Imputation results were compared across subjects genotyped on different arrays by arbitrarily assigning subjects from one originating study as cases and subjects from the other originating study as controls. Associations between SNP genotype dosage (fractional value ranging from 0 to 2.0 that corresponds to the estimated reference allele count) and the assigned case-control status were tested using a logistic regression model implemented in PLINK (Purcell et al. 2007). To eliminate any potential bias from residual population stratification, we applied EIGENSTRAT (Price et al. 2006) analysis to each set of study comparisons using a set of autosomal SNPs, which included only those having R2 < 0.2 within a 1,500 window size and omitted known regions of high LD, as implemented elsewhere (Fellay et al. 2007). The first ten principal components were included as covariates in all regression models.
