Background: Family-based designs enable assessment of genetic associations without bias from population stratification. When parents are not readily available - especially for diseases with onset later in life - the case-sibling design, where each case is matched with one or more unaffected siblings, is useful. Analysis typically accounts for within-family dependencies by using conditional logistic regression (CLR).
Methods: We consider an alternative to CLR that treats each case-sibling set as a nuclear family with both parents missing by design. One can carry out maximum likelihood analysis by using the Expectation-Maximization (EM) algorithm to account for missing parental genotypes. We compare conditional logistic regression and the missing-parents approach under several risk scenarios.
Results: We show that the missing-parents approach improves power when some families have more than one unaffected sibling and that under weak assumptions the approach permits the incorporation of supplemental cases who have no sibling available and supplemental controls whose case sibling is unavailable (e.g., due to disability or death).
Conclusion: The missing-parents approach offers both improved statistical efficiency and asymptotically unbiased estimation for genotype relative risks and genotype-by-exposure interaction parameters.