In applying these ideas to breast and prostate cancers, Machiela et al did not find significant associations of polygenic scores [14]. While this could be explained by the genetic architecture of the diseases, a possible explanation (noted by the authors) is the lower sample size together with the low heritability. Their breast cancer study used a total sample of 2287 subjects, approximately half of which were cases and half controls, which was split into training and testing subsets in a 9∶1 ratio for 10-fold cross-validation. The marker panel consisted of 161,702 nearly independent SNPs. Assuming a prevalence of 3.6% and sibling relative risk of 2.5 [29], this design has only 17% power to detect an association of the polygenic score, even if the markers explain the full heritability. If the sample were split in a 1∶1 ratio, the power would increase to 37%.
