Statistical methods to characterize causal mechanisms by which air pollution affects the recurrence of cardiovascular events

项目摘要 我们这个时代最大的危害之一是空气污染，有充分的证据表明，接触细颗粒物（例如， PM2.5）增加心血管疾病（CVD）和死亡的风险。然而，有两个关键的知识 差距。首先，现有的研究主要认为，第一次不良健康事件的发生是健康状况恶化， 因此，总体疾病负担和PM2.5暴露对疾病进展的影响仍然存在 未知其次，据我们所知，没有研究评估暴露的因果途径， 空气污染物会影响复发性心血管事件。如果不能更好地了解疾病的进展 和临床调解人，我们的能力，告知监管政策和预防疾病是严重阻碍。 有限的注意力已经给予了发展的方法来评估因果关系的时间变化的前， 尤其是在像死亡这样的终结事件面前。一个相关的方法差距是， 识别（时变）介质或估计复发事件时变暴露的介导效应 结果。这一建议解决了因果推理中的这两个关键方法论差距， 目的是阐明（i）PM2.5对CVD负担和进展的影响;以及（ii） 空气污染暴露影响这些事件的途径。实现这些目标将得到以下方面的推动： 对前所未有的数据收集进行新的分析，包括：（a）已经统一和相互联系的A部分 医疗保险数据（2000年至2019年3300多万受试者），包括个人（例如，年龄，性别，种族，入院日期- 任何可能的住院原因的识别，死亡日期）和邮政编码（例如，每日PM2.5、O3和 NO2水平;许多潜在的混杂因素）水平;和，（B）这些数据的增强版本，包括个体 代表性样本的相关B部分数据（医生访视和任何原因的门诊程序）， 1500万医疗保险登记者（2012年至2019年），以及（c）医疗保险当前贝内官方调查（MCBS）数据， 在同一研究期间，作为次要分析，以解释由于未测量的混杂因素导致的可能偏倚。 在方法论方面，我们提出了稳健的边际和结构嵌套模型，允许估计 随时间变化的PM2.5暴露对CVD事件复发的瞬时和延迟效应（目的1）。 我们进一步提出了一个半参数的方法来确定临床上重要的途径，暴露于空气中， 污染物增加了心血管疾病住院和死亡复发的风险，并估计相应的 目的2（Aim 2）。我们将实现我们的方法并将其应用于上述丰富的数据 来源，特别侧重于（目标3）：（i）估计PM2.5暴露对两个相关的因果影响 在存在死亡率的情况下，复发事件的因果被估量;（ii）确定关键介质并表征 暴露于空气污染会增加CVD进展和死亡风险的临床途径;以及， (iii)识别和调查在生物学上合理的疾病群。最后，我们将创建和 传播开放源码、经同行审查的统计软件，以确保易于使用和获取（目标4）。

Project Summary One of our era's greatest scourges is air pollution, and it is well documented that exposure to fine particles (e.g., PM2.5) increases the risk of cardiovascular disease (CVD) and death. However, there are two critical knowledge gaps. First, existing studies have mainly considered the occurrence of the first adverse health event as health out- comes; hence, the impact of overall disease burden and of PM2.5 exposure on disease progression both remain unknown. Second, to our knowledge, there are no studies assessing the causal pathways by which exposure to air pollutants impacts recurrent cardiovascular events. Without a better understanding of disease progression and clinical mediators, our ability to inform regulatory policy and prevent disease is severely hampered. Limited attention has been given to developing methods for assessing the causal effect of time varying ex- posures, especially in the presence of a terminating event like death. A related methodological gap is the ability to identify (time varying) mediators or estimating mediated effects of time varying exposure for a recurrent event outcome. This proposal addresses these two critical methodological gaps in causal inference, the overarching goals being to elucidate (i) the impact of PM2.5 on the burden and progression of CVD; and, (ii) the key causal pathways by which air pollution exposure impacts such events. Accomplishing such goals will be facilitated by new analyses of an unprecedented data collection consisting of (a) an already harmonized and linked Part A Medicare data (33+ million subjects 2000 to 2019) at both the individual (e.g., age, gender, race, date of hospital- ization for any of the possible causes for hospitalizations, date of death) and zip code (e.g., daily PM2.5, O3 and NO2 levels; many potential confounders) levels; and, (b) an augmented version of these data including individu- ally linked Part B data (doctor visits, and outpatient procedures for any cause) for a representative sample of over 15 million Medicare enrollees (2012 to 2019) and, (c) the Medicare Current Beneficiary Survey (MCBS) data for the same study period, as a secondary analysis to account for possible bias due to unmeasured confounding. In methodological terms, we propose robust marginal and structural nested models that allow estimation of both instantaneous and delayed effects of time varying exposure to PM2.5 on recurrence of CVD events (Aim 1). We further propose a semiparametric approach to identify clinically important pathways by which exposure to air pollutants increases the risk of recurrence of CVD hospitalizations and death, and estimate the corresponding mediated effects (Aim 2). We will implement our methods and apply them to the above-described rich data source, focusing specifically on (Aim 3): (i) estimating the causal effects of PM2.5 exposure on two relevant causal estimands for recurrent events in the presence of mortality; (ii) identifying key mediators and characterize clinical pathways through which exposure to air pollution increases risks of CVD progression and death; and, (iii) identifying and investigating the disease groups that are biologically plausible. Finally, we will create and disseminate open-source, peer-reviewed statistical software to ensure ease-of-use and accessibility (Aim 4).