Statistiscal methods for population health research on chemical mixtures

描述（由申请人提供）：接触化学混合物是常规而非例外。例子包括饮食、药剂和空气污染。在公共卫生领域，开发估算复杂混合物毒性的统计方法被认为是重要的。通过确定混合物中某些特定物质的毒性，我们将指导关于生物作用机制的假设的发展，这些假设可以在实验模型中进行测试，我们将更好地为公共政策提供信息。我们收集了有关健康结果、可吸入颗粒物（PM10）化学成分和混杂因素的国家数据集。PM10可以被看作是一种混合物，由两种PM大小（PM2:5和PM102:5）、PM2:5的50多种化学成分以及具有相同化学特征的化学成分组（见图1左侧和中间面板）所代表。基于这一科学问题，我们提出：a .1)一类具有非精确层次结构预测器的贝叶斯层次模型；A.2)在效应估计中考虑调整不确定性的创新贝叶斯方法；A.3)检验空间和时间数据回归模型的新方法，方法是将暴露效应估计分解为与不同时空变化尺度相关的分量，当模型规范正确时，这些分量应该是相同的。我们将把方法应用于国家空气污染、天气和健康数据库，以便在全国范围内估计PM混合物的毒性（a .4）。在实质性结果是复杂统计计算产物的领域，使科学发现可重复的能力日益重要。在A.5中，我们将开发用于传播数据、统计方法和科学结果的计算工具。我们的方法和相关的计算工具将推进暴露于化学混合物的人口健康研究的统计和计算方法。这里开发的方法将为处理复杂的混合物提供一个通用平台，并可应用于其他生物学问题，例如，识别由功能途径或其他先前假设组织的假定疾病基因。公共卫生相关性：接触化学混合物是普遍现象，而不是例外，饮食、药剂和空气污染是常见的例子。在本建议中，我们将开发一类新的统计模型，用于估计与接触化学混合物有关的不良健康影响。通过确定混合物中某些特定物质的毒性，我们将指导关于生物作用机制的假设的发展，这些假设可以在实验模型中进行测试，我们将更好地为公共政策提供信息。

DESCRIPTION (provided by applicant): Exposure to chemical mixtures is the rule rather than the exception. Examples include diet, pharmaceutical agents and air pollution. In public health, is deemed important to develop statistical methods for estimating the toxicity of a complex mixture. By identifying the toxicity of some specific agents in the mixture, we will guide the development of hypotheses on biological mechanisms of action that can be tested in experimental models and we will better inform public policy. We have assembled national datasets on health outcomes, chemical components of PM10 (particulate matter 10m) and confounders. PM10 can be viewed as a mixture represented by a hierarchically structured predictor characterized by two PM sizes (PM2:5 and PM102:5), more than fifty chemical constituents of PM2:5, and groups of chemical constituents that have the same chemical characteristics (see Figure 1 left and center panels). Motivated by this scientific problem, we propose to develop: A.1) a novel class of Bayesian hierarchical models with an imprecise hierarchically structured predictor; A.2) innovative Bayesian methods to account for adjustment uncertainty in effect estimation; A.3)new methods for checking regression models for data in space and time by decomposing exposure effect estimates into components associated with distinct spatio-temporal scales of variation that should be the same when the model specification is correct. We will apply methods to national databases of air pollution, weather, and health for estimating the toxicity of the PM mixture on a national scale (A.4). The ability to make scientific findings reproducible is increasingly important in areas where substantive results are the product of complex statistical computations. In A.5, we will develop computational tools for disseminating data, statistical methods, and scientific results. Our methods and associated computational tools will advance statistical and computational methods for population health studies of exposure to chemical mixtures. The methods developed here will provide a general platform for dealing with complex mixtures and can be applied to other biological questions, for example, to identify putative disease genes that are organized by functional pathways or other prior hypotheses. PUBLIC HEALTH RELEVANCE: Exposure to chemical mixtures is the rule rather than the exception, diet, pharmaceutical agents and air pollution are common examples. In this proposal we will develop a novel class of statistical models for estimating adverse health effects associated with exposure to chemical mixtures. By identifying the toxicity of some specific agents in the mixture, we will guide the development of hypotheses on biological mechanisms of action that can be tested in experimental models and we will better inform public policy.