Causal Inference With Race/Ethnicity: Analysis And Interpretation.

Abstract

A large body of research is devoted to understanding mechanisms that underlie racial health disparities. This work is often based on mediation procedures that may not be justified with race as the exposure of interest. We will present contemporary concerns related to causal interpretation of race/ethnicity in epidemiologic studies, and comment on valid and invalid interpretations of key empirical findings. Additionally, we will discuss the use of common analytic approaches to assess mechanisms by which racial health disparities occur, and compare these to more general methods including inverse probability weighted marginal structural models, the structural transformation method (also known as sequential g-estimation), doubly robust g-estimation of a structural nested model, and doubly robust targeted minimum loss based estimation. We use simulation data, and an empirical dataset of nearly 1 million pregnancies to assess the role of pre-pregnancy obesity in explaining the racial disparity in infant mortality. Using standard methods, we estimated that pre-pregnancy obesity explained roughly 6.5% of the racial disparity in infant mortality. In contrast, more general methods yielded estimates between 10% and 20%. Our empirical findings suggest that use of standard approaches for mediation can have important consequences on empirical findings.