Statistical Methods for Outcome-Dependent Sampling

描述（由申请人提供）：我们将开发和评估改进的统计方法，用于设计和分析生物医学研究，这些研究采用一般偏倚抽样设计方案，单变量和多变量结果辅助依赖抽样（OADS）和两阶段OADS设计。这种设计的优点是它同时允许前瞻性和回顾性样本，其中前瞻性样本提供队列研究的好处，而回顾性样本使研究人员能够将资源集中在信息量最大的地方，即根据结果和辅助协变量信息明智地选择一些子集。需要新的统计方法来实现潜在的统计效率。将扩展简单的ODS设计，使抽样概率取决于连续的结果和连续的辅助协变量。我们还在实践中经常遇到的预算和精度/功率限制下开发了最佳的两阶段OADS设计。将制定工具和基准，以便在研究的规划阶段区分可用的抽样选择。这些是固定精度/功率情况下的相对预算指数和固定预算情况下的相对收益指数。所提出的方法在癌症和环境研究中特别有用，因为辅助暴露信息和昂贵的暴露评估是经常面临的挑战。

DESCRIPTION (provided by applicant): We will develop and evaluate improved statistical methods for the design and analysis of biomedical studies conducted with general biased sampling design schemes, the univariate and multivariate outcome-auxiliary- dependent sampling (OADS) and the two-stage OADS designs. The advantage of such designs is that it allows both prospective and retrospective samples at the same time where the prospective sample provides the benefits of a cohort study and the retrospective sample enables investigators to concentrate resources on where there is the greatest amount of information, i.e., some judiciously chosen subsets based on the outcome and auxiliary covariate information. New statistical methods is needed to achieve the potential statistical efficiency. Extension of the simple ODS design to allow the sampling probability to depend on a continuous outcome and a continuous auxiliary covariates will be developed. We also develop optimal two-stage OADS designs under commonly encountered budget and precision/power constraints in practice. Tools and benchmark for distinguishing available sampling options in the planning stage of the study will be developed. These are the relative-budget-index for fixed precision/power case and the relative-gain-index for fixed budget case. The proposed methods are particularly useful in cancer and environmental research where auxiliary exposure information and expensive exposure assessment are frequent challenges. The proposal consists of six projects. The first project deals with semiparamtric efficient inference for two-stage OADS design where the first stage data can be either from a simple random sample or from an ODS sample itself. The second project concerns the optimal two-stage OADS design for a fixed budget and the development of a formal evaluation criteria (RGI) that measures the closeness of an alternative design to the optimal one. The third project concerns the optimal two-stage OADS design for a given precision/power and the development of a formal evaluation criteria (RBI) that measures the closeness of an alternative design to the optimal one (the one with the minimal budget). The fourth project considers a multivariate OADS and multivariate two-stage OADS design and develop the semiparametric inferences for correlated responses under the multivariate OADS. The fifth project concerns a partial linear model for the nonlinear exposure effects in both fixed and random effects regression analysis under an OADS and two-stage OADS design. The sixth project investigates a variable selection and hypothesis testing techniques for data from two-stage OADS design. The strengths and weaknesses of proposed methods will be critically examined via theoretical investigations and simulations. Cost-effective sampling strategies in a given setting will be investigated. Comparisons with existing methods will be conducted. Related software will be developed. Data sets from epidemiologic and environmental studies on the effects of environmental exposures, and on cancer and other diseases will be analyzed. These include the Cancer Risk in Uranium Miners Study, the Magnetic Fields and Breast Cancer Risk Study, the Collaborative Perinatal Project, the Family Heart Study, and the DDE-antiandrogen Study. PUBLIC HEALTH RELEVANCE: We propose and investigate some new study designs/analytical methods that will allow biomedical study to be conducted less costly in practice while still providing a good statistical power to detect the effect of interests. These designs allow investigators to conduct their study more efficiently for a given budget and hence can help improve the overall efficiency and productivity of the public health research.