We evaluated the performance of ERS using three metrics. In each case, the contribution of ERS was measured in the presence of base covariates and micronutrients retained in the model. First, we used linear regression with the continuous phenotype outcome and continuous version of the ERS, with R2 and the predicted residual sums of squares (PRESS) statistic measuring model fit. Second, we dichotomized the levels of the phenotypes as high vs. low (200 mg/dL for total cholesterol; 40 mg/dL (male) or 50 mg/dL (female) for HDL; 130 mg/dL for LDL; and 150 mg/dL for triglycerides [38]), and conducted logistic regression analysis with this dichotomized outcome and with continuous ERS as predictor. We used area under the receiver operating characteristic (ROC) curve or AUC to assess predictive ability of the ERS with these binary endpoints. In each of the above two metrics we compared a sequence of models, with only base covariates, base covariates + micronutrients, base covariates + micronutrients + ERS. Note that the above two metrics measure overall prediction, aggregated over all subjects. A bootstrap resampling (2000 iterations) was
