Analysis for this study was broken down into two parts, distinguished by whether SNPs were selected from family-based candidate gene association studies or from case-control GWAS analyses (Fig. 1). In the first part of this study, SNPs that were previously associated with AD in candidate gene studies in the COGA high-density family-based association sample were used to create a genetic risk score to assess prediction of AD in independent individuals from the COGA and SAGE GWAS samples. The second part of the study assessed the discriminatory, or predictive, accuracy of SNP panels selected from GWAS results using varying “significance” criteria. We controlled for allele frequency and linkage disequilibrium (LD) pattern differences across ethnicities by assessing risk scores in just the EA subsets in both parts of the study. In order to select independent discovery and validation samples, individuals from the COGA GWAS EA sample independent of the COGA family-based association sample were used to assess predictive accuracy of candidate gene sum scores. The FSCD and COGEND portions of the SAGE GWAS EA sample were extracted for use as a sample
