The goal of this analysis was to identify subsets of expression traits independently associated with each risk factor. To increase the robustness of the analysis the population was randomly divided into 2 sub-samples of equal size which were used for screening and validation purpose respectively. The screening step was focused on the subsets of expression traits that were associated with each covariate-adjusted risk factor in univariate analysis at P<3.9×10−6 (Bonferroni correction for 12,808 expressions). Each risk factor and corresponding subset of expression traits were included as dependent and predictor variables respectively in a forward stepwise regression model to identify expression traits that were independently associated with the risk factor (P<0.01). Gene expressions selected at the screening step were then jointly tested in the validation sample for association with the risk factor by multiple regression analysis. This screening/validation procedure was repeated 250 times and for each risk factor, expression traits associated (P<0.01) with the risk factor in more than 25% of the replicates are reported.
