Abstract:
|
Genetic case-control association studies have been widely-used for many human complex diseases, where Cochran-Armitage trend test with 1 degree-of-freedom under the assumption of an additive model is a standard practice to detect the association of the disease with candidate alleles. However, we show that the commonly-used additive score is biased when the case proportion is not equal to the population disease prevalence, which almost always holds for case-control study. In this study we first examine the statistical source that leads to this bias, and then develop an unbiased adjusted additive score that successfully maintains the additive model characteristics. We also prove that, the unbiased additive score is more close to a dominant model for common case-control studies where case proportions are higher. We conduct simulation studies to confirm above arguments. We further investigate the effect on power of the misuse of the biased additive score under different situations.
|