Fighting Obesity in Mississippi: Using Small Area Estimation Techniques to Identify Targeting Areas

Background and objective: Mississippi ranks number one in the nation in adult obesity rate in the past five consecutive years. Obesity contributes to the major chronic diseases in the State of Mississippi. To take targeted actions against this threatening epidemic, it is needed to identify counties of high obesity rates. However, the sample sizes of Behavioral Risk Factor Surveillance System data in 98% of Mississippi counties are too small for reliable direct estimation. Objective of this study is to develop a small area estimation (SAE) method to overcome this hurdle. Methods: A hierarchical generalized linear model is developed to incorporate individual level BRFSS survey data (2007-2009) with county level census data. This mixed model includes random county-specific effects not accounted for by other variables in the model. Synthetic estimation technique is also applied to age-adjust county prevalence to compare with model estimates. To validate, county level estimates are aggregated to compare with direct state level estimation. Results: For year 2009, county prevalence range from 30.5% to 44.2%, median is 35.4%. Twelve counties (14.6%) had obesity prevalence over 40%. Age, gender, race, education level and employment status have significant effect on obesity status. Model estimates have high precision, and the differences between aggregated SAE estimates and State direct estimates are minimal. Conclusion: With proper modeling and appropriate choice of auxiliary data, small area estimation methods can be instrumental in accessing health status at sub-state levels.