Perhaps our most striking result was that nearly half of disaster-related PTSD occurred among the 5% of respondents with highest predicted risk scores in our model. This result is broadly consistent with several other recent studies showing that subsequent PTSD can be predicted with good accuracy using data collected in the immediate aftermath of trauma about pre-trauma risk factors, objective trauma characteristics, and early post-traumatic responses (Galatzer-Levy et al. 2014; Kessler et al. 2014; Karstoft et al. 2015). These findings contradict the previously-held view that the individual predictors in epidemiological models of PTSD have ORs too weak and inconsistent to be clinically useful in targeting people for preventive interventions (Brewin, 2005a), making it necessary to use assessment tools in the aftermath of trauma focused on current symptoms rather than risk factors (Brewin, 2005b). The error in this earlier way of thinking was in failing to appreciate that multivariate model-based predictions can be strong even when coefficients of individual predictors are weak. It is noteworthy in this regard that our high concentration of PTSD risk among the 5% of respondents with
