Results did not show significant AUCs for the candidate gene sum scores, suggesting that sum scores of this limited set of SNPs are not predicting better than chance. The individual variants contributing to the sum scores did not yield significant results in the independent samples in which discriminative ability was assessed. Results from the GWAS analyses resulted in significant, albeit small, AUC estimates for p-value thresholds of 0.01 to 0.50. These results support a polygenic model involving hundreds of variants of small effect contributing to risk for AD that is consistent with previous findings on schizophrenia and bipolar disorder (Purcell et al., 2009). Less stringent thresholds allowed for the selection of more true findings with effect sizes that would not otherwise have reached genome-wide significance. Combining nominally associated SNPs in aggregate improved clinical validity because these true loci could outweigh noise from null loci.
