The challenge we address here is that of expanding the control group to include genotyped individuals from a variety of studies that may not have been ascertained from the same population, and thus may not be genetically matched to the primary “within-study” cases and controls. Inappropriate genetic matching of cases and controls in the presence of population structure can lead to inflation in the false-positive (i.e. type I) error rate, unless properly accounted for in the analysis. A variety of statistical methods exist for the detection of and adjustment for population structure in GWA studies [Devlin and Roeder, 1999; Patterson et al., 2006; Price et al., 2006; Pritchard et al., 2000]. Principal components analysis (PCA) was originally applied to genetic data to infer worldwide axes of human genetic variation from allele frequency differences between populations [Cavalli-Sforza et al., 1993; Menozzi et al., 1978]. The EIGENSTRAT method makes use of axes of genetic variation, estimated from genome-wide genotype data, to continuously adjust the genotypes and phenotypes by amounts attributable to ancestry along each of these axes [Patterson et al., 2006]. By
