The development of a clinically useful tool for predicting the development of psychological disorder following injury

Objectives. To identify factors significantly associated with post-traumatic stress disorder (PTSD), anxiety, and depression at 3 months post-injury; to develop a generic model to predict the occurrence of PTSD, anxiety, and depression at 3 months postinjury; and to validate this model in a test data set of patients. Design. Prospective cohort study. Methods. Participants were 823 patients attending an emergency department (ED) following accidental injury. Baseline questionnaires were completed, with 1 and 3 months postal follow-ups. Predictor variables demonstrating significant associations with two of the three outcome measures (3-month HAD anxiety and depression scores and PTSD symptoms) were included in multivariate regression models for each outcome. Non-significant predictor variables were removed until all remaining independent variables made the most significant contribution to each of the three models. Models were validated using a test dataset. Results. Previous history of mental health problems, neuroticism score and having PTSD symptoms at 1 month predicted adverse outcomes at 3 months. When used on the test datasets, the areas under the receiver operating curve (ROC) curve for the models predicting outcomes at 3 months were: PTSD 0.91 (sensitivity 88.5%); anxiety 0.87 (sensitivity 93.7%); and depression 0.87 (sensitivity 96.7%). Conclusions. The final model performed moderately well across the three outcomes and may be useful clinically as a generic rule-out tool to identify those who will not require follow up, watchful waiting or intervention