this process using the weights from assessing them one at a time (ERS1) and jointly (ERS2). It appears that in terms of overall prediction, ERS1 and ERS2 were very similar in performance (Table 3), however, ERS2 was often slightly better in terms of risk stratification (Table 4). It is not possible to conclude definitively, without extensive and exhaustive simulation studies, which one performs better. Also, one could modify ERS2 by filtering potentially correlated predictors through variable selection, and reducing its variability. Although high risk groups were identified by the ERS in the present study, the ERS showed only modest improvement in lipid-related risk prediction of above and beyond the effect of traditional risk factors including sociodemographic and dietary factors (e.g., AUC improvements of 0.72 to 0.82, Table 3 and Table S5). This finding may not be surprising because a marker with an OR of 3 or lower is usually a poor tool for classifying or predicting risk for individuals [48]. In fact, the improvements of risk prediction/classification by the ERS are similar to the AUC improvements for coronary heart disease risk prediction by genetic risk scores (GRS) found in the Atherosclerosis Risk in Communities (ARIC) (from 0.742 to 0.749), Rotterdam
