Flexible models for elucidating health disparities

Measurement of disparities in health-related outcomes between sociodemographic subpopulations is of increasing interest in both public health research and U.S. public health policy. The goals of Healthy People 2010 (HP2010) specifically include reference to the elimination of racial and geographic disparities in health outcomes, and recent guidance from the CDC has reinforced this by making specific recommendations concerning statistical methods for measuring such health disparities.

We provide an example of the type of statistical model required to fully elucidate the complex etiology of health disparities in accord with the recommendations of HP2010 and the CDC. We develop novel Bayesian statistical methodology that: 1) accommodates individual-level data in a multilevel modeling framework (needed to measure contextual effects such as whether the effect of race at the individual level differs for areas with differing racial demographics); 2) incorporates spatial information, resulting in more reliable and precise estimates that identify geographic differences in racial disparities; and 3) produces flexible joint probabilities, which allows inference about all marginal and conditional probabilities associated with the model, including alternative measures of disparity (e.g., the absolute and relative) and subgroup-specific component rates of disparity.

We illustrate our approach with North Carolina Detailed Birth Record data from 1999-2003, focusing on the variables low birth weight, maternal race, maternal tobacco use, and infant sex. We demonstrate the potential of our model to answer policy questions and inform intervention and mitigation strategies.