The prevalence of childhood obesity tripled during the last 3 decades in the United States (1
); data for 2009 through 2010, showed that 16.9% (approximately12.0 million) of US children aged 2 to 19 years were obese (2
). Besides disparities in childhood obesity among various racial/ethnic groups (2
), research shows significant disparities by geographic area: by state (5
), city (6
), and community (7
). Small-area data can reveal wide disparities in obesity outcomes and facilitate community-based initiatives for obesity prevention (8
). Having reliable data for each community or small area allows state, county, and local decision makers and health professionals to tailor programs for preventing childhood obesity to conditions and factors that affect their community (9
), identify priority areas for action, and optimize the use of limited resources.
Local public health practitioners often lack small-area data on childhood obesity. National health surveys, such as the National Health and Nutrition Examination Survey (NHANES) (www.cdc.gov/NCHS/nhanes.htm
), the National Survey of Children’s Health (NSCH) (www.cdc.gov/nchs/slaits/nsch.htm
), and the Youth Risk Behavior Survey (YRBS) were designed to provide data on national or state childhood obesity. Direct estimates of obesity rates in small areas or communities cannot be calculated on the basis of data gathered through these surveys. Use of the surveillance methods for obtaining national (ie, large-area) data to obtain small-area data on childhood obesity is prohibitively expensive.
There are, however, cost-effective methods of generating health-related data, particularly on obesity, for small-area populations (10
). Recently, considerable research has been done on multilevel, model-based, small-area estimation methods (10
). These methods can produce data on variations in the multilevel influence of local social and physical environments on health outcomes among people in small areas by using various demographic characteristics (eg, age, sex, race/ethnicity). Another advantage is that model-based small-area estimation methods borrow information from both individual-level data within the survey sample and from area-level covariates external to the original sample, and they tend to generate smoothed estimates with better precision (16
). Malec et al constructed a 2-stage hierarchical model with NHANES III to generate state-level prevalence estimates of adult overweight (11
). Li et al used Massachusetts Behavioral Risk Factor Surveillance System (BRFSS) data to generate multilevel model-based zip code-level estimates to prioritize communities for obesity prevention (10
). More recently, Congdon extended this framework for multilevel small-area estimation modeling by using BRFSS data with county-level covariates and predicted heart disease prevalence estimates for zip code tabulation area levels (13
). We used a similar approach in this study to construct a multilevel model with county- and zip code-level covariates using NSCH 2007; we then predicted census block-group level small-area estimates (SAEs) of childhood obesity by combining the estimated model parameters and block-group level covariates with population counts for children, by age, sex, and race/ethnicity.
The objectives of our study were to 1) identify and evaluate individual and geographic factors that influence childhood obesity; 2) use multilevel small-area estimation methods to generate cost-effective data on the prevalence of childhood obesity at the block-group level for the United States; and 3) characterize the geographic disparities in childhood obesity by block groups, counties, and states.