We used an arbitrary significance level of 0.01 to account for false positive rate. One reason is that we wanted to allow environmental pollutants that had even modest associations to be included in the ERS. We conducted sensitivity analyses using significance levels of 0.05 and 0.001 and applied these different thresholds to the AUC as shown in Table 3. Under the significance level of 0.05, 30 pollutants (vs. 13 under the significance of 0.01) for total cholesterol; 16 (vs. 9) for HDL; 5 (vs. 5) for LDL; and 34 (vs. 27) for triglycerides were identified. However, the improvement in the AUC and OR were minimal. Using a significance level of 0.001, the number of pollutants identified decreased substantially, especially for LDL. The decrease in AUC was mainly for LDL while the decrease in OR was found for all phenotypes. Therefore we chose the intermediate threshold of 0.01. Even higher significance levels (e.g., alpha of 0.1) have been used as “pruning criteria” in genetic risk scores [57], [58], therefore, genetic markers conferring only modest levels of disease risk could be aggregated
