Our final analysis data set included 3,742 AAs (2,017 UHS cases and 1,725 population controls) and 6,845 EAs (1,142 UHS cases and 5,703 population controls). We previously found that combining imputed SNPs derived from cases and controls genotyped on different arrays has the potential to create artifactual differences leading to biased genotype-phenotype associations.(Johnson et al., 2013) To circumvent this potential bias, imputation was conducted using the common set of genotyped SNPs available across all EAs (344,736 SNPs) or AAs (336,412 SNPs) as the input genotypes. We have shown this approach eliminates this potential bias and retains high imputation quality. (Johnson et al., 2013)
