on alcoholism risk in Caucasians and African-Americans. However, it is fairly uncommon in non-Asian samples (< 5%) and is poorly captured by content on commercially available GWAS platforms, due to lack of LD with neighboring SNPs. We assessed the discriminatory accuracy of this ADH1B SNP for AD and found that it alone has an AUC of 0.538 (p = 7.58 × 10−4) in COGA. The inclusion of this SNP in the candidate gene sum score increased the AUC from 0.498 to 0.503, but this AUC was not significant (p = 0.885), presumably due to the very low allele frequency in this population. This suggests that including known variants that replicate in the validation sample used for prediction could have a greater AUC. Expanding the panel to include additional replicated variants could increase the AUC further.
