Bayesian Small Area Estimates of Diabetes Prevalence by U.S. County, 2005

10.6339/JDS.2010.08(1).650

Betsy L. Cadwell；Theodore J. Thompson；James P. Boyle；Lawrence E. Barker

Bayesian analysis ； BRFSS ； census data ； diabetes mellitus ； multilevel model ； small area estimation

Journal of Data Science

8卷1期（2010 / 01 / 01）

173 - 188

英文

County specific estimates promote understanding of national and state patterns of the diabetes burden and can help better target diabetes programs. Using Bayesian multilevel models, the authors estimated the prevalence of self reported diagnosed diabetes for adults aged 20 years or older for each of the United States' 3,141 counties/county equivalents. These estimates provide the first comprehensive county level estimates of diabetes for the U.S. and provide opportunities for the practical targeting of interventions and new lines of investigations into area level risk factors for diabetes. The ranks' posterior distribution was used to identify counties with extreme diabetes burden. Counties with high (low) diabetes burden were identified as those for which at least 95% of the posterior distribution for the rank was above (below) the median. In 2005, 428 (480) counties had high (low) diabetes burden. Design-based estimates could be obtained for 232 large population counties; model-based estimates compared favorably with these design-based estimates.