(B-190) Estimating Prevalence for 50 Common Nonfatal Health Conditions by Race/Ethnicity and County in the United States, Incorporating a Novel Correction for Healthcare Access Disparities

Background: United States (US) law requires clinical trial sponsors to submit plans for population-representative enrollment. However, prevalence data sources by race and ethnicity (R/E) are usually fragmented due to absence of universal data integration for most conditions, and biased due to reduced healthcare access experienced by many demographic groups. We previously presented a novel approach to approximate prevalence by R/E for fatal health conditions, leveraging mortality estimates using the vital registration system. However, that method cannot be used for nonfatal conditions.

Objectives: Obtain reliable approximations of US prevalence by R/E, for common nonfatal health conditions.

Methods: We estimated average national-level age- and sex-adjusted prevalence disparities between five standard US R/E groups, using Medical Expenditure Panel Survey (MEPS) data from 2014-2022, fitting a survey-compatible logistic regression model for each condition. The models adjusted for standard healthcare access mediators (insurance coverage and English fluency), and a novel variable that compares participants’ number of diagnosed conditions with their self-reported VR-12 physical health score, relative to age/sex peers. We also fitted sensitivity models adjusting only for age, sex, and panel.

The full-model disparity estimates were combined with IHME’s Global Burden of Disease estimates, which provided prevalence magnitude and age/sex/year/geographic resolution.

Results: Over 150 distinct health conditions were sufficiently common in MEPS for deriving disparity estimates. In sensitivity analyses that did not adjust for healthcare access, prevalence estimates for non-White groups appeared to be biased downward. For example, averaged across all conditions Blacks and Hispanics had ~30% and ~35% lower age/sex/panel-adjusted prevalence vs. Whites, respectively, despite both groups reporting worse average age-adjusted VR-12 health scores than Whites. After incorporating healthcare access covariates, the average difference was reduced to ~10%. The novel covariate utilizing health scores was particularly impactful. Nearly 50 of the conditions were mapped to nonfatal causes modeled by IHME, enabling generation of detailed estimates.

Conclusions: Surveys with sufficiently rich data collection offer a way to substantially mitigate access-related biases in health burden estimates. In particular, direct information about individuals’ health, even via simple self-reported questionnaires, can be leveraged to mitigate biases not captured elsewhere. These findings can inform the design of future surveys, to further improve bias mitigation.