Testing the genomic control parameter

摘要

In genome-wide case-control association studies, the genomic control factor, λ, is generally computed to estimate the inflation of chi-square statistics due to population heterogeneity. For a given estimated λ value, all chi-square values are then divided by λ before their associated significance level is computed. We hypothesize that in some studies λ may not be significantly larger than 1, in which case division by λ leads to decreased power. Thus, we treat λ as a test statistic and evaluate the p=value associated with the null hypothesis, H0: λ = 1, versus the one-sided alternative, H1: λ > 1. Our analyses of published disease data show that, owing to the large number of SNPs generally used, observed λ values are in most cases significantly larger than 1.