Semiparametric inference for case-control studies with complex sampling

概述(申请人提供)：复杂抽样的基于人群的病例对照研究，例如分层多阶段整群抽样，现在越来越多地被用于研究遗传变异和基因-环境(G-E)在人类疾病病因学中的作用。当通过简单的随机抽样选择病例和对照时，已经开发了基于回顾的Logistic回归估计器，以利用各种协变量分布假设来在此类研究中获得效率。然而，当用复杂的抽样选择病例或对照时，这些方法可能会导致无效的推断。虽然大多数基于单核苷酸多态性(SNP)的复杂抽样关联研究考虑了复杂设计引起的并发症，但许多基于单倍型的遗传关联研究倾向于在估计单倍型频率和/或回归系数时忽略它们。在这个项目中，我们将开发统计方法，以考虑复杂样本设计中基于单倍型的关联研究的设计复杂性。具体地说，被回溯法的效率优势所吸引，我们将探索Hardy-Weinberg均衡和G-E独立性的假设，并开发适用于具有复杂样本设计的病例对照研究的有效估计量。另一方面，当这些假设失败时，使用上述假设的分析可能会产生误导。因此，上述假设将通过提出经验贝叶斯型收缩估计来放松，作为偏差和效率之间的权衡。建议的方法将通过各种复杂样本设计下的模拟以及两个以总体为基础的病例对照研究进行评估。此外，还将开发一个统一的软件包，以广泛传播研究成果。

DESCRIPTION (provided by applicant): SUMMARY Population-based case-control studies with complex sampling, e.g. stratified multistage cluster sampling, are now increasingly used to study the role of genetic variants and gene-environment (G-E) interplay in the etiology of human diseases. Retrospective- based logistic regression estimators have been developed to exploit various covariate- distributional assumptions to gain efficiency in such studies when cases and controls are selected with simple random sampling. These methods, however, can lead to invalid inferences when cases or controls are selected with complex sampling. Although most single nucleotide polymorphism (SNP)-based association studies with complex sampling account for the complications induced by complex designs, many of haplotype-based genetic association studies with complex sampling tend to ignore them in the estimation of haplotype frequencies, regression coefficients or both. In this project, we will develop statistical methods for taking into account the design complications in haplotype-based association studies with complex sample designs. Specifically, attracted by the efficiency advantage of the retrospective method, we will explore the assumptions of Hardy- Weinberg equilibrium and G-E independence, and develop an efficient estimator suitable for the case-control study with a complex sample design. On the other hand, analysis with above assumptions can be misleading when these assumptions fail. Thus, the above assumptions will be relaxed by proposing an empirical Bayes-type shrinkage estimator as a trade-off between bias and efficiency. The proposed methods will be evaluated using simulations under various complex sample designs as well as two population-based case-control studies. Furthermore, a unified software package will be developed to widely disseminate the research outcomes.