Methods/Statistics
Old tools to address persistent problems: assessing health disparities and differences with attributable fractions Alexander Keil* Alexander Keil Maria E Kamenetsky Maya Spaur Jessica M Madrigal Whitney R Robinson
Understanding disparities or differences in health across social and economic strata within populations is key to improving the overall health of populations. One focus of health disparities research in epidemiologic studies has been the differential impact of individual exposures among groups across which a health disparity exists. Current tools for assessing and understanding health disparities in epidemiologic data include evaluation of effect measure modification (EMM, does the impact of an additional unit of exposure vary across groups on some scale?) and decomposition analysis (to what extent is the disparity across groups mediated by the association between group membership and the exposure?). Despite common usage, these tools are limited. EMM evaluation fails to account for how the differential distribution of exposure across groups may impact disparities, while the decomposition analysis focuses directly on effects of group membership, which carries possibly untenable assumptions for making causal inferences and may not represent a modifiable factor.
Here, we make a modest proposal to adopt an existing, underutilized tool for assessing how exposures of interest impact health and health disparities: attributable fractions. Attributable fractions estimate the impact of eliminating exposure in a population, making them useful for studies of modifiable exposures that may or may not be equally distributed across population groups. Using synthetic examples, we demonstrate that attributable fractions can identify exposures that contribute to health disparities in settings in which there is no EMM. We demonstrate how attributable fractions can be extended, using tools like g-computation, to multiple exposures and settings where eliminating exposure is not feasible. This work demonstrates a set of tools for studying health disparities that focus on modifiable factors and thus better contribute to public health decision makers’ efforts to reduce health disparities.