Multi-level re-weighted regression models for small area analysis

Despite heightened national attention regarding the existence of disparities in health by race, progress in reducing these disparities has been slow. Part of the problem may be that the agenda to reduce disparities in health have been set on the national and state levels and are based on national and state level data. This is potentially problematic because these levels are far removed from the individual level where health outcomes are realized. County health departments, community-based organizations, and policymakers need local health data for program evaluation, program planning, and resource allocation.

The Behavioral Risk Factor Surveillance System (BRFSS) is commonly used to estimate the prevalence of chronic disease, however, the small sample sizes of county-level data does not allow for direct calculation of county-level prevalence rates. To address this issue multilevel re-weighted regression models were developed to obtain race-specific county level prevalence estimates by combining three years (1999-2001) of the BRFSS data with 2000 U.S. Census data. Social, demographic, and behavioral risk factors contributing to the prevalence of the disease were modeled on three levels; state, county, and individual. Separate models were developed for each disease outcome; asthma, diabetes and hypertension. Comparison of prevalence estimates obtained from the multilevel re-weighted regression models and commonly used existing data extrapolation methods; the synthetic method, spatial data smoothing, temporal estimation, and regression analysis will be obtained by calculating discrepancies between model estimates and “gold standard” chronic disease prevalence estimates obtained through local data collection.