Longitudinal Analysis of Neighborhood Food Environment and Diabetes Risk in the Veterans Administration Diabetes Risk Cohort

Key Points Question Is there an association between the presence of fast-food restaurants and availability of supermarkets in neighborhoods and the risk of developing type 2 diabetes? Findings In this longitudinal cohort study of 4 100 650 veterans, the relative availability of fast-food restaurants compared with all restaurants within neighborhoods was associated with an increased risk of diabetes across a spectrum of rural to high-density urban settings, whereas the availability of supermarkets had an inverse association with incident diabetes in suburban and rural communities only. Meaning These findings suggest that policies to shift the mix of fast-food restaurant and supermarket distribution in neighborhoods may be associated with reduced diabetes risk.


Introduction
Diabetes is a major cause of morbidity and mortality in the US. 1,2 In 2018, the Centers for Disease Control and Prevention estimated that 34.2 million adults aged 18 years and older in the US (13% of the population) had diabetes. 1 Although the risk of diabetes has increased in all parts of the country since the late 1990s, there are substantial geographical disparities in diabetes prevalence and incidence. 1,3,4 County-level analyses have highlighted age-adjusted prevalence of diabetes ranging from 1.5% (mostly counties in the West) to as high as 33.0% in others (mostly counties in the Southeast). 1 Substantial geographical disparities have also been observed between neighboring counties with similar demographic profiles, suggesting a heterogeneous impact of neighborhoodlevel factors on diabetes. [5][6][7][8] A growing body of literature has examined the role of the food environment on the risk of diabetes. [9][10][11][12][13] Longitudinal studies have found that better neighborhood resources, such as suitability for physical activity and availability of healthy food, were associated with reduced diabetes risk, whereas higher density of food stores selling less healthy foods was associated with increased diabetes risk, likely related to behavioral changes due to changes in access to neighborhood resources. However, these studies had limited geography to a handful of urban environments, thus limiting generalizability and prohibiting exploration of how such associations vary across other environments (ie, urban, suburban, or rural settings). The small size of this research body and methodological variability in (1) data sources and definitions of food environment and (2) methods used to operationalize how food environments are measured by urbanicity, as well as insufficient geographical scope in published studies to date, has restricted our ability to understand and characterize the association between the food environment and diabetes across the US. To our knowledge, no study to date has examined associations between neighborhood food environment measures and diabetes incidence using objectively measured food establishment data at the US Census tract level nationwide while stratifying by urbanicity.
The Diabetes Location, Environmental Attributes, and Disparities (Diabetes LEAD) network is a Centers for Disease Control and Prevention-funded collaboration between multiple academic institutions aiming to study the role of community level factors on diabetes incidence. 14 We explored the association of neighborhood food environments, specifically the presence of fast-food establishments and supermarkets, on type 2 diabetes incidence among a cohort of US veterans using the Veteran's Affairs (VA) electronic health record (EHR).

Individual-Level Data
This cohort study was approved by New York University and the VA institutional review boards, which waived the need for informed consent given the retrospective nature of the study and deidentified data, in accordance with 45 CFR §46. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.
Data used for this study are from the US Veterans Administration Diabetes Risk (VADR) cohort, a cohort of veterans without type 2 diabetes constructed by the New York University Grossman School of Medicine and George Mason University through the VA national EHR. 15 The VADR is a national cohort of US veterans enrolled in the VA for primary care. Veterans were passively enrolled  20 These priority groups were used to create a low-income or disability flag that was used as a proxy for socioeconomic status. The low-income or disability flag was categorized hierarchically as with a disability, low-income but without a disability, and none of the above.

Neighborhood-Level Data
All baseline neighborhood-level attributes were defined at cohort entry date. Relative food environment measures were identified as the primary exposures and included (1) the percentage of total food-serving establishments that were fast-food establishments, calculated by taking the 5-year mean number of fast-food restaurants in US Census tracts relative to all restaurants, and (2) the proportion of total retail food outlets that were supermarkets, also calculated by taking the 5-year mean number of supermarkets in US census tracts with assigned buffers relative to other food stores.
Information on how these measures were created have been published elsewhere. 18,21 A sensitivity analysis was done using absolute food environment measures, defined as 5-year mean fast-food restaurants and supermarkets density (count per square kilometer) to test the robustness of our models. All US Census tracts were categorized by the Diabetes LEAD network into 1 of 4 community types using a modified measure of the 2010 Rural-Urban Commuting Area: high-density urban (HDU), low-density urban (LDU), suburban or small town (suburban), and rural. 17 Neighborhood food characteristics were assigned according to the availability of each type of food outlet within buffers created around US Census tracts. Buffer sizes were determined by the Diabetes LEAD network group through a consensus of experts as follows: 1-mile walking buffer for HDU communities, 2-mile driving buffer for LDU communities, 6-mile driving buffer for suburban communities, and 10-mile driving buffer for rural communities. 22 Assigning food environment variables to addresses using each community type's information with different buffers helped standardize these variables and made it feasible to compare findings between different community types.
Other neighborhood-level covariates were generated by the Diabetes LEAD network and included in the models. Neighborhood social and economic environment (NSEE) was created as a community type-stratified z score sum of the following census-derived measures from the American

Statistical Analysis
Neighborhood-level covariates and type 2 diabetes incidence were described in the full sample and also stratified by community type. To estimate the associations of the food environment with type 2 diabetes risk, adjusted hazard ratios (HRs) and 95% CIs were calculated using piecewise exponential (PWE) models with 2-year intervals of person-time and county-level random effects. 23 One-year intervals were used as a sensitivity analysis, and the models yielded similar results (data not shown).
PWE models assume that the hazard of the outcome is constant across intervals, which means that follow-up times follow an exponential distribution within each interval. Two separate models were fitted: the first with the 2 relative food measures and the second with the 2 absolute density food measures. The models also adjusted for individual (age [continuous], sex, race and ethnicity, and disability or low-income flag) and neighborhood-level (NSEE quartiles, land use environment, percentage non-Hispanic Black residents, percentage Hispanic residents, and food environment) covariates. Even though weight is highly correlated with diabetes, we decided not to include it in our models for being on the causal pathway between food and type 2 diabetes. Only individuals with available data on all covariates were included in the PWE models. A sensitivity analysis was done stratifying by region (Northeast, South, Midwest, and West). Models were fitted in the full sample and stratified by community types. Two-tailed t tests were used to assess significance, which was set at a threshold of P < .05. Statistical analyses were done using SAS statistical software version 9.4 (SAS Institute). Data analysis was performed from October 2020 to March 2021.

Results
A total of 4 100 650 individuals were included in the analysis.

Discussion
This cohort study of US veterans is the first, to our knowledge, to prospectively examine the association between neighborhood food environment and type 2 diabetes risk nationally and by community type, using exposure measures tailored to community type. The availability of fast-food restaurants relative to all restaurants was associated with a higher risk of type 2 diabetes in all community types, whereas supermarkets were associated with a lower type 2 diabetes risk in suburban and rural communities. Our models did not find a significant association between supermarket availability and type 2 diabetes incidence in urban communities.
To date, only a limited number of studies 24,25 have examined the association between neighborhood food environment and type 2 diabetes incidence using longitudinal data. Our study found that the association of food environment with type 2 diabetes incidence varied by level of urbanicity but did not vary further by region. Other studies 24,25 focused only on urban communities reported an association between type 2 diabetes incidence and food environment that was different from urban communities strata findings from our cohort. One study, using data from the Multi-Ethnic Study of Atherosclerosis, found that better neighborhood resources was associated with a reduced risk of type 2 diabetes incidence (HR, 0.62; 95% CI, 0.43-0.88), 24 but in this case access to     neighborhood resources combined both physical activity and healthy food establishments. Another study, using data from the Jackson Heart Study, found that higher density of unfavorable food stores was associated with a higher risk of type 2 diabetes incidence (HR, 1.34; 95% CI, 1.12-1.61). 25 This association was larger than the association we found in urban communities. However, the data used were geographically focused in and around a single urban area. 25 Another study 26 using a crosssectional analysis and self-reported food environment found no significant association of food environment with insulin resistance and type 2 diabetes.

JAMA Network Open | Public Health
Unlike other community types, our findings suggest that the relative availability of supermarkets in urban communities was not associated with type 2 diabetes. This could be explained by access to both public transportation and cars, specifically in LDU communities, which could increase the ability to access supermarkets, regardless of availability within the residential neighborhoods. Thus, interventions targeting the placement or zoning of supermarkets may be more appropriate in suburban and rural communities.
In our study, the association between the relative availability of fast-food restaurants and type 2 diabetes incidence was similar in all community types. Results from our sensitivity analysis indicated that the association between the absolute availability of fast-food restaurants and type 2 diabetes incidence was larger in suburban and rural communities compared with LDU communities and was null in HDU communities. Given the high population density in HDU communities, the absolute count of food outlets per kilometer may mirror population density, rather than quality of food environment. In addition, these urban centers have higher socioeconomic status than other areas in HDU communities. Taken together, our findings suggest that policies specific to fast-food restaurants, such as policies restricting the siting of fast-food restaurants and healthy beverage default laws, 27,28 may be effective in reducing type 2 diabetes risk in all community types. In urban areas where population and retail density are growing, it will be even more important to focus on these policies.

Strengths and Limitations
Strengths of this study include examining the association between type 2 diabetes and neighborhood food environment in a large, national, longitudinal cohort using robust statistical methods. PWE models allowed us to test the longitudinal association of food environment with the risk of incident type 2 diabetes, accounting for the multilevel structure of the data and stratifying by different community types. Neighborhood characteristics were assigned using walking and driving Figure buffers around individuals' addresses, the buffer sizes of which were designed to be congruent with community types.
This study has several limitations. The analysis was observational, and the exposure was not randomly assigned to participants. Given that the study used EHR data, we were unable to capture residual lifestyle confounders, such as diet, physical activity, and comorbidities. Follow-up frequency was not constant across all cohort participants. The study may also not be generalizable to nonveteran populations; US veterans have substantially greater financial and health burdens than the civilian population and are at an increased risk of disability, obesity, and other chronic conditions. 29,30 Although most of the cohort was composed of non-Hispanic White men, it did also include a sizable number of women and participants of other races and ethnicities, which were included in our models. We were unable to account for participants' individual household income; however, a low-income or disability flag was used as a proxy for socioeconomic status, and we further adjusted for neighborhood-level socioeconomic factors. Neighborhood characteristics were assigned using patients' baseline address regardless of the possibility of moving. However, it was previously found that even if people move, they tend to live in neighborhoods with similar characteristics. 24 We were unable to ensure whether and how frequently participants were using the stores in their neighborhood. We were also unable to identify those in our cohort who also received care outside the VA and may have been diagnosed with diabetes.

Conclusions
In this study, neighborhood food environment was associated with type 2 diabetes risk among US veterans in multiple community types, suggesting potential avenues for action to address the burden of type 2 diabetes. Tailored interventions targeting availability of supermarkets may be more appropriate in suburban and rural communities than urban communities, whereas restrictions on fast-food restaurants could possibly help in all community types. These actions, combined with increasing awareness of the risk of type 2 diabetes and the importance of healthy diet intake, might be associated with a decrease in the burden of type 2 diabetes among adults in the US.