Delta-centralization fails to control for population stratification in genetic association studies

To investigate the validity of simulations and assumptions used to underpin the delta-centralization (DC) method for correcting for population stratification in genetic association studies; to assess the effectiveness of DC compared to genomic control (GC) under valid simulation conditions; and to highlight other studies employing similarly flawed simulations.

DC and GC use data from unlinked null loci to correct the test statistic at the target locus, but differ in the way the correction is performed. We compare DC and GC under two simulation approaches: an invalid approach adopted by the originators of DC, which permits subpopulation allele frequency matching of null markers to the target locus; and a valid approach, based on the Balding-Nichols model, which does not allow subpopulation matching.

DC works under invalid simulation conditions (subpopulation allele frequency matching), but not under our valid ones. We use theoretical arguments to assert that there are no valid conditions under which DC might work. We identify several studies which have adopted similarly invalid simulation approaches in this field.

DC fails to control for population stratification and should not be used. Other results from studies which have used the same invalid simulation approach should be treated cautiously.