Flexible causal inference methods for estimating longitudinal effects of air pollution on chronic lung disease

摘要 本申请为指导定量研究职业发展奖已提交与 支持马林斯基博士作为生物统计学交叉点的定量研究人员的职业生涯的目标， 流行病学和环境健康数据科学。培训和研究计划建立在博士。 马林斯基在统计学和计算机科学方面的定量跨学科背景，特别是他的 在因果推理和机器学习方面的专业知识。总体研究目标是开发新的 因果推理的统计方法，满足重要的分析挑战，在观察 环境流行病学，并将这些方法应用于空气污染和慢性肺部疾病的研究， 使用的数据来自长期的多种族动脉粥样硬化研究（梅萨）。这些方法将用于 估计几种环境空气污染物（臭氧、细颗粒物和氮氧化物）对 肺气肿的进展和肺功能在延长的时间段内的下降。严格调查 这些关系对于促进我们对病因和机制的理解都很重要 潜在的肺部疾病，并告知有关污染浓度水平的监管政策。重点 将扩展和调整方法，从观测纵向数据中进行因果推理， 先前已开发，以适应时变混杂和量化不确定性， 未测量的混杂因素，但从未应用于空气污染和慢性肺疾病的复杂纵向数据 疾病这些将被用来估计长期肺部疾病的后果假设的变化， 空气污染暴露水平。研究计划的目标1扩展了现有方法以应对挑战 具体到空气污染流行病学，即通过利用机器学习的进步来估计鲁棒性 暴露倾向和灵活的剂量反应函数。研究计划的目标2利用了这些 方法来调查关于上述污染物之间的关系的假设， 梅萨数据中的肺部疾病指标，并确定脆弱的亚群。目标3将延长 统计文献中的反事实敏感性分析方法，该方法量化了 将未测量的混杂因素应用于梅萨的设置，并将该方法应用于梅萨数据。应用 通过以下领域的监督和教学指导，制定指导和职业发展计划： 空气污染科学、环境流行病学、气候、纵向研究设计和其他相关主题 为梅萨数据建立可靠的分析模型。马林斯基博士将得到 在空气污染科学和测量、肺病、生物统计学方面具有相当专业知识的指导团队 方法和健康的环境决定因素。该奖项将确立马林斯基博士作为一个独立的 研究员在这个跨学科领域，使他能够成功地竞争R01资金。

Abstract This application for a Mentored Quantitative Research Career Development Award has been submitted with the goal of supporting Dr. Malinsky’s career as a quantitative researcher at the intersection of biostatistics, epidemiology, and data science for environmental health. The training and research plan build on Dr. Malinsky’s quantitative interdisciplinary background in statistics and computer science, in particular his expertise in causal inference and machine learning. The overarching research goal is to develop novel statistical methods for causal inference that meet important analytical challenges in observational environmental epidemiology and apply these methods to the study of air pollution and chronic lung diseases, using data from the longstanding Multi-Ethnic Study of Atherosclerosis (MESA). The methods will be used to estimate the effects of several ambient air pollutants (ozone, fine particulate matter, and oxides of nitrogen) on progression of emphysema and decline in lung function over an extended time period. Rigorously investigating these relationships is important both for advancing our understanding of the etiology and mechanisms underlying lung disease and to inform regulatory policies concerning pollution concentration levels. The focus will be on extending and adapting methods for causal inference from observational longitudinal data, which have been previously developed to accommodate time-varying confounding and quantify uncertainty due to unmeasured confounding, but never applied to complex longitudinal data on air pollution and chronic lung disease. These will be used to estimate the long-term lung disease consequences of hypothetical changes to air pollution exposure levels. Aim 1 of the research plan extends existing methods to address challenges specific to air pollution epidemiology, namely by exploiting advances in machine learning to estimate robust exposure propensities and flexible dose-response functions. Aim 2 of the research plan leverages these methods to investigate hypotheses about the relationships between the aforementioned pollutants and measures of lung disease in the MESA data and identify vulnerable subpopulations. Aim 3 will extend an approach to counterfactual sensitivity analysis in the statistical literature that quantifies uncertainty due to unmeasured confounding to the setting of MESA and apply this approach to the MESA data. The application delineates plans for mentoring and career development via supervision and didactic instruction in the areas of air pollution science, environmental epidemiology, climate, longitudinal study design, and other topics relevant to the construction of credible analysis models for the MESA data. Dr. Malinsky will be supported by a mentoring team with considerable expertise in air pollution science & measurement, lung disease, biostatistical methods, and environmental determinants of health. The award will establish Dr. Malinsky as an independent investigator in this interdisciplinary area and enable him to successfully compete for R01 funding.