In case-control studies, hidden population stratification can lead to markers that are associated with disease in the population but are unlinked to the disease loci. Such associations are often referred to as 'spurious'. A common approach to overcoming the problem of hidden population stratification is to condition on the parental genotype of the case, effectively comparing the diseased offspring's genotype to those genotypes that could have occurred but did not. If the marker is associated and linked with the disease loci the probability of offspring genotype, given parental genotype, will be 'distorted' from that expected under Mendelian inheritance. In the complete data case, tests based on this 'distortion' are insensitive to hidden population stratification. In practice however, the attractiveness of this approach is compromised by the fact that parental genotypes are often unobserved. Previous approaches to this problem have focused on recovering information from incomplete offspring-parent triads and haven't considered potential biases in tests due to the missingness distribution of parental genotype. In this talk we present an approach to this problem that explicitly considers this missingness distribution. In particular we consider the case where the missingness of parental genotype can depend on the underlying unobserved genotype of the parent, i.e., the so-called informatively missingness case. Simulations confirm the potential for large biases in existing procedures in the presence of informatively missing parental genotype.(This is joint work with Glen Satten.)
Return to Biostatistics Working Group