Genotypic association analysis using discordant-relative-pairs

In practice, family-based design has been widely used in disease-gene association analysis. The major advantage of such design is that it is not subject to spurious association due to population structure such as population stratification (PS) and admixture. A disadvantage is that parental genotypes are hard to obtain if the disease is late onset for which a discordant-relative-pair design is useful. Designs of such kind include full-sib-pair, half-sib-pair, first-cousin-pair, and so on. The closer the relatedness of the pair, the less possible that they are subject to population stratification. On the other hand, the association test using close relative-pairs may be less powerful due to over-matching. Trade-off between these two factors (population structure and over-matching) is the major concern of this study. Some tests, namely McNemar's test, matched Cochran-Armitage trend tests (CATTs), matched maximum efficient robust test (MERT), and Bhapkar's test, are used for testing disease-gene association based on relative-pair designs. These tests are shown to be valid in the presence of PS but not admixture. Numerical studies show that the McNemar's test, additive CATT, MERT, and Bhapkar's test are robust in power, but none of them is uniformly more powerful than the others. In most simulations, the power of any of the tests increases as the pair is more distant. The proposed methods are applied to two real examples.