Bayesian integration of biomarkers and spatial exposure data

描述(申请人提供)：许多环境污染物在空间和时间上都有明显的模式。在这些污染物对健康影响的流行病学研究中，主要挑战之一是准确地描述这些模式，以便根据在任何特定时间点或在给定时间段内累积接触污染物的量对研究参与者进行排名。我们提出了新的贝叶斯统计模型，以结合来自各种来源的空间和时间暴露信息，包括源排放、命运和运输模型、暴露问卷和生化测量，如参与者血液样本中的污染物浓度。贝叶斯方法还对每个参与者暴露的量的不确定性程度进行了定量估计。这种不确定性的特征对于理解将污染物暴露与有害健康影响联系起来的空间流行病学分析的可靠性至关重要。我们建议使用C8健康项目的数据进一步开发和实施这些新的贝叶斯模型，C8健康项目最近对超过69,030名接触全氟辛酸盐的社区居民进行了研究，这些居民是从西弗吉尼亚州的一个主要生产设施释放出来的。以前的分析已经将这些研究参与者中的全氟辛酸暴露与妊娠高血压/先兆子痫、睾丸癌和肾癌联系在一起；许多其他流行病学分析目前正在进行中。这些分析都依赖于相同的一套暴露估计，即根据每个人的居住和职业历史以及自来水消耗量估计其过去暴露的程度，并结合一套复杂的相互关联的空间命运和运输模型，估计过去全氟辛酸盐在环境中的移动。概率分布将被用来表征C8健康项目中每个同意参与者的各种暴露模型组件所造成的不确定性。然后将使用蒙特卡罗技术来确定这些不确定性对暴露估计的综合影响。然后，这些诱导的先验分布将与2005-2006年为研究参与者获得的全氟辛酸盐血清测量相结合；由此产生的后继暴露估计将用于重新分析妊娠高血压综合征/先兆子痫之间的联系。这些贝叶斯模型应该是1。)改进流行病学分析的暴露排名和2.)可以直接描述空间不确定性对流行病学调查结果可靠性的影响。

DESCRIPTION (provided by applicant): Many environmental pollutants have distinct patterns in both space and time. One of the main challenges in epidemiologic studies of the health effects of these pollutants is characterizing those patterns accurately enough to rank study participants by the amount of pollutant exposure at any particular point in time or cumulatively over a given period of time. We propose new Bayesian statistical models to combine spatial and temporal exposure information from a variety of sources including source emissions, fate and transport models, exposure questionnaires, and biochemical measurements such as pollutant concentrations in participant's blood samples. Bayesian methods also produce quantitative estimates of the extent of uncertainty regarding each participant's amount of exposure. This characterization of uncertainty is critical for understanding the reliability of spatial epidemioloic analyses linking pollutant exposures to adverse health effects. We propose to further develop and implement these new Bayesian models using data from the C8 Health Project, a recent study of over 69,030 community residents exposed to perfluorooctanoate released from a major production facility in West Virginia. Previous analyses have associated perfluorooctanoate exposures in these study participants to pregnancy-induced hypertension/preeclampsia, testicular cancer, and kidney cancer; numerous other epidemiologic analyses are currently underway. These analyses all rely on the same set of exposure estimates, which estimate the extent of past exposure to each individual based on his/her residential and occupational history and tap water consumption, in conjunction with a complex set of linked spatial fate and transport models that estimate the past movement of perfluorooctanoate in the environment. Probability distributions will be used to characterize the uncertainty contributed by the various exposure model components for each consented participant in the C8 Health Project. Monte Carlo techniques will then be used to determine the combined effects of those uncertainties on the exposure estimates. These induced prior distributions will then be combined with perfluorooctanoate blood serum measurements obtained for the study participants in 2005-2006; the resulting combined posterior exposure estimates will be used to reanalyze the pregnancy-induced hypertension/preeclampsia association. These Bayesian models should 1.) improve exposure rankings for epidemiologic analyses and 2.) allow for direct characterization of the effects of spatial uncertainty on the reliability of epidemiologic findings.