Multilevel Regression and Poststratification for Small Area Estimation of Population Health Outcomes: Validation and Application

• Activity Number: 615
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Xingyou Zhang*+ and James B. Holt and Shumei Yun and Hua Lu and Kurt Greenlund and Janet B. Croft
• Companies: CDC and CDC and Missouri Department of Health and Senior Services and CDC and CDC and CDC
• Keywords: Behavioral Risk Factor Surveillance System (BRFSS) ; Missouri County-level Study ; small area estimation ; multilevel regression and poststratification ; external validation ; population health outcomes

Abstract

Population health outcomes have demonstrated substantial variations at a variety of geographic levels: from local community to state. Most health surveys have been designed to monitor national or state population health. In order to meet the data needs at local levels in public health practice, small area models have been applied to national health surveys to generate local small area estimates of population health outcomes. But very few model-based small area estimates have been evaluated via external validation. Thus, we applied multilevel logistic models with 2011 BRFSS and poststratification with 2010 census data to generate county-level COPD and current smoking, diabetes and obesity prevalence estimates for 115 Missouri counties. We compared them with those direct survey estimates of the 2011 Missouri County-level Study, which has a sample size of 400 or more for each county. Our model-based estimates are within the 95% CIs of the direct survey estimates in at least 95 counties. The external validation confirmed that the multilevel regression and poststratification approach could generate both reliable and accurate county-level estimates for the majority of Missouri counties.