergency-department clinician specializing in PTSD identified 89% of them.
The work was published online in Nature Communications on November 9. The researchers describe it as the first study to show that spontaneous answers to a single routine-care question, “What brought you to the ED?”, may carry enough psychological information to forecast PTSD risk one month later.
That matters because emergency departments are often the first, and sometimes only, contact point with the health system after a physical trauma. The source article notes that about 43 million patients are admitted to U.S. emergency departments each year with physical injuries, and that more than a quarter develop PTSD according to past studies.
Why emergency screening is difficult
PTSD screening can be valuable when it identifies people who may benefit from early attention, but emergency care is defined by limited time and competing clinical priorities. Standard questionnaires can be impractical in a busy department, particularly when a patient is being evaluated for an acute injury.
PREVIA is intended to use information that is already collected rather than adding another formal assessment to the intake process. Patients’ accounts were recorded and transcribed, then analyzed for patterns associated with later PTSD risk.
The researchers say generative AI is well suited to this type of analysis because such models learn relationships among words from large quantities of language data. In this case, the system looks at how a person describes what happened and searches for language patterns that the study associates with later PTSD risk.
Potential role, not a diagnosis
The result does not mean an AI-generated risk flag is a PTSD diagnosis. The study describes a screening approach: a way to identify people who may need follow-up or a closer clinical assessment. PTSD is a complex condition, and a patient’s narrative is only one source of information available to clinicians.
Still, the reported results suggest that free-form language could become a useful addition to emergency-department decision support. A system that works from a routine narrative could be especially attractive if it can identify risk without slowing intake or requiring patients to complete another lengthy form immediately after an injury.
- The study analyzed narratives from 145 emergency-department patients.
- PREVIA was reported to identify more than 95.4% of at-risk patients.
- In a 30-case comparison, the system flagged all patients who later developed PTSD.
- The researchers frame the tool as a screener that could support, rather than replace, clinical judgment.
What comes next
The central promise is earlier recognition. If risk can be identified during or soon after an emergency visit, clinicians and health systems may have a clearer opportunity to connect patients with appropriate follow-up before symptoms become entrenched.
But the study’s sample was limited to 145 patients, and its strongest reported comparison involved 30 later PTSD cases from that group. Wider use would require validation in additional patient populations and real-world emergency settings. It would also require careful attention to how a risk estimate is explained, documented and acted on by clinicians.
For now, the research offers evidence that an ordinary clinical conversation can contain useful information that has historically been difficult to use systematically. Rather than asking patients for more during a stressful visit, the approach seeks to learn from the words they already use.
This article is based on reporting by Medical Xpress. Read the original article.
Originally published on medicalxpress.com
