Predicting PTSD symptoms in firefighters using a fear-potentiated startle paradigm and machine learning

This study develops a fear-potentiated startle paradigm (FPS) and a machine learning approach to accurately predict PTSD symptoms using electrogram data. A three-phase fear-potentiated startle paradigm was designed to assess the conditioning, generalization, and extinction of fear. Electrooculogram and electrocardiogram signals were collected during the FPS. A total of 1107 Chinese firefighters participated in the study. The Chinese version PCL-C was administered to all subjects. A cutoff of 38 or higher is used to indicate PTSD symptoms. Electrogram features were extracted and selected to build a machine learning model to classify individuals. The machine learning model was 5-fold cross validated. The importance of the selected features was calculated. Classification performance metrics were evaluated for the machine learning model. The machine learning model can identify firefighters with a PCL-C score of 38 or above with sensitivity and specificity both above 0.85 when 5-fold cross validated on a 1107-person sample. The area under the receiver operating characteristic curve of the model is 0.93. Features related to fear generalization are found to be the most important. The proposed fear-potentiated startle paradigm and machine learning approach can accurately predict PTSD symptoms in Chinese firefighters, which can improve the screening and diagnosis of PTSD.