Rate Stabilizer Tool (RST): Smoothing the Way to Robust Sub-County Estimates of Chronic Disease Outcomes

BACKGROUND: There is a need for sub-county measures of chronic disease outcomes; a number of methodological challenges constrain their application and use in the public health setting. Small populations and low counts often accompany sub-county geographic units, calling for advanced statistical methods to generate robust measures of health outcomes. To facilitate the calculation of statistically robust sub-county measures, we have developed the Rate Stabilizer Tool (RST) to produce estimates for specific chronic disease outcomes. Our objectives: To assess the utility of the RST using individual georeferenced chronic disease mortality data. Specifically, we were interested in: evaluating the practicality of an in-tool age adjustment feature; and understanding the value added from employing empirical Bayesian methods to generate smoothed estimates.

METHODS:  The RST integrates: US Census data, georeferenced individual-level health data, and empirical Bayesian methods to generate standardized, map-ready, age-adjusted, spatially smoothed estimates along with measures of uncertainty. The RST was tested by several local health departments (LHDs) in North Carolina (NC) in partnership with the NC State Center for Health Statistics. The group tested the RST with multiple years of georeferenced mortality data with specific chronic disease ICD-10 codes to generate age-adjusted US Census Tract level death rate estimates. We compared age-adjusted rates that were un-smoothed with those that were smoothed using Bayesian Hierarchical modeling at both the county and tract level. The smoothed rates included 95% confidence intervals which allowed for calculating measures of uncertainty.

RESULTS:  After successfully acquiring five years of georeferenced mortality data for a multi-county area, LHD staff used the RST to assess its performance and utility. After quickly generating estimates at the county and tract level, staff were able to compare of maps important variation within counties at these two different spatial scales. Staff also detected significant differences between smoothed versus non-smoothed rates, particularly in census tracts with fewer cases and small populations.

CONCLUSIONS:  This pilot of the Rate Stabilizing Tool (RST) demonstrated the ability of staff in local health departments to generate statistically robust age-adjusted heart disease death rates at the census tract level using advanced statistical methods. We noted substantial information gains when observing sub county vs county estimates, especially in those with a mix of urban and rural populations. Next steps for the RST will focus on implementing and incorporating a spatial empirical Bayesian smoothing technique, thereby providing health departments with the ability to generate optimal measures for documenting geographic disparities of health outcomes in their communities.