PROJECT SUMMARY/ ABSTRACT Currently, no accurate prognostic model of posttraumatic stress following trauma exposure for emergency department (ED) patients is available at the point-of-care without requiring clinical screening or diagnostic interviews. The proposed research is based on the rationale that the 139 million annual ED visits provide a critical window to proactively plan risk-based follow-up care at an early stage, where patients are still in contact with the health care system. While clinical interviews are still the gold standard to screen for acute stress symptoms following trauma exposure, their feasibility in clinical practice as routine screenings in the ED is severely limited given the acute care priorities in the ED. The long-term goal of the proposed research is to develop a prognostic model that is accurate, scalable, practical, and feasible with low additional burden on the highly taxed ED procedures. The overall objective is to use advanced computational methods to extract objective markers for posttraumatic stress from video and audio data to build a clinical readout at the point-of- care that will enable ED clinicians to prognosticate the risk for posttraumatic stress disorder (PTSD). Based on our preliminary data, we hypothesize that voice and speech content, head movement, pupil dilation, gaze, and facial landmark features of emotion provide probabilistic information that will allow us to identify digital biomarkers for PTSD. This hypothesis will be tested by pursuing two specific aims directed at analyzing digital biomarkers to predict 1) who is at risk to develop PTSD and 2) to combine digital biomarkers with routinely available electronic health records to predict at the point of care who will develop PTSD one month after ED discharge to plan follow-up specialty care and who is at risk for chronic PTSD. This proposed prospective longitudinal study will chart PTSD symptoms in a cohort of 350 trauma survivors. The proposed research is of high clinically significance. The prognostic model will facilitate risk-targeted early interventions for curtailing delayed treatment, assist clinicians in prioritizing treatment allocation and reduce downstream health care costs. This research project aims to deliver an objective, accurate, and reliable digital measure for patients’ well-being. Such digital biomarkers will enable more efficient discharge planning and will promote early prevention strategies. The mental besides the physical well-being of trauma-survivors admitted to the ED after a life-threatening event is of high value and is the foundation of a well-functioning, high-quality emergency care system. The SARS-CoV-2 pandemic, future disasters, or other large-scale emergencies underscore the critical need to support highly charged EDs through computational methods to better determine risks of long- term mental health care needs without disrupting the standard operating procedures of acute care.

PROJECT NARRATIVE We propose a novel digital phenotyping approach using deep learning to identify objective digital biomarkers from video-recorded interviews to predict posttraumatic stress disorder risk after Emergency Department (ED) discharge to better target early interventions and reduce delayed treatment. We will determine how candidate digital biomarkers that are designed for practical feasibility under the demanding acute care priorities in the ED add to the prognostic value of psychometric assessments and electronic health records. The proposal contributes to the NIMH’s mission aiming to “enable practical, scalable, and sustainable mental health screening and triage, and providing interventions at scale”.1