Genetic model testing and statistical power in population-based association studies of quantitative traits

The optimal method for considering different genetic models in association studies is not clear. We compared analytical strategies that use different genetic models to analyze genotype-phenotype information from association studies of quantitative traits in unrelated individuals. We created simulated datasets where the minor alleles are causal with an additive, dominant, or recessive mode of inheritance over a range of allele frequencies. We then computed power to detect these causal alleles using one or a combination of statistical models in a standard regression framework, including corrections for the multiple testing incurred by analyzing multiple models. Our results show that, as expected, maximal power is achieved when we test a single genetic model that matches the actual underlying mode of inheritance of the causal allele. When the inheritance pattern of the causal allele is unknown, the co-dominant model, a single two degrees of freedom test, has good overall performance in any of the three simple modes of inheritance simulated. Alternatively, it is slightly more powerful to analyze all three genetic models together (additive, dominant, and recessive), but only if the significance thresholds used to correct for analyzing multiple models are appropriately determined (such as by permutation). Finally, a commonly employed approach, testing the additive model alone, performs poorly for recessive causal alleles when the minor allele frequency is not close to 50%. Our observations were confirmed by analyzing an existing genetic association dataset in which we detect the effect of a KCNJ11 variant on insulinogenic index in unrelated non-diabetic individuals.