Abstract 33: Instrumental Variables Regression to Account for Bias in Associations Between the Food Environment and Diet Quality in the Coronary Artery Risk Development in Young Adults (CARDIA) Study

Abstract

Introduction: Although observational evidence suggests that the food environment is associated with diet behaviors, the literature is inconsistent. Most studies ignore potential bias due to healthier people selecting healthier residential neighborhoods and healthier food outlets locating in areas where there is demand. Hypotheses: Greater availability of fast food restaurants (vs. full-service restaurants) and convenience stores and grocery stores (vs. supermarkets) is associated with lower diet quality, accounting for potential bias. Methods: We used 20 years (1985-6; 1992-3; 2005-6) of data from CARDIA study (n=26,203 observations) with geographically-linked, time-varying neighborhood-level data. Using instrumental-variables regression with a generalized method of moments estimator, we estimated the longitudinal association between the percent of neighborhood fast food restaurants, full-service restaurants, convenience stores, grocery stores, and supermarkets (per total food outlets within 1-km network distance of respondents’ residences at each exam) and an a priori diet quality score, with higher scores indicating higher diet quality. We adjusted for individual-level sociodemographics, health behaviors, baseline study center, exam year, and total other food outlets. We included theoretically-justified instrumental variables related to neighborhood-level sociodemographics, road connectivity, physical activity facilities, and housing prices. Results: Across exam years the mean diet quality score was 64.3 (SD=12.8; range: 24-107). Relative to the percent of neighborhood full-service restaurants, a 10% increase in the density of fast food restaurants (mean=12.3; SD=13.0) was associated with a lower a priori diet quality score over time (β=-3.5; 95% CI=-6.0, -0.90; p=0.01), after adjusting for potential residential selection and purposeful store placement. However, relative to the percent of supermarkets, neither the percent of convenience stores (β=-0.85; 95% CI=0.12, -1.93; p=0.12), nor grocery stores (β=-0.72; 95% CI=-1.70, -0.26; p=0.15), was associated with diet quality. Conclusions: Compared to standard regression models, our findings suggest that efforts to reduce the availability of neighborhood fast food restaurants, but not other types of food outlets, may be associated with minor improvements to diet behaviors. This suggests that the selective nature of where people and stores locate may influence associations of the food environment with diet quality.