Genome-wide association studies of polygenic risk score-derived phenotypes may lead to inflated false positive rates

Abstract

In a recent study, a polygenic risk score (PRS) for Alzheimers disease was used to construct a new phenotype for a subsequent genome-wide association study (GWAS). Here we show that the applied method, in which the same genetic variants are used to construct the PRS-derived phenotype as well as to assess their effect in a GWAS of the same phenotype, leads to inflated false positive rates. We illustrate this bias by simulation. We first simulate an initial discovery cohort, and run a GWAS of a disorder like Alzheimers disease. We then simulate a target cohort, in which we construct a PRS based on the initial GWAS results. Following the published study, we select the bottom and top 5% of individuals in the PRS distribution and define them as controls and cases. Lastly, we run a GWAS on the new PRS-derived phenotype using all genetic variants. We show that at a significance threshold of 5 x 10-8, false positive rates are inflated up to 0.004 (an 80,000-fold increase compared to 5 x 10-8). We also show that such inflation can be prevented by excluding all variants that were used to construct the PRS (as well as all variants in linkage disequilibrium), when a GWAS on a PRS-derived phenotype is conducted.