If there are major genes accounting for substantial covariation in X and Y, the genetic component of U may not be constant across SNPs and the assumption of a constant regression slope b is violated. This problem could be avoided by conditioning the prognosis associations on the major genes, thereby estimating the bias through all confounders except those genes. The resulting correction is then appropriate for all SNPs including those in major genes (Fig. 3). A similar approach can be taken to polygenic scores aggregating the small effects of several individual SNPs.
