Flexible Statistical Methods for Complex Survival Data in Biomedical Studies

描述（由申请人提供）：本研究的广泛、长期目标是开发新的统计方法，用于分析流行病学研究和临床试验的生存数据。在生存数据分析的统计建模和推断方面已经取得了重大进展;然而，新的研究设计、先进的技术以及医学研究日益增长的规模和复杂性仍然存在许多悬而未决的问题和新出现的挑战。在这项研究中，我们将探讨两个一般类的半参数模型，转换模型和加速失效时间模型，用于分析复杂的生存数据。这些模型不仅是对考克斯比例风险模型的补充，而且还提供了一般的回归框架和可能更好的生存数据建模策略。因此，它们通过提供全面的生存分析在许多生物医学应用中发挥着重要作用。我们寻求开发统计上合理的方法，不仅正确利用数据信息和结构，而且功能强大，计算效率高。 受纽约大学妇女健康研究（NYUWHS）和砷对健康影响纵向研究（HEALS）研究人员合作工作中出现的问题的激励，我们的方法学发展包括以下四个具体目标：（1）探讨嵌套病例对照研究中的线性转换模型;（2）通过统一的基于似然的方法，研究加速失效时间（AFT）模型在病例队列（CC）和巢式病例对照研究中的有效估计;（3）为AFT治愈模型开发半参数贝叶斯推断方法，用于分析来自具有易感和非易感（治愈）受试者的混合人群中的队列研究或临床试验的生存数据;（4）研究队列研究或临床试验中删失生存数据的部分线性回归建模及相关推断方法。 拟议项目的结果将与许多生物医学研究相关并适用。在所有的具体目标中，我们将研究所提出的估计的理论性质，并开发可靠的数值算法来实现所提出的估计方法。还将作出特别努力，为从业人员开发和传播软件。我们将进行广泛的模拟研究，以评估相关的理论和有限样本性能的估计。我们还将研究所提出的方法在已发表数据集上的性能，将其与现有方法进行比较，并展示其在主要临床和流行病学研究中的应用，包括NYUWHS和HEALS。1 公共卫生相关性：拟议的研究旨在开发新的统计方法，用于分析各种研究设计下的生存数据，混合人群，以及复杂的协变量效应。我们所提出的研究的完成将为临床和流行病学研究中经常遇到的复杂生存数据提供可靠和有效的统计方法。这些方法可以促进科学家对复杂疾病病因学的理解，并最终导致更好地设计疾病预防，预后和治疗策略，以改善人类健康。1

DESCRIPTION (provided by applicant): The broad, long-term objectives of this research are the developments of new statistical methodology for the analysis of survival data from both epidemiological studies and clinical trials. Significant progress has been made in statistical modeling and inference in survival data analysis; however, there are still many open questions and emerging challenges posed by new study designs, advanced technologies, as well as the growing scale and complexity of medical studies. In this proposed research, we will explore two general classes of semiparametric models, the transformation model and the accelerated failure time model, for analyzing complex survival data. These models not only are complements to Cox's proportional hazards model, but also provide general regression frameworks and possibly better strategies for modeling survival data. Thus, they play important roles in many biomedical applications by offering comprehensive survival analysis. We seek to develop statistically sound methods that not only make proper use of data information and structure but also are powerful and computationally efficient. Motivated by problems arising from the investigators' collaborative work on the New York University Women's Health Study (NYUWHS) and the Health Effects of Arsenic Longitudinal Study (HEALS), our methodology developments include the following four specific aims: (1.) To explore a broad class of linear transformation models in nested case-control (NCC) studies; (2.) To investigate efficient estimation of the accelerated failure time (AFT) model in case-cohort (CC) and nested case-control studies through a unified likelihood-based approach; (3.) To develop semiparametric Bayesian inference methods for the AFT cure model for the analysis of survival data from cohort studies or clinical trials in an admixture population with susceptible and non-susceptible (cured) subjects; (4.) To study partially linear regression modeling and the associated inference procedures for censored survival data from cohort studies or clinical trials. Results from the proposed project will be relevant and applicable to many biomedical studies. In all the specific aims, we will study the theoretical properties of the proposed estimators, and develop reliable numerical algorithms for implementing the proposed estimation methods. Special effort will also be devoted to developing and disseminating software for practitioners. We will carry out extensive simulation studies to evaluate relevance of the theory and the finite sample performance of the proposed estimators. We will also investigate the performance of the proposed methods on published datasets, compare them with existing approaches and demonstrate their applications in major clinical and epidemiological studies, including the NYUWHS and the HEALS. 1 PUBLIC HEALTH RELEVANCE: The proposed research aims to develop novel statistical approaches for analyzing survival data under various study designs, from admixed populations, and with complex covariates effects. The completion of our proposed research will provide reliable and efficient statistical methods for complex survival data that are commonly encountered in clinical and epidemiological studies. These methods can facilitate scientists' understanding of etiology of complex diseases and eventually lead to better design of disease prevention, prognosis and treatment strategies to improve human health. 1