About this project
We develop person-specific models that forecast short-term occurrences of PTSD symptoms (e.g., re-experiencing, physical arousal, avoidance). Models learn from in-the-moment ecological momentary assessment (EMA) surveys and passive sensing from phones and wearables (e.g., Garmin, Empatica). This high-resolution daily-life data allows us to test not only when symptoms are about to occur, but also why—capturing triggers such as interpersonal stressors, contextual factors, and momentary emotional states.
The current aim is to demonstrate the feasibility and prediction accuracy for each individual. In later phases, these models will lay the foundation for personalized just-in-time adaptive interventions (JITAIs)—digital support systems that deliver targeted therapeutic tools exactly when an individual’s personal model detects they are at high risk.
Why does it matter?
PTSD is a highly prevalent and debilitating disorder, with symptoms that fluctuate widely across hours and days. Despite evidence-based treatments, many individuals either do not access care or continue to experience significant distress. Traditional approaches often overlook these dynamic, person-specific fluctuations.
Our approach detects risk moments in real-time. By combining EMA and passive sensing, we can identify fine-grained symptom patterns and their antecedents, such as stressors, interpersonal context, environmental exposure, sleep patterns, and more. This knowledge not only enhances scientific understanding of PTSD processes but also provides the foundation for scalable interventions that can reach those who need support the most.
How does it work?
Everyday data: Six EMA prompts per day (about 90 seconds each) across 30 days.
Short-term prediction: Each participant receives a personal model that predicts symptom occurrence in the near future, using machine-learning methods trained on their own data.
Context modeling: Passive sensing via smartphones and wearables captures geolocation, activity, heart rate, and daily rhythms. Integrating these signals enables us to examine how interpersonal and contextual triggers influence symptom trajectories.
Stability testing: Two weeks after the first EMA wave, participants complete a second 10-day EMA period, allowing us to test the reliability of personal models over time.
Next steps
This proof-of-concept will inform larger studies aiming to integrate predictive models into real-time digital interventions. By advancing personalized and scalable mental health tools, our goal is to transform PTSD care—from static diagnosis to dynamic, moment-to-moment support.