Using GWAS Data for Enhanced Mendelian Randomization Studies

描述（由申请方提供）：本提案解决了使用观察数据对治疗效果进行因果推断的困难。这一领域的方法学进展对确定最有希望的干预杠杆点的努力产生了巨大的影响，从而优化了有关影响行为，生物标志物或心理社会风险因素的一系列干预或治疗的临床建议。我们的建议发展了孟德尔随机化（MR）设计的先前工作，这是使用遗传变异作为工具的工具变量的特殊情况。我们通过利用全基因组关联研究（GWAS）已经收集的数据来扩展和加强MR应用。在GWAS数据中嵌入MR研究将导致更强大的研究和对这些研究有效性的更好评估，从而导致更可信的MR效应估计。我们建议利用GWAS数据，在MR研究中进行四项重要创新。首先，GWAS允许对背景遗传特征进行全面表征，以控制由于群体分层而导致的可能偏倚。第二，关于多个候选遗传工具的信息可用于创建多基因风险评分。这些将提供比单基因更强大的工具，提高统计能力和机会，以批判性地评估有效MR研究的假设。第三，我们可以利用多个工具，对每个工具的有效性进行过度识别测试。过度识别测试是一个标准的计量经济学工具，用于评估工具变量，但他们迄今尚未被应用于MR研究。最后，我们展示了一种新的方法来评估假设MR使用基因-环境和基因-基因相互作用。我们使用当前关于心理社会表型（抑郁和焦虑症状以及社会融合）对糖尿病和冠心病影响的研究作为例子。心理社会对这两种疾病的影响在观察性研究中得到了很好的支持，但以前对心理社会风险因素进行干预的随机试验结果令人失望。因此，这是一个理想的领域，在其中实施新的方法来分析观测数据。我们的分析是基于先前收集的两个队列的GWAS数据，护士健康研究和卫生专业人员随访研究。我们已经确定了许多可能影响每种心理社会表型的候选基因，我们还将利用GWAS研究的优势来计算多基因或多基因风险评分，以用作MR研究中的工具。MR可以是一个非常强大的工具来估计因果关系的影响，这种方法的应用已经非常迅速地增加。然而，MR依赖于很少经过严格检验的强有力的假设;部分原因是评估这些假设的工具还没有。我们的工作旨在解决这一差距，并促进明智的应用MR提供可靠的效果估计的观测数据。 公共卫生相关性：孟德尔随机化是一种方法，可以提供新的机会，从观察数据中获得可信的效果估计。该方法使用影响暴露于特定生物标志物、行为或其他类型的风险因素的基因型。这种基因型的变异可能提供自然实验来估计风险因素对随后健康结果的影响。我们使用新的工具来加强这项研究设计，并评估这项设计的效应估计值是否无偏。我们的方法利用了许多流行病学样本在收集全基因组关联研究数据方面已经进行的大量投资。为了说明这些方法，我们着重于估计心理社会困扰和社会融合对冠心病和糖尿病的影响。

DESCRIPTION (provided by applicant): This proposal addresses the difficulty of using observational data to drawing causal inferences on treatment effects. Methodological advances in this area have enormous implications for efforts to identify the most promising leverage points for interventions and thereby optimize clinical advice regarding a range of interventions or treatments affecting behavioral, biomarker, or psychosocial risk factors. Our proposal develops prior work on Mendelian Randomization (MR) designs, which are a special case of instrumental variables that use genetic variants as instruments. We extend and strengthen MR applications by taking advantage of data already collected in the context of Genome Wide Association Studies (GWAS). Embedding MR studies in GWAS data will result in more powerful studies and better assessments of the validity of those studies, leading to more credible MR effect estimates. We propose capitalizing on the GWAS data to allow four important innovations in MR studies. First, GWAS allows comprehensive characterization of background genetic characteristics to control for possible bias due to population stratification. Second, information on multiple candidate genetic instruments can be used to create multi-gene risk scores. These will provide far stronger instruments than single genes alone, improving both the statistical power and the opportunities to critically evaluate the assumptions for a valid MR study. Third, we can take advantage of the multiple instruments to conduct over-identification tests for the validity of each instrument. Over-identification tests are a standard econometric tool for evaluating instrumental variables but they have to date not been applied in MR studies. Finally, we demonstrate a novel approach to evaluating the assumptions of MR using gene- environment and gene-gene interactions. We use as an example current research on the effects of psychosocial phenotypes (symptoms of depression and anxiety and social integration) on diabetes and coronary heart disease. Psychosocial effects on both diseases are well-supported in observational studies, but previous randomized trials intervening on psychosocial risk factors had disappointing results. Thus, this is an ideal area in which to implement new approaches to analyzing observational data. Our analyses are based on previously collected GWAS data in two cohorts, the Nurses' Health Study and the Health Professionals Follow-up Study. We have identified a number of candidate genes likely to influence each of the psychosocial phenotypes and we will also take advantage of GWAS studies for the phenotypes of interest to calculate multi- gene or polygenic risk scores to use as instruments in MR studies. MR can be an extremely powerful tool to estimate causal effects, and applications of this approach have increased very rapidly. However, MR rests on strong assumptions that are rarely critically tested; in part this is because tools to evaluate these assumptions have not been available. Our work is intended to address this gap and foster judicious applications of MR to provide credible effect estimates in observational data. PUBLIC HEALTH RELEVANCE: Mendelian Randomization is a method that may offer new opportunities to derive credible effect estimates from observational data. This method uses genotypes that influence exposure to a particular biomarker, behavior, or other type of risk factor. Variations in this genotype may provide natural experiments to estimate the effect of the risk factor on subsequent health outcomes. We use new tools to make this study design stronger and to evaluate whether the effect estimates from this design are unbiased. Our methods take advantage of large investments already made by many epidemiologic samples in collecting data for genome wide association studies. To illustrate the methods, we focus on estimating the effects of psychosocial distress and social integration on coronary heart disease and diabetes.