Posttraumatic Stress Disorder Hyperarousal Event Detection Using Smartwatch Physiological and Activity Data

Posttraumatic Stress Disorder (PTSD) is a psychiatric condition affectingnearly a quarter of the United States war veterans who return from war zones.Treatment for PTSD typically consists of a combination of in-session therapyand medication. However; patients often experience their most severe PTSDsymptoms outside of therapy sessions. Mobile health applications may addressthis gap, but their effectiveness is limited by the current gap in continuousmonitoring and detection capabilities enabling timely intervention. The goal ofthis article is to develop a novel method to detect hyperarousal events usingphysiological and activity-based machine learning algorithms. Physiologicaldata including heart rate and body acceleration as well as self-reportedhyperarousal events were collected using a tool developed for commercialoff-the-shelf wearable devices from 99 United States veterans diagnosed withPTSD over several days. The data were used to develop four machine learningalgorithms: Random Forest, Support Vector Machine, Logistic Regression andXGBoost. The XGBoost model had the best performance in detecting onset of PTSDsymptoms with over 83% accuracy and an AUC of 0.70. Post-hoc SHapley AdditiveexPlanations (SHAP) additive explanation analysis showed that algorithmpredictions were correlated with average heart rate, minimum heart rate andaverage body acceleration. Findings show promise in detecting onset of PTSDsymptoms which could be the basis for developing remote and continuousmonitoring systems for PTSD. Such systems may address a vital gap injust-in-time interventions for PTSD self-management outside of scheduledclinical appointments.