Measurement Error, Missing Data and Semiparametrics

描述（由申请人提供）：该提案反映了我们的持续 对解决测量误差、数据缺失等问题感兴趣 回归设置。提出的研究课题自然产生于两个 重要研究： (i)结肠癌肿瘤发生项目，包括一系列 在细胞水平上研究结肠癌肿瘤发生的实验，以及（ii） 前列腺癌结局研究，这是最大的多中心观察性研究 一项研究，旨在调查未受控制的 男性的真实世界环境和健康相关生活质量（HRQOL） 被诊断为前列腺癌 我们建议研究五个主要研究课题： 开发有效的广义估计方程（GEE）程序， 半参数边际回归模型 估计两个变量之间的动态相关性，例如细胞凋亡（细胞凋亡）。 死亡）和DNA加合物损伤，作为连续协变量，细胞 结肠隐窝内的位置。 开发适应新型协变量测量的方法 使用二级混合模型的预测结果作为 主模型中的协变量。 利用多种多样的多重性对不完全数据进行简化分析 插补程序，包括非参数和半参数插补。 采用近似鞍点方法对统计推断进行了研究， 适用于分析少量受试者/群集/单位的数据。 该提案的主要重点是开发高效、易于计算的 方法，而不施加不必要的参数假设。特别强调 将给予相关的观测收集纵向和 集群研究

DESCRIPTION (provided by applicant): This proposal reflects our continuing interest in solving problems of measurement error, missing data in general regression settings. The proposed research topics naturally arise from two important studies: (i) Colon Cancer Tumorigenesis Project, which consists of a series of experiments to study colon cancer tumorigenesis at the cellular level, and (ii) Prostate Cancer Outcomes Study, which is the largest multicenter observational study to investigate the link between medical practices in the uncontrolled real-world environment and health-related quality of life (HRQOL) among men diagnosed with prostate cancer. We propose to study five primary research topics: Developing efficient generalized estimating equation (GEE) procedure under non- and semi-parametric marginal regression models. Estimating a dynamic correlation between two variables, such as apoptosis (cell death) and DNA adduct damage, as a function of a continuous covariate, cell positions within a colon crypt. Developing methodologies to accommodate a new type of covariate measurement error problem caused by using predictions from a secondary mixed model as the covariate in the primary model. Accommodating analyses of incomplete data through utilizing various multiple imputation procedures including non- and semi-parametric imputations. Adopting approximate saddle point methods to statistical inferences that are suitable to analyze data from a small number of subjects/clusters/units. The major focus of the proposal is development of efficient, easily calculable methods without imposing unnecessary parametric assumptions. Special emphasis will be given to correlated observations collected from longitudinal and clustered studies.