Modeling Complex Exposures and Reproductive Outcomes

描述(由申请者提供)：调查人员经常不遗余力地获得仔细、详细的暴露测量，这些测量可能有多个维度，并可能随着时间的推移而变化，例如饮食、压力或血压。在评估个人暴露史与健康事件发生时间或发生率之间的关联时，重要的是降低这种多变量暴露史数据的维度，以增加统计能力。尽管如实践中通常所做的那样，用简单的摘要替换多变量暴露信息有时可以提高可解释性和统计能力，但通常不清楚如何最好地汇总手头的信息。此外，将详细数据简化为天真的总结往往与尽可能获得最准确的暴露评估的研究目标背道而驰。我们感兴趣的是开发和应用统计方法，使以证据为基础的摘要能够以最大限度地提供有关感兴趣的结果的信息的方式客观地构建。根据怀孕、感染和营养(PIN)研究的数据，我们提出了一个多情节暴露过程和生殖结果的贝叶斯分层模型。PIN研究是一项关于早产的前瞻性队列研究。关于暴露的时间、持续时间和强度的数据在一个框架内使用一个共同的脆弱术语来汇总，该框架解释了过程中随时间的变化、相关的暴露测量和缺失的数据。对曝光效果的推断可以基于使用有效的MCMC算法获得的后验密度。这些方法将使用妊娠、感染和营养研究数据来应用于阴道出血和妊娠持续时间，但在生殖健康、流行病学和其他领域有广泛的应用。

DESCRIPTION (provided by applicant): Investigators often go to great lengths to obtain careful, detailed measures of exposure, which may have multiple dimensions and may change over time, e.g. diet, stress, or blood pressure. In assessing the association between an individual's exposure history and the time or rate of occurrence of a health event, it is important to reduce the dimensionality of this multivariate exposure history data in order to increase statistical power. Although replacing the multivariate exposure information with a simple summary, as is typically done in practice, can sometimes improve interpretability and statistical power, it is typically not clear how best to summarize the information at hand. In addition, reducing detailed data into naive summaries often runs counter to the study goals of obtaining the most accurate assessment of exposure possible. We are interested in developing and applying statistical methods, allowing evidence-based summaries to be constructed objectively in a manner that maximizes information about the outcomes of interest. Motivated by data from the Pregnancy, Infection, and Nutrition (PIN) study, a prospective cohort study of preterm birth, we propose a Bayesian hierarchical model for a multiple episode exposure process and a reproductive outcome. Data on timing, duration, and intensity of exposure are summarized using a shared frailty term within a framework that accounts for changes in the process over time, correlated exposure measures, and missing data. Inferences on exposure effects can be based on the posterior density obtained using an efficient MCMC algorithm. The methods will be applied to vaginal bleeding and duration of gestation using the Pregnancy, Infection, and Nutrition study data but have broad applications in reproductive health, epidemiology, and other areas.