Missing data
Since some of the covariates contained missing data, we imputed missing data through multiple imputation by chained equations approach.(30) We assumed that the pattern of missing data was “missing at random,” and created 20 imputed data sets (after 25 burn-in iterations). All analyses were performed for each of the 20 data sets, and we combined the results of the analyses of imputations based on Rubin’s principles.(31) Details regarding missing covariate data are shown in Supplemental Table 1 .