Statistical Methods for Outcome-Dependent Sampling

描述（由申请人提供）：本基金申请的主要目的是开发和评估改进的统计方法，用于在随机效应模型和非线性协变量效应下采用结果依赖抽样（ODS）方案进行的流行病学研究的设计和分析。将开发简单ODS设计的扩展，以通过允许采样概率取决于结果和辅助协变量来进一步改进研究。抽样策略，导致一个更具成本效益的设计在给定的设置将进行调查。所提出的方法对癌症和环境研究特别有用，因为辅助信息和昂贵的暴露评估是经常面临的挑战。该提案包括四个项目：第一个项目涉及具有连续成果的ODS设计的线性混合模型回归分析。两种新的方法进行推断的回归参数的随机效应模型将被研究：半参数的经验似然方法观察ODS样本，和伪似然方法进一步提高效率时，结果变量的值是已知的基础队列。第二个项目涉及一个有效的ODS设计的抽样概率取决于一个连续的结果和辅助协变量。再次，提出了两种新的方法来处理不同的可用数据结构。第三个项目涉及依赖于结果和辅助协变量的ODS的广义线性混合效应模型分析。最后一个专题研究ODS设计下逻辑回归模型中非线性协变量效应的非参数建模的新统计方法。提出了两种方法：非参数局部经验似然法用于推断通过ODS设计收集数据时的非线性协变量效应，局部经验似然法用于ODS设计依赖于结果和离散辅助协变量的更一般设置。还考虑了处理连续辅助协变量的另一种方法。 每种方法的优点和缺点将通过理论研究和模拟研究进行严格审查。将与现有方法进行比较。将开发相关软件。将分析正在进行的关于环境暴露影响以及对癌症和其他疾病影响的流行病学研究的数据集。

DESCRIPTION (provided by applicant): The main objective of this grant application is to develop and evaluate improved statistical methods for the design and analysis of epidemiologic studies conducted with outcome-dependent sampling (ODS) schemes under random effects models and nonlinear covariates effects. Extension of the simple ODS design to further improve the study by allowing the sampling probability to depend on the outcome and auxiliary covariates will be developed. Sampling strategies that lead to a more cost-effective design in a given setting will be investigated. The proposed methods are particularly useful for cancer and environmental research because auxiliary information and expensive exposure assessment are frequent challenges. The proposal consists of four projects: The first project deals with linear mixed model regression analysis for an ODS design with a continuous outcome. Two new methods for making inferences about regression parameters in random effects models will be studied: a semi-parametric empirical likelihood approach for observed ODS sample, and a pseudo-likelihood approach for further improving efficiency when the values of the outcome variable are known for the underlying cohort. The second project concerns an efficient ODS design where the sampling probability depends on a continuous outcome and auxiliary covariates. Again, two new methods dealing with different available data structure are proposed. The third project deals with generalized linear mixed effects model analysis for an ODS that depends on both outcome and auxiliary covariates. The last project investigates new statistical methods for nonparametric modeling of nonlinear covariates effects in the logistic regression model under ODS designs. Two methods are proposed: a nonparametric local empirical likelihood method for inference about nonlinear covariates effects when data is collected via ODS designs and a local empirical likelihood method for more general settings where the ODS design depends on outcome and discrete auxiliary covariates. Another method for dealing with continuous auxiliary covariates is also considered. The strengths and weaknesses of each proposed method will be critically examined via theoretical investigations and simulation studies. Comparisons with existing methods will be conducted. Related software will be developed. Data sets from ongoing epidemiologic studies of the effects of environmental exposures, and on cancer and other diseases will be analyzed.