Family-based association studies.

We review case-control designs for studying gene associations in which relatives of case patients are used as control subjects. These designs have the advantage that they avoid the problem of population stratification that can lead to spurious associations with noncausal genes. We focus on designs that use sibling, cousin, or pseudosibling controls, the latter formed as the set of genotypes not transmitted to the case from his or her parents. We describe a common conditional likelihood framework for use in analyzing data from any of these designs and review what is known about the validity of the various design and analysis combinations for estimating the genetic relative risk. We also present comparisons of efficiency for each of the family-based designs relative to the standard population-control design in which unrelated controls are selected from the source population of cases. Because of overmatching on genotype, the use of sibling controls leads to estimates of genetic relative risk that are approximately half as efficient as those obtained with the use of population controls, while relative efficiency for cousin controls is approximately 90%. However, we find that, for a rare gene, the sibling-control design can lead to improved efficiency for estimating a G x E interaction effect. We also review some restricted designs that can substantially improve efficiency, e.g., restriction of the sample to case-sibling pairs with an affected parent. We conclude that family-based case-control studies are an attractive alternative to population-based case-control designs using unrelated control subjects.