Predicting PTSD development with early post-trauma assessments: a proof-of-concept for a concise tree-based classification method

<b>Background:</b> Approximately 70% of individuals globally experience at least one traumatic event in their lifetimes, potentially leading to posttraumatic stress disorder (PTSD). Understanding the development of PTSD and devising effective prevention and treatment strategies are crucial. This proof-of-concept study aimed to design a concise tree-based adaptive test using the Classification and Regression Trees (CART) framework to predict PTSD development.<b>Methods:</b> Utilizing data from a longitudinal neuroimaging study, adult trauma survivors were enrolled from local hospital emergency departments within 48 h of experiencing trauma. Participants who completed psychological evaluations within 2 weeks post-trauma and a PTSD diagnosis assessment at 3 months were included in the analytic sample (<i>n</i> = 143). A total of 131 features including demographic, trauma-related, and behavioural and clinical symptoms were collected during this initial two-week post-trauma period. The performance of the CART model was benchmarked against two of the most powerful and widely used machine learning algorithms in the field, Random Forest (RF) and Gradient Boosting (GB) models.<b>Results:</b> The CART model, which incorporates just three critical questions from established assessments, predicted PTSD development with performance closely matched to that of the RF and GB models. The CART model achieved an accuracy of 0.641 and an AUC of 0.663, which showed only slightly worse performance compared to the RF and GB models. Its efficiency in utilizing a minimal set of questions for prediction highlights its potential for practical application in early PTSD detection and intervention strategies.<b>Conclusion:</b> The CART framework demonstrates a streamlined and efficient method for predicting PTSD onset in trauma survivors. While showing promise for practical application, further validation and refinement are necessary to enhance its predictive performance and establish its broader utility in early intervention strategies.