Disorder-specific and transdiagnostic vulnerability to posttraumatic stress symptoms: A machine learning approach

A wide range of biological, cognitive, affective, and behavioral risk factors have been studied in relation to posttraumatic stress disorder. Previous work has often isolated a single risk factor or a small number of risk factors, making it is difficult to know which may be the most important to study or target in interventions. We used a supervised machine learning technique, elastic net, to test the associations between posttraumatic stress symptoms (PTSS) and several self-reported risk factors at the full-scale, subscale, and item levels in a large online sample (N = 1,186) of individuals who endorsed experiencing a DSM-5 Criterion A traumatic event, allowing for a broader and more granular understanding of the associations between transdiagnostic risk factors and PTSS. In our full-scale model, posttraumatic cognitions, β = .28; anxiety sensitivity, β = .21; and posttraumatic maladaptive beliefs, β = .18, explained the largest amount of variance in PTSS. At the subscale level, heightened threat perceptions of harm, β = .30; negative cognitions about the self, β = .23; and cognitive sensitivity, β = .14, explained the largest amount of variance in PTSS. Meanwhile, at the item level, not feeling safe, not knowing oneself, and self-blame for a traumatic event had the highest importance ratings. The identified variables may be important targets in future longitudinal and treatment research.