Response to: Prediction of Outcome after Traumatic Brain Injury Using Clinical and Neuroimaging Variables: A Methodological Mistake

We appreciate the comments made by Dr. Sabour about our manuscript. We agree that the term ‘prediction’ can be misleading. We used the term as a general one, since our study employed an explanatory rather than a predictive model.1 We estimated which factors were associated with an unfavorable outcome (defined as a Glasgow outcome scale ≤4) rather than predicting the prognosis of individual patients. Validation using divided data sets or bootstrapping, as recommended, is rarely performed for this type of study and cannot be applied to our study because of the smallness of the sample.2 We admit that the confidence intervals for ‘motorcycle accident’ and ‘being an accident victim’ are too wide, which was due to the sample size being insufficient for a multivariable model including all the potential variables. In particular, the effect of being an accident victim requires careful interpretation and a detailed analysis to ensure that the number of samples in each group is sufficient, because it can be affected by the cause and severity of injury. We acknowledge this limitation and hence metioned in the Discussion that a large multicenter study is needed to confirm our findings. However, the factors ‘motorcycle accident,’ ‘being an accident victim,’ and ‘diffuse axonal injury on brain magnetic resonance imaging’ were found to be consistently and significantly correlated with unfavorable outcomes in both univariate and multivariate analyses. Not only are these factors statistically significant, but it is also plausible that they could affect the outcome, as described in the Discussion.