UC Berkeley

The adverse effects of air pollution on human life is of serious concern for today’s society. Two population groups that are especially vulnerable to air pollution are pregnant women and their growing fetuses, and the focus of this thesis is to study the effects of air pollution on these populations. In order to address the methodological limitations in prior research, we quantify the impact of air pollution on various adverse pregnancy outcomes, utilizing machine learning and novel causal inference methods. Specifically, we utilize two semi-parametric, double robust, asymptotically efficient substitution estimators to estimate the causal attributable risk of various pregnancy outcomes of interest. Model fitting via machine learning algorithms helps to avoid reliance on misspecified parametric models and thereby improve both the robustness and precision of our estimates, ensuring meaningful statistical inference. Under assumptions, the causal attributable risk that we estimate translates to the absolute change in adverse pregnancy outcome risk that would be observed under a hypothetical intervention to change pollution levels, relative to currently observed levels. The estimated causal attributable risk provides a quantitative estimate of a quantity with more immediate public health and policy relevance.