Abstract
Background: The likelihood of developing Posttraumatic Stress Disorder (PTSD) depends on the interaction of individual risk factors and cumulative traumatic experiences. Hence, the identification of individual susceptibility factors warrants precise quantification of trauma exposure. Previous research indicated that some traumatic events may have more severe influences on mental health than others; thus, the assessment of traumatic load may be improved by weighting event list items rather than calculating the simple sum score. Objective: We compared two statistical methods, Random Forests using Conditional Interference (RF-CI) and Least Absolute Shrinkage and Selection Operator (LASSO), based on their ability to rank traumatic experiences according to their importance for predicting lifetime PTSD. Methods: Statistical models were initially fitted in a sample of N(1) = 441 survivors of the Northern Ugandan rebel war. The ability to correctly predict lifetime PTSD was then tested in an independent sample of N(2) = 211, and subsequently compared with predictions by the simple sum score of different traumatic event types experienced. Results: Results indicate that RF-CI and LASSO allow for a ranking of traumatic events according to their predictive importance for lifetime PTSD. Moreover, RF-CI showed slightly better prediction accuracy than the simple sum score, followed by LASSO when comparing prediction results in the validation sample. Conclusion: Given the expense in time and calculation effort by RF-CI and LASSO, and the relatively low increase in prediction accuracy by RF-CI, we recommend using the simple sum score to measure the environmental factor traumatic load, e.g., in analyses of gene × environment interactions.