The survival models were initially estimated separately for each TE type (the 27 listed TEs, other TEs and private TEs) across all countries combined, with dummy variables to control for between-country differences in prevalence. The implicit assumption was that the associations (odds ratios; ORs) of predictors with outcomes were constant across countries. Given the large number of TE types considered, the person-year data files for the 29 TE types were then pooled into more highly aggregated data files, one for the set of TEs in each factor uncovered in the exploratory factor analysis. Inspection of coefficients in the 29 trauma-specific models was used to confirm the validity of this pooling. Interaction tests subsequently were used to evaluate the assumptions of constant ORs across countries and TE types within each factor. Logistic regression coefficients and their standard errors were exponentiated to create ORs and 95% confidence intervals.
