Here, we make use of a related statistical technique to adjust for population structure with an expanded control cohort. Axes of genetic variation are defined by application of classical multidimensional scaling (MDS) techniques [Cox and Cox, 1994] to a matrix of identity by state (IBS) values between all pairs of samples in the study (cases and the expanded control cohort), using genome-wide genotype data. Such an approach has been used to identify population outliers in the WTCCC [The Wellcome Trust Case Control Consortium, 2007] and for SNP selection and subsequent visualization of population structure [Miclaus et al., 2009]. In a logistic regression framework, we identify which of the resulting axes of genetic variation are associated with disease. These are then treated as covariates in the logistic regression model, providing a basis for testing for association with disease, adjusted for the effects of underlying population structure. We present a simulation study to investigate the false-positive error rate of this procedure, and provide evidence that we can correct for substantial population structure between cases and controls from the original study and the
