This prospective longitudinal study used machine learning to analyze pre-deployment data from 473 active-duty Army personnel deployed to Afghanistan to predict post-deployment PTSD. The dataset included polygenic, epigenetic, metabolomic, endocrine, inflammatory, and clinical lab markers; neurocognitive testing; and symptom self-reports. Machine-learning models (random forest and support vector machine) predicted PTSD diagnosis 90-180 days post-deployment (AUCs of 0.78 and 0.88, respectively) and longitudinal PTSD symptom trajectories (AUCs of 0.85 and 0.87, respectively). Pre-deployment sleep quality, anxiety, depression, sustained attention, and cognitive flexibility were among the strongest predictors, complemented by blood-based biomarkers. The models show promise for determining deployment readiness and informing pre-deployment interventions.

Katharina Schultebraucks, Meng Qian, Duna Abu-Amara, Kelsey Dean, Eugene Laska, Carole Siegel, Aarti Gautam, Guia Guffanti, Rasha Hammamieh, Burook Misganaw, Synthia H. Mellon, Owen M. Wolkowitz, Esther M. Blessing, Amit Etkin, Kerry J. Ressler, Francis J. Doyle III, Marti Jett, Charles R. Marmar