摘要：It has recently been alerted that adjustment of covariates in genetic association analyses using case-control data may lead to decreased power for rare phenotypes but increased power for common phenotypes. We propose a unified profile likelihood method to incorporate external phenotype prevalence data and gene-covariate independence, allowing adjustment of additional covariates. Our method guarantees that adjustment of covariates can lead to increased power for testing genetic  association, regardless of phenotype prevalence. A key theoretical novelty in our method is that we replace with their large sample limits the Lagrange multipliers involved in the maximization of the profile likelihood, which leads to both numerical stability and straightforward development of theoretical properties while maintaining the asymptotic efficiency. The proposed method can be applied to fit any commonly used penetrance model such as the logit and probit models. We show through  extensive simulation studies that the power of our proposed method is higher than the standard model-fitting methods with or without covariate adjustment, and can be considerably higher when the phenotype is common and the covariate effect is strong. We illustrate the proposed methods through analyses of a case-control genetic association study on human high density lipoprotein cholesterol level.