Geographic variations in obesity: A multilevel study of BRFSS data

Our goal is to understand the geographic variations in obesity at the county level in the US, and to identify county characteristics that predict higher and lower rates of obesity. Data from the Behavioral Risk Factor Surveillance Survey (BRFSS) was obtained for years 2001-2005. Counties were selected if an average of 25 African Americans and 25 Whites completed the BRFSS survey each year (N=164 counties, 367,485 interviews). Census data were extracted for each county and five composite measures of county characteristics were derived using factor analysis: 1) Economic distress; 2) City size; 3) Economic privilege; 4) Black concentration; and 5) Segregation-crowding. Multilevel linear models predicting body mass index (BMI=kg/m2) were constructed using the HLM program (SSI, Lincolnwood, IL). The level 1 model included age, gender, race/ethnicity, education, and income (p's <0.0001). A level 2 model of intercepts (differences in mean county BMI) was constructed using the census variables. County intercept was predicted by City size (p<0.0001), Economic privilege (p<0.0001), and Segregation-crowding (p<0.0001). Increases in County population, concentration of Economic privilege, and degree of Segregation-crowding are associated with decreases in mean BMI. Mean BMI ranged from a low of 25.0 in New York City to a high of 29.1 in Washington County, Mississippi. A logit-link model using obesity (BMI°İ30) as the outcome yielded similar results. Obesity ranged from 13.5% in New York City to 36.7% in Washington County, Mississippi. There is considerable variation in obesity at the County level, and this variation can be modeled using multilevel linear models.