Causal machine learning methods for studying the effects of environmental exposures on childhood cancer using natural experiments

项目概要： 我的目标是建立一个独立的研究计划，在因果推理方法的发展， 调查儿童癌症的环境原因。这个K 01将使我能够进行集中， 深入的研究将为该计划奠定基础，并获得环境，生物， 和流行病学培训，以最大限度地提高我工作的严谨性和影响力。 研究：我们建议开发新的因果机器学习（ML）方法，以实现严格的分析 环境自然实验（NE），用于估计环境暴露对 儿童癌症。研究环境暴露与 儿童癌症受到挑战的困扰，并产生不一致的结果。我们认为， 最近地方环境监管计划的激增创造了大量相关的内斯， 这是研究这些关系的另一种有力方法。然而，现有的NE分析方法 不太适合环境卫生背景。特别是，现有的方法在存在罕见的 结果，如儿童癌症（目标1），他们不能提供洞察的时间， 儿童最容易受到任何有害的接触影响（目标2）。我们提出了因果ML方法， 克服这些挑战，并将其应用于NE，以研究与交通有关的空气毒物对儿童的影响 白血病我们还提供实现这些方法的开源软件（目标3）。 职业发展和培训：鉴于我之前在统计和数据方面的广泛培训和经验， 科学，该奖项资助的培训的主要目的将是获取主题 熟练程度，这将为我提供所需的见解，以创造更有效和更有影响力的 环境卫生方法。具体来说，我将追求的生物学和流行病学知识， 儿童癌症和环境健康和暴露生物学。培训将通过一个 结合（1）如上所述的动手合作研究;（2）密集的跨学科 导师制，导师专门从事环境健康，儿科肿瘤学，癌症生物学和 流行病学和统计学;（3）精心挑选的流行病学系课程， 环境健康，和细胞生物学在哈佛;和（4）相关的会议，讲习班和研讨会。我 我将特别强调在我的所有培训领域建立一个专家合作者网络。 环境：哈佛医学院是世界顶级研究团队的所在地， 癌症和环境健康。由于哈佛在这些领域处于科学发现的前沿， 其无与伦比的资源，充满活力的学术氛围，以及对合作科学的促进 它整合了跨学科的知识，为这些主题的培训提供了理想的环境。

Project Summary: My goal is to build an independent research program in the development of causal inference methods for investigating environmental causes of childhood cancer. This K01 will enable me to conduct the focused, intensive research that will lay the groundwork for that program and to acquire the environmental, biological, and epidemiological training needed to maximize the rigor and impact of my work. Research: We propose to develop new causal machine learning (ML) methods that enable rigorous analysis of environmental natural experiments (NE) for estimation of the causal effects of environmental exposures on childhood cancer. Classical approaches to studying relationships between environmental exposures and childhood cancer are plagued with challenges and are yielding inconsistent findings. We contend that the recent proliferation of local environmental regulatory programs has created ample relevant NEs, which provide a powerful alternative approach to study these relationships. However, existing methods for NE analysis are poorly-suited for environmental health contexts. In particular, existing methods fail in the presence of rare outcomes like childhood cancer (Aim 1), and they are not able to provide insight into the timing at which children are most susceptible to any adverse exposure effects (Aim 2). We propose causal ML methods that overcome these challenges and apply them to a NE to study the effects of traffic-related air toxics on childhood leukemia. We also provide open source software implementing these methods (Aim 3). Career Development and Training: Given my extensive prior training and experience in statistics and data science, the primary aim of the training funded by this award will be the acquisition of subject-matter proficiency, which will provide me with the insights needed to create more effective and impactful environmental health methods. Specifically, I will pursue knowledge in the biology and epidemiology of childhood cancer and in environmental health and exposure biology. The training will be achieved through a combination of (1) hands-on collaborative research as described above; (2) intensive cross-disciplinary mentorship, with mentors specializing in environmental health, pediatric oncology, cancer biology and epidemiology, and statistics; (3) carefully-selected coursework in the Departments of Epidemiology, Environmental Health, and Cell Biology at Harvard; and (4) relevant conferences, workshops, and seminars. I will place special emphasis on establishing a network of expert collaborators in all my areas of training. Environment: The Harvard Medical Campus is home to the top research teams worldwide in both childhood cancer and environmental health. Due to Harvard’s position at the forefront of scientific discovery in these fields, its unparalleled resources, its vibrant intellectual atmosphere, and its promotion of collaborative science that integrates knowledge across disciplines, it provides an ideal environment in which to train on these topics.