摘要：Objectives. We assessed neighborhood confounding on racial/ethnic obesity disparities among adults in New York City after accounting for complex sampling, and how much neighborhood factors (walkability, percentage Black or Hispanic, poverty) contributed to this effect. Methods. We combined New York City Community Health Survey 2002–2004 data with Census 2000 zip code–level data. We estimated odds ratios (ORs) for obesity with 2 sets of regression analyses. First, we used the method incorporating the conditional pseudolikelihood into complex sample adjustment. Second, we compared ORs for race/ethnicity from a conventional multilevel model for each neighborhood factor with those from a hybrid fixed-effect model. Results. The weighted estimate for obesity for Blacks versus Whites (OR = 1.8; 95% confidence interval = 1.6, 2.0) was attenuated when we controlled neighborhood confounding (OR = 1.4; 95% confidence interval = 1.2, 1.6; first analysis). Percentage of Blacks in the neighborhood made a large contribution whereas the walkability contribution was minimal (second analysis). Conclusions. Percentage of Blacks in New York City neighborhoods explained a large portion of the disparity in obesity between Blacks and Whites. The study highlights the importance of estimating valid neighborhood effects for public health surveillance and intervention. The prevalence of obesity among adults in the United States increased dramatically between 1980 and 2010 (16% to 36%). 1,2 The obesity epidemic has disproportionately affected some racial and ethnic minorities. 3 In New York City (NYC) in 2011 there was a 2-fold higher prevalence of obesity among non-Hispanic Black and Hispanic adults compared with non-Hispanic Whites. 4 Studies show that neighborhood factors play an important role in shaping racial and ethnic obesity disparities. 5,6 This multilevel perspective posits that minorities are more likely to live in environments that promote high energy consumption and low physical activity including limited access to healthy foods and parks (known as an obesogenic environment), which in turn influences individual-level energy consumption and expenditure. 5–8 Estimating neighborhood effects on individual-level obesity is challenging because individuals living in the same neighborhood are more likely to be similar compared with those living in other neighborhoods, which violates the principle of statistical independence of observations and can produce biased estimates. 9 A common method used in the current literature to account for confounding attributable to neighborhood clustering is multilevel modeling. 10 However, it requires stringent conditions being met including independence between random effects and covariates, a large sample size per cluster, and sufficient within-cluster variation for individual-level covariates. 10 In addition, multilevel modeling cannot properly incorporate complex sample design into estimation unless the neighborhood sample size is large and the sampling bias is weak, which limits its utility. 11 Unweighted estimates, which are usually reported from multilevel analyses, cannot be generalized to the target population from which the samples were originally drawn. As a result, although various neighborhood characteristics such as access to healthy food, built environment, neighborhood poverty, and residential segregation have been independently associated with individual-level obesity, the assumptions for multilevel modeling and the bias attributable to not adjusting for complex sample design have been rarely examined. Another gap in the current literature is that most studies tend to report the statistical significance of neighborhood effects without assessing the magnitude of their contributions to individual-level obesity. Recently a new method, the conditional pseudolikelihood method for complex survey data, has been developed to account for both confounding caused by clustering and complex sample design. 11,12 Furthermore, it controls for full confounding by neighborhood, including both observed and unobserved neighborhood-level factors. 11,12 We applied this method to obtain unbiased estimates of obesity disparities by race/ethnicity in NYC after adequately accounting for neighborhood confounding and complex sample design. In addition, to address the limitations in current research, we estimated the total neighborhood effect on racial/ethnic disparities in obesity and assessed the extent to which this effect was attributed to each of 3 neighborhood characteristics (neighborhood walkability, residential segregation, and neighborhood poverty).
