BOX 3-2Population Stratification (Confounding)

When the risk of disease varies between two ethnic groups, any genetic or environmental factor that also varies between the groups will appear to be related to disease. This phenomenon is called “population stratification” in epidemiologic studies investigating the effect of a genetic factor on disease, and it is a form of confounding. Population stratification refers to the presence of sub-groups—for example ethnic groups—in the sample, which could potentially cause a spurious association between genetic variations and trait. Concerns about population stratification have raised doubts about the credibility of some reported findings in candidate gene studies and have led to calls for the routine use of related controls in case-control studies of genetic factors to eliminate the possibility of population stratification (Lander and Schork, 1994; Altshuler et al., 1998). In fact, although population stratification is frequently used as an explanation for nonreplicable associations in the literature, there are few actual examples to support this assumption (Risch, 2000) and many agree that the problem probably has been overstated (Cardon and Bell, 2001). For example, Wacholder et al. (2000) argued that population stratification to an extent large enough to distort results is unlikely to occur in many realistic situations. Population stratification is a manifestation of confounding—that is, the distortion of the relationship between the exposure of interest and disease due to the effect of a true risk factor that is related to the exposure (Wacholder et al., 2000). Thus, in population stratification ethnicity acts as a surrogate for the true risk factor, which may be environmental or genetic. This means that controlling for ethnicity can reduce the confounding bias.

Ardlie et al. (2002) evaluated four moderately sized case-control studies for the presence of population structure and concluded that carefully matched case-control samples in U.S. and European populations are unlikely to contain levels of population stratification that would result in significantly inflated numbers of false positive associations. However, methods have been developed by which unlinked genetic markers can be used to detect stratification and even correct for it when it is present (Pritchard and Rosenberg, 1999; Satten et al., 2001).