Whenever possible, it is preferable to obtain all relevant genotype data and correct for confounding biases directly34–38; post-hoc correction of test statistics is no substitute for diligent quality control. However, in the event that only summary data are available, or if a conservative correction is desired, we propose that the LD Score regression intercept provides a more robust quantification of the extent of inflation from confounding bias than λGC (or intergenic λGC, Supplementary Table 8). Since λGCincreases with sample size in the presence of polygenicity (even without confounding bias)3, the gain in power obtained by correcting test statistics with the LD Score regression intercept instead of λGC will become even more substantial for larger GWAS. Extending this method to non-European populations such as East Asians or West Africans is straightforward given appropriate reference panels, but extension to admixed populations is the subject of future research.
