BRINGING MODERN DATA SCIENCE TOOLS TO BEAR ON ENVIRONMENTAL MIXTURES

Project Summary: Bringing Modern Data Science Tools to Bear on Environmental Mixtures Environmental exposures often cumulate in particular geographies, and the nature of the complex mixtures that characterize these exposures remains understudied. In addition, adverse environmental exposures often occur in communities facing multiple social stressors such as deteriorating housing, inadequate access to health care, poor schools, high unemployment, crime, and poverty ? all of which may compound the effects of environmental exposures. Our central objective is to develop new data architecture, statistical, and machine learning methods to assess how exposure to environmental mixtures shapes educational outcomes in the presence or absence of social stress. We focus on air pollution mixtures, childhood lead exposure, and social stressors. We will implement our proposed work in North Carolina (NC), a state characterized by diverse environmental features, industrial activities, and airsheds typified by varying pollution emission sources and resulting pollutant mixtures. To accomplish this central objective, we will first develop, document, and disseminate methods for building space-time environmental and social data architectures. We will implement this for all of NC, incorporating data on air pollution, lead exposure risk, and social exposures from 1990-2015+ (dataset 1). Second, we will refine methods for linking unrelated datasets to build a space-time child movement and outcome data architecture (dataset 2). Third, we will connect exposures (dataset 1) and outcomes (dataset 2) data via shared geography and temporality into a single, comprehensive geodatabase. Fourth, we will implement increasingly complex methods to assess the effect of environmental mixtures in the presence or absence of social stressors on early childhood educational outcomes. We will document and disseminate all of the underlying methodological work via public website. The proposed work leverages a rich array of data resources already available to the investigators (with some significantly post-processed) and allows tracking of children across space and time. Our team brings tools from modern data science (hierarchical Bayesian methods with variable selection, spatial point process models, machine learning) to bear on the critical question of how environmental mixtures shape child outcomes directly and differentially in the presence of social stress.