Statistical methods for GxE interactions: Measurement error & time varying exposu

描述（由申请人提供）：该项目的主要重点是开发用于检测和估计复杂疾病的基因-环境（GE）相互作用的新方法。遗传学研究的最新进展已经成功地完全确定了与癌症、心脏病等复杂疾病相关的遗传变异。为了进一步了解疾病的病因，研究遗传和环境危险因素之间的相互作用是重要的。研究基因-环境相互作用的一个重要挑战来自环境暴露评估的困难。大多数环境风险因素（如饮食、体力活动和空气污染）的测量不精确，自我报告的饮食或体力活动可能存在严重的系统偏差。许多环境暴露是随时间变化的，它们对健康结果的影响可能相当复杂。现有的统计方法，处理这些类型的复杂的环境评估主要集中在主效应，并没有开发的G-E相互作用。考虑到这些实际挑战，我们的目标是开发统计方法，解释GE相互作用的测量误差和随时间变化的暴露。在存在环境测量误差的情况下，我们将首先评估忽略测量误差的原始测试的有效性。然后，我们将回归校准方法扩展到具有经典测量误差的暴露和具有系统偏差的测量的交互模型。后者的典型例子包括饮食和身体活动自我报告评估。预计拟议的校准分析在测试GE相互作用方面更强大。对于时变的环境因素，如空气污染，我们提出了新的功能数据分析方法，允许灵活的建模环境的主效应和G E的相互作用。功能模型框架利用曝光的时间模式，可以潜在地提高检测G E相互作用的能力。拟议的方法学研究是由大规模流行病学研究的科学问题（例如，妇女健康倡议），并将直接适用于这些项目。

DESCRIPTION (provided by applicant): The major focus of this project is the development of novel methodologies for the detection and estimation of gene-environment (G E) interactions for complex diseases. Recent advances in genetic studies have success- fully identified genetic variants that are associated with complex diseases such as cancer, heart disease and others. To further understand disease etiology, it is important to study the interplay between genetic and environ- mental risk factors. An important challenge to studying gene-environment interactions comes from the difficulty in environmental exposure assessments. Most environmental risk factors, such as diet, physical activity and air pollution, are measured imprecisely and self-reported diet or physical activity may suffer from substantial sys- tematic bias. Many environmental exposures are time-varying and their effects on health outcomes can be rather complicated. Existing statistical methods that deal with these types of complex environmental assessments have focused on main effects, and little has been developed for G E interactions. With these practical challenges in mind, our goal is to develop statistical methodologies that account for mea- surement error and time-varying exposures for GE interactions. In the presence of environmental measurement error, we will first evaluate the validity of na¿1ve tests that ignore measurement error. We then extend regression calibration methods to interaction models for both exposures with classical measurement error and measure- ments subject to systematic bias. Typical examples for the latter include diet and physical activity self-report assessments. The proposed calibrated analyses are expected to be more powerful for testing GE interactions. For time-varying environmental factors such as air pollution, we propose novel functional data analysis methods that allow flexible modeling of environmental main effect and G E interactions. The functional model framework utilizes temporal patterns of exposures and can potentially improve power to detect G E interactions. The proposed methodological research is motivated by scientific problems from large-scale epidemiological studies (e.g., the Women's Heath Initiative) and will be directly applied to these projects.