Population Structure in Genetic Association Studies

Standard genetic association tests using case-control data are based on certain assumptions about the population from which study subjects were sampled. Two types of departure from these assumptions have been studied: population stratification and cryptic relatedness. Both types of departure have been called population structure. Each can lead to erroneous inferences due to differences between a test statistic’s actual null distribution and the nominal one valid only for populations without structure. The differences can reflect either confounding bias or variance distortion. For each type of structure, adjusted test statistics have been proposed whose actual null distributions, in the presence of the structure, equal the nominal ones appropriate for unstructured populations. This paper reviews models for population stratificationand cryptic relatedness, and uses them to examine the effects of each on the Armitage trend test for case-control data. Specifically, population stratification can cause confounding bias but not variance distortion, while cryptic relatedness can cause variance distortion but not confounding bias. Consequently the adjusted statistics developed for population stratification (e.g. the latent variable methods of Pritchard et al. (1999, 2001); Satten et al. (2001); Schork et al. (2001); Wang et al. (2005)), address potential confounding bias but not variance distortion. Conversely, the adjusted statistics developed for cryptic relatedness (e.g. the Genomic Control (GC) methods of Devlin and Roeder (1999), Setakis et al. (2006) and Zheng et al. (2006)) address variance distortion but not confounding bias. These differences may explain the anomalous behavior of adjusted statistics when applied to populations with structure of a type that differs from the one for which the method was designed. They indicate that care is needed to specify the nature of the underlying structure anticipated for a given population, and to use appropriate methods to adjust for it.