Case-control designs provide an attractive approach for measured genes. A conventional case-control study selects controls from the cases' source population. However, when estimating the effect of a candidate gene, this approach can give biased results if population stratification exists. One can attempt to reduce this potential bias by using familial-based controls instead of population controls. In this project we first describe the issue of confounding by population we present an overview of the different family design options. We then give results from a simulation study comparing the bias, relative efficiency, and power of case-control studies of measured genes among designs using sibling, cousin, and parental genotype control(the trasmission-disequilibrium test and generalizations thereof based on conditional logistic regression). The case-parent design was found to be slightly biased towards the null (a bias that decreases with disease rarity), but more efficient than the other designs. Of the latter, using cousin controls was found to be more efficient than using sibling controls and to offer some nonstatistical advantages (potentially closer matching on age, birth cohort, and birth order), although using siblings retains some advantages over using cousins (e.g., ease of locating controls).
Showing the most recent 10 out of 922 publications
