The approach proposed here could be seen as a next step in the PsySTART program designed to refine the selection of risk factors and optimize the weighting scheme used to combine information about these risk factors into a composite risk score. These refinements would require data to be collected from a much larger sample than in the WMH analysis. The sample should include a baseline assessment of a broad range of risk factors obtained in the immediate aftermath of disaster. Participants should be followed over time to determine who develops PTSD or other post-disaster mental disorders. Much more sophisticated data analysis methods should be used to analyse these data than in the WMH analysis. In particular, machine learning methods designed to maximize out-of-sample prediction accuracy should be used to develop the final model (Kessler et al. 2014), leading to optimal selection of the risk factors to include in subsequent assessments and to optimal weighting of these measures to assess risk of post-disaster psychopathology. We were unable to use these methods in the WMH analysis because of our small sample size.
