Methods for Epidemiology Studies

已经对使用当前和未来全基因组关联研究的数据来改进预测疾病风险的模型的性能的可能性进行了调查。发展了一种新的数学范式，根据训练数据集的样本大小、潜在易感基因座的数量及其影响大小的分布来表征多基因模型的预测性能。然后，该范例被应用于对包括癌症在内的10种不同复杂特征的风险预测模型的性能进行预测。这些预测表明，未来将需要极大的全球遗传风险模型，其样本量甚至比一些迄今最大的全球遗传风险模型的数量级还要大，以建立具有显著改进的预测性能的遗传风险模型。开发了一种新的方法，利用病例对照全基因组关联研究的数据来评估基因与环境的相互作用，该研究使用了公开可用的遗传对照。结果表明，在一系列假设下，如果从内部病例对照研究获得关于环境暴露的数据，即使这种研究中的对照没有基因分型，也可以表征这种研究的基因-环境联合影响。开发了一种新的方法来评估SNP标记与反映疾病发展不同阶段的有序性质的疾病结果的相关性。提出了两种可供选择的检验方法，即最大得分检验(Max)和自适应P值组合检验(ADAPT-P)，目的是在可能的替代模型(SNPs可能涉及不同阶段)的效率和稳健性之间取得平衡。仿真研究表明，在各种现实场景下的一系列测试中，MAX和ADAPT-P具有最稳健的性能。一种基于排列的重采样方法被开发出来，用于使用代谢组数据来检验暴露(例如吸烟)通过中间生物标记物对疾病(例如肺癌)风险的中介作用的假设。广泛的模拟研究被用来检验所提出的测试的有效性和威力。发展了以人群为基础的病例对照研究的分析方法，具有复杂的抽样设计。提出了两种结合样本权重信息的方法，即根据设计变量对样本权重的样本期望进行建模。与使用原始样本权重的标准估计器相比，这些方法具有更高的效率和更小的有限样本偏差。这些方法被用于美国肾癌病例对照研究，以确定危险因素。一个项目开发了一个线性-指数回归模型(LEXPIT)，将线性和非线性风险效应结合在一起，从二元结果的研究中估计绝对风险。LEXPIT是二项线性模型和Logistic回归模型的推广。LEXPIT线性项的系数估计调整后的风险差异，而指数非线性项估计剩余赔率比。LEXPIT对于风险关联的流行病学研究可能特别有用，在这些研究中，对多个混杂变量的调整是常见的。该方法被应用于评估在Kaiser Permanente Northern Califronia接受筛查的167,171名妇女患宫颈癌前病变或与不同巴氏试验和人类乳头瘤病毒检测结果相关的癌症的绝对五年风险。LEXPIT模型发现，由于HPV阴性的巴氏试验异常，而Logistic回归没有检测到这一点，风险增加。开发了R包BLM，以提供免费且易于使用的软件来拟合LEXPIT模型。

Investigations have been conducted for the potential for using data from current and future genome-wide association studies for improving performance of models for predicting disease risks. A new mathematical paradigm was developed to characterize predictive performance of polygenic models in terms of sample size for training datasets, number of underlying susceptibility loci and distribution of their effect-sizes. The paradigm was then applied to make projections for performance of risk prediction models for ten different complex traits, including cancers. These projections revealed that in the future extremely large GWAS, with sample size of a larger order magnitude than even some of the largest GWAS to date, would be needed for building genetic risk models with substantially improved predictive performance. A new method was developed for assessing gene-environment interactions using data from case-control genome-wide association studies that uses publicly available genetic controls. It was shown that under a set of assumptions it possible to characterize joint gene-environment effects from such studies if data on environmental exposures are available from an internal case-control study even if controls in such a study are not genotyped. New methods was developed for evaluating association of SNP markers with disease outcome of ordinal nature reflecting various stages of the progression of a disease. Two alternative tests, the maximum score test (MAX) and the adaptive P-value combination test (Adapt-P), are proposed with the aim of striking a balance between efficiency and robustness over possible alternative models by which a SNPs might be involved in the various stages. Simulation studies were used to demonstrates that MAX and Adapt-P have the most robust performance among all a range of tests under various realistic scenarios. A permutation-based resampling method was developed for using metabolomic data for testing the hypothesis of mediation of the effect of an exposure (e.g smoking) on the risk of a disease (e.g lung cancer) through intermediate biomarkers. Extensive simulation studies were used to examine validity and power of the proposed test. Methods were developed for analysis of population-based case-control studies with complex sampling designs. Two methods were developed for incorporating the information included in the sample weights by modeling the sample expectation of the weights conditional on design variables. These methods have higher efficiency and smaller finite sample bias compared with the standard estimators that use original sample weights. The methods were to the U.S. Kidney Cancer Case-Control Study to identify risk factors. A project developed a linear-expit regression model (LEXPIT) to incorporate linear and nonlinear risk effects to estimate absolute risk from studies of a binary outcome. The LEXPIT is a generalization of both the binomial linear and logistic regression models. The coefficients of the LEXPIT linear terms estimate adjusted risk differences, while the exponentiated nonlinear terms estimate residual odds ratios. The LEXPIT could be particularly useful for epidemiological studies of risk association, where adjustment for multiple confounding variables is common. The method was applied to estimate the absolute five-year risk of cervical precancer or cancer associated with different Pap and human papillomavirus test results in 167,171 women undergoing screening at Kaiser Permanente Northern Califronia. The LEXPIT model found an increased risk due to abnormal Pap test in HPV-negative that was not detected with logistic regression. An R package blm was developed to provide free and easy-to-use software for fitting the LEXPIT model.