Several criteria were used to select SNPs for the genetic sum score. An initial list of 114 SNPs across 21 genes was generated based on prior association with AD (Table 1). SNPs associated only with early onset AD were not included in the list so that SNPs in the candidate gene panel would be applicable to the wide range of ages of individuals in the COGA and SAGE validation samples. Because assessment of clinical validity was to be performed in EA individuals, SNPs that were associated only in the AA subset were removed from the list. Forty-two of the SNPs showing association in the original papers (Table 1) were present on the Illumina Human 1M DNA Analysis BeadChip. Because we wanted to include SNPs that were captured on the current GWAS arrays, we used proxy SNPs for SNPs that were not genotyped on the arrays rather than use imputed SNPs or remove the SNPs altogether. Proxy SNPs on the Illumina chip with an r2 > 0.70 were found for 32 additional SNPs based on LD calculations in the HapMap CEU
