Community-Based Studies in Cancer and Environment

这是一个更新的应用方法研究在统计问题相关的社区或观察性研究癌症和环境流行病学。针对多变量生存数据，提出了一类一般的半参数转换模型，提供了一个统一的似然框架，使回归系数具有总体水平的解释，并方便地对依赖参数进行建模。具体而言，PI计划开发：(1)多变量生存数据的半参数正态变换Cox模型：将开发成对依赖的灵活建模，并提出一类新的边缘化脆弱性模型。研究半参数极大似然估计方法。我们将进一步考虑广义半参数正态变换生存模型，包括比例风险模型和比例几率模型作为特殊情况；(2)多变量生存数据的加速半参数正态变换模型：通过探索失效时间的平均结构，提出一种新的半参数正态变换AFT模型。我们将开发一致的和数值稳定的推理程序，并调查他们的大样本结果；(3)具有不可忽略治愈分数的多元生存模型：提出了半参数脆弱治愈模型和半参数正态转换治愈模型两类新模型。我们将探讨这两类模型的理论性质，并研究可行的推理程序。将考虑空间设置的延伸；(4)具有协变量测量误差和延迟输入的聚类循环事件模型：将提出两类新的循环事件模型，以解释聚类内相关性、误测协变量和延迟输入。模型的性质将被调查，并提出各种推理程序，经验数据分析将在所有具体目标中发挥核心作用。现有数据来自PI参与的几个癌症和环境健康项目，即nci资助的癌症护理结果研究和监测联盟（CanCORS）、青年工人戒烟干预研究（一项随机社区癌症预防试验）、NIH SEER癌症发病率数据库、西肯尼亚疟疾研究和家庭过敏原研究。虽然本研究的动机是应用问题，但在生存分析、多元统计分析、治愈模型和测量误差建模等方面对一般统计理论和方法做出了有益的贡献。

This is a renewal application for methodological research in statistical issues related to community-based or observational studies in cancer and environmental epidemiology. A general class of semiparametric transformation models is proposed for multivariate survival data, providing a unified likelihood framework that allows the regression coefficients to have population-level interpretations and the dependence parameters to be conveniently modeled. Specifically, the PI plans to develop: (1) Semiparametric normal transformation Cox models for multivariate survival data: Flexible modeling of pairwise dependence will be developed and new class of marginalized frailty models will be proposed. Semiparametric maximum likelihood estimation approach will be investigated. We will further consider generalized semiparametric normal transformation survival models, encompassing the proportional hazards and the proportional odds models as special cases; (2) Accelerated semiparametric normal transformation models for multivariate survival data: We will propose a new semiparametric normal transformation AFT model by exploring the mean structure of the failure times. We will develop consistent and numerically stable inference procedures, and investigate their large sample results; (3) Multivariate survival models with non negligible cure fractions: Two new classes of models, semiparametric frailty cure models and semiparametric normal transformation cure models will be proposed. Theoretical properties for both classes of models will be explored and feasible inference procedures will be investigated. The extension to spatial setting will be considered; (4) Clustered recurrent event models with covariate measurement errors and delayed entry: two new classes of recurrent event models will be proposed to account for within-cluster correlations, mismeasured covariates and delayed entry. Properties of the models will be investigated and various inference procedures will be proposed Empirical data analyses will playa central role in all the specific aims. Available data stem from several cancer and environmental health projects that the PI has been involved in, namely, the NCI-funded Cancer Care Outcomes Research and Surveillance Consortium (CanCORS), the Young Worker Smoking Cessation Intervention study (a randomized community-based cance prevention trial), the NIH SEER cancer incidence database, the West Kenya Malaria Study and the Home Allergen Study. Although motivated by applied problems, the proposed research offers useful contributions to general statistical theory and methodology in survival analysis, multivariate statistical analysis, cure models and measurement error modeling.