Population stratification in epidemiologic studies of common genetic variants and cancer: quantification of bias

Some critics argue that bias from population stratification (the mixture of individuals from heterogeneous genetic backgrounds) undermines the credibility of epidemiologic studies designed to estimate the association between a genotype and the risk of disease. We investigated the degree of bias likely from population stratification in U.S. studies of cancer among non-Hispanic Caucasians of European origin.

An expression of the confounding risk ratio-the ratio of the effect of the genetic factor on risk of disease with and without adjustment for ethnicity-is used to measure the potential relative bias from population stratification. We first use empirical data on the frequency of the N-acetyltransferase (NAT2) slow acetylation genotype and incidence rates of male bladder cancer and female breast cancer in non-Hispanic U.S. Caucasians with ancestries from eight European countries to assess the bias in a hypothetical population-based U.S. study that does not take ethnicity into consideration. Then, we provide theoretical calculations of the bias over a large range of allele frequencies and disease rates.

Ignoring ethnicity leads to a bias of 1% or less in our empirical studies of NAT2. Furthermore, evaluation of a wide range of allele frequencies and representative ranges of cancer rates that exist across European populations shows that the risk ratio is biased by less than 10% in U.S. studies except under extreme conditions. We note that the bias decreases as the number of ethnic strata increases.

There will be only a small bias from population stratification in a well-designed case-control study of genetic factors that ignores ethnicity among non-Hispanic U.S. Caucasians of European origin. Further work is needed to estimate the effect of population stratification within other populations.