Abstract
Objective.
To examine relationships between the food environment and obesity by community type.
Methods.
Using electronic health record data from the U.S. Veterans Administration Diabetes Risk (VADR) cohort, we examined associations between the percentage of supermarkets and fast food restaurants with obesity prevalence from 2008-2018. We constructed multivariable logistic regression models with random effects and interaction terms for year and food environment variables. We stratified models by community type.
Results.
Mean age at baseline was 59.8 (SD=16.1) years; 93.3% identified as men; and 2,102,542 (41.8%) were classified as obese. The association between the percentage of fast food restaurants and obesity was positive in high-density urban areas (OR=1.033; 95% CI: 1.028, 1.037), with no interaction by time (p=0.83). The interaction with year was significant in other community types (p’s<0.001), with increasing odds of obesity in each follow-up year. The associations between the percentage of supermarkets and obesity was null in high-density and low-density urban areas, and positive in suburban (OR=1.033; 95% CI: 1.027, 1.039) and rural (OR=1.007; 95% CI: 1.002, 1.012) areas, with no interactions by time.
Conclusions.
Many healthy eating policies have been passed in urban areas; our results suggest such policies might also mitigate obesity risk in non-urban areas.
Keywords: Epidemiology, Food, Health Policy, Obesity, Aging
INTRODUCTION
In 2017–2018, an estimated 42.4% of U.S. adults were obese.(1) Previous work has shown considerable geographic variation in obesity prevalence across the U.S., with higher obesity prevalence among adults in rural areas than among those in urban areas.(2) The prevalence of diabetes, coronary heart disease, and other chronic diseases is also higher in rural versus urban areas.(3, 4) These geographic disparities in obesity risk may be due to differences in individual- and neighborhood-level sociodemographic characteristics across the rural-urban spectrum,(5) as well as poorer diet behaviors among rural residents.(6) The latter is influenced by a number of community-level characteristics, which provide potentially meaningful targets for policy change and intervention.
One such target for policy change is the neighborhood food environment. A large body of work indicates that the proximity of supermarkets to homes is associated with healthier diets and lower risk of obesity,(7, 8) but results are not always consistent.(9) There is also some evidence to suggest that greater access to fast food restaurants is associated with worse diets and higher prevalence of obesity,(10, 11) but, again, some studies report otherwise.(12, 13) The hypothesized pathway is that those shopping at convenience stores and fast food restaurants, where the availability of energy-dense, nutrient-poor food and beverage items is high,(14-16) are likely to make less healthy food purchases. In contrast, supermarkets and wait-service restaurants sell a mix of healthy and less healthy food and beverage items.(17, 18)
The mixed findings in the literature are potentially due to several methodological differences, including differences in the operationalization of food environment measures. For example, most studies quantify food availability using absolute measures,(19) such as the number of supermarkets or fast food restaurants in a given geographic area. Fewer studies use relative measures (e.g., proportion of a type of food outlet), which capture the range of alternative options in the food environment, and may more closely reflect consumers’ food shopping experience compared to absolute measures (e.g., counts).(20) Other studies differ in their definition of areal units and their specification of buffer size, which influences the relationship between the food environment and diet and diabetes outcomes.(21) Furthermore, most of these studies were conducted in small geographic areas, and thus may lack generalizability and have limited statistical power to detect changes in nutrition-related health outcomes. One study examined whether the proximity of supermarkets and fast food restaurants was associated with changes in body mass index (BMI) among 1.7 million veterans across the U.S. between 2009 and 2014, and found no relationship. Yet, this study did not explore differences by community type (e.g., urban, suburban, rural) or associations with obesity prevalence.
There are several reasons why the association between food environment exposures and obesity may differ by community type. A lack of consideration of differences in the range of communities introduces several methodological challenges for food environment studies. Examining differences in the relationship between the food environment and obesity by community type is important because many place-based factors cluster at the community level,(22, 23) and this clustering might differ by community type (e.g., high walkability and healthy food availability may cluster in urban areas but not in rural areas). Some communities may experience little variation in food environment exposures, such as those living in rural areas where supermarkets are scarce,(24) which may result in non-positivity when evaluating these exposures in a statistical model (i.e., not all community types exhibit variation in food environment exposures). It is also possible, for example, that the construct of food availability has a different meaning in urban, suburban, and rural areas (also known as measurement non-invariance, or differential item functioning), which makes it inappropriate to test relationships with health outcomes without taking into account potential effect modification by community type.
Overall, there is a lack of studies accounting for the role of community type, with no national studies to our knowledge assessing the impact of the food environment on obesity prevalence by community type. We sought to address this gap by examining the relationship between the relative availability of fast food restaurants and supermarkets and obesity status over time in a cohort of five million U.S. veterans using the Veteran’s Affairs (VA) electronic health record (EHR), and stratifying by community type using a four-level categorization of community type developed by the Diabetes LEAD Network (i.e., high-density urban, low-density urban, suburban/small town, rural).(21, 25)
METHODS
Data sources
Data used for this study were from the U.S. Veterans Administration Diabetes Risk (VADR) cohort, a national diabetes-free cohort of U.S. veterans enrolled in the VA for primary care, constructed by the NYU Grossman School of Medicine and George Mason University through the VA national EHR.(26) Veterans were passively enrolled into VADR if they were free of type-2 diabetes (T2D) as of January 1, 2008 and had at least two primary care visits at least 30 days apart prior to cohort entry (to establish a cohort of patients who seek VA care regularly). Patients were considered to have T2D and were excluded if they met the following criteria prior to entering the cohort: two encounters with T2D ICD-9/10 codes, a prescription for T2D medication other than metformin or acarbose alone, or one in/outpatient encounter with T2D ICD-9/10 codes and two elevated A1C (≥6.5%). The cohort enrolled 7,044,740 veterans and followed them through December 31, 2018; median follow-up time was 5.5 person-years (IQR 2.6-9.8 years).
Addresses were geocoded using ArcGIS StreetMap Premium.(27) P.O. Box addresses and addresses with missing information (n=667,398) and individuals whose addresses were not located in the continental U.S. (n=63,443) were excluded. Patients with first documented address occurring more than two years after the cohort entry date and with inconsistent clinic visit history were excluded due to ambiguity regarding classification of these addresses as baseline addresses (n=1,577,857). The geo-location information of valid baseline addresses were linked to neighborhood characteristics obtained from the Retail Environment and Cardiovascular Disease (RECVD) study and the decennial Census from 2000 and 2010,(28-30) including exposure variables.
We used food establishment data from the Retail Environment and Cardiovascular Disease (RECVD) study,(28) which classified neighborhood amenities using the National Establishment Time Series (NETS) Database. The NETS data were licensed from Walls & Associates (Walls & Associates, Denver, CO), who prepared annual establishment information collected by Dun and Bradstreet (D&B, Short Hills, NJ). The RECVD team re-geocoded the NETS data to improve locational accuracy and assigned establishments to subcategories using Standard Industrial Classification codes, employee and sales information, and chain names obtained from Technomic/Restaurants and Institutions (R&I) and TDLinx®. Details on classification methods have been described elsewhere.(31)
Outcome
To measure obesity, we first calculated body mass index (BMI), defined as weight in kilograms divided by the square of height in meters at each patient visit. If a patient had more than one patient visit in a given year, we used the average of the two most recent weight measurements per year. Then we classified BMI values into four categories, including underweight (<18.5 kg/m2), normal (18.5 to <25 kg/m2), overweight (25.0 to <30 kg/m2), and obese (≥30.0 kg/m2). Our primary repeated-measures outcome variable was dichotomized into obesity (yes/no). We modeled obesity status because it is a more relevant outcome for clinical practice, and because it is easier to interpret changes in obesity odds as an outcome, than BMI units.
Covariates
Age at baseline was calculated by subtracting the date of birth from cohort entry date. In the EHR, sex was reported as male and female, and marital status as married/living with a partner or single. Race/ethnicity was reported as non-Hispanic American Indian/Alaska Native, Asian, Black, Native Hawaiian and Other Pacific Islander, and white; and Hispanic. Patients in the VA EHR are assigned to different priority groups based on their military service history, disability status, income, and whether or not they qualify for Medicaid or other VA benefits. These priority groups were used to create a proxy for socioeconomic status, categorized hierarchically into three groups: disabled, low income/non-disabled, and none of the above.
Neighborhood socioeconomic environment (NSEE) was defined as the sum of z-scores of six Census variables (% persons with less than a high-school education, % persons unemployed, % of households earning less than $30,000/year, % of households in poverty, % of households on public assistance, and % of households with no cars), modeled on previous work and scaled to be between 0 and 100.(32) Land use environment was defined as a the sum of z-scores of seven indicators (average block length, average block size, intersection density, street connectivity, household density, % developed land, establishment density); an increase in the land use environment variable indicates more compact development.(33) Using American Community Survey (ACS) 5-year data from 2004-2008 to 2012-2016, we also included percentage Hispanic population and percentage non-Hispanic Black population of participants’ Census tracts. To classify community type, we used a measure developed by the Diabetes LEAD Network, which combined Rural Urban Commuting Area codes and land area of participants’ residential census tract to create a four-level variable representing high-density urban, low-density urban, suburban/small town, and rural communities; geographically smaller tracts in metropolitan cores were classified as high-density urban and geographically larger tracts were classified as low-density urban.(25)
Exposures
Our exposures included the relative availability of supermarkets and fast food restaurants around participants’ residences, including a relative measure of the percentage of supermarkets out of total food stores, and the percentage of fast food restaurants out of total restaurants. The supermarkets category included three mutually-exclusive subcategories: supermarkets, supercenters, and medium-sized grocers. Fast food restaurants were defined as quick-service restaurants offering low-preparation-time foods for take-away or cafeterias (no wait service). The total food stores and total restaurants variables include all types of food stores and restaurants, respectively, including the numerators (i.e., supermarkets, fast food restaurants).
We operationalized the exposures by calculating a network buffer around the population-weighted centroid of the census tracts of participants’ home addresses. Street network data was obtained from ArcGIS StreetMap Premium, and network buffers (i.e., using line-based road networks) were created using the “generalized” polygon option and default settings in ArcGIS Pro 2.4.2 and ArcGIS Pro 2.1, respectively. The Diabetes LEAD Network based their buffer distances for network buffers on data from the National Household Food Acquisition and Purchase Survey (FoodAPS),(34) which calculated the average driving distance between participants’ residential addresses and their primary food store. The FoodAPS data also assigns participants to rural (yes/no) and non-metro (yes/no) categories, which align with our four-level community-type variable. Based on the FoodAPS mean distances within rural and non-metro categories, 1-, 2-, 6-, and 10-mile buffer distances were assigned to participants residing in high-density urban, low-density urban, suburban/small town, rural census tracts, respectively in order to account for differences in geographic scale across these community types.
Statistical analysis
We excluded participants with fewer than two primary care visits prior to cohort entry (n=1,333,878), those who did not have a valid address (n=667,398), those who did not reside in the contiguous U.S. (n=1,577,857), those with fewer than two BMI values in the follow-up period (n=1,215,361) and those missing covariate data (n=515,034) and exposure data (n=1,463), for a final sample size of 3,067,627 participants. To examine the association between each food environment exposure and repeated measures of obesity status, we used separate generalized linear mixed models with logistic link and a random effect for county and adjusting for covariates, including age at baseline (i.e., cohort entry date). To capture varying associations with duration of follow-up time, a proxy for time since exposure, we included an interaction term for time (year since baseline) and food environment exposure variables. To assess for differences in estimates by community type and mitigate potential confounding bias, we stratified all models by community type. To maximize interpretability, we scaled our exposure variables so that estimates corresponded to a 20 percentage point increase in the relative availability of supermarkets and fast food restaurants. We used R version 4.2.2 for all statistical analyses. We used GGPREDICT in R to visualize the predicted probability of obesity by food environment exposure over time, including zero, 20%, 40%, 60%, 80%, and 100% supermarket and fast food restaurant availability (separately). For ease of interpretation and comparison of association by community types, we further summarized the predicted probability of obesity (using the coefficients from the regression model) at comparable levels of our primary exposures (i.e., 0% and 100%).
To assess whether variation in the frequency of VA clinic visits across participants influenced the distribution of BMI values, we used a Pearson correlation to test the association between participants’ BMI value and the time (years) between BMI values recorded at clinic visits.
RESULTS
The mean age of participants was 58.5 (SD=16.0) years; 2,844,774 (92.7%) identified as men and 222,853 (7.3%) identified as women (Table 1). About 0.8% of 37, participants were American Indian or Alaska Native; 0.9%, Asian; 16.8%, Black or African American; 5.2%, Hispanic, Latino, or Spanish; 0.8%, Native Hawaiian or other Pacific Islander; and 75.4%, white. A higher percentage of participants in our study sample resided in low-density urban areas (38.0%) compared to high-density urban (n=516,599 (13.0%)), suburban/small town (21.4%), rural (27.6%) areas. A total of 1,316,027 of total participants (42.9%) were classified as obese, with a lower percentage in high-density urban (39.6%) areas relative to low-density urban (42.1%), suburban/small town (44.0%), and rural (44.6%) areas (Table 2). Similarly, we observed a lower percentage of participants classified as overweight or obese in high-density urban areas (77.8%) relative to low-density urban (80.8%), suburban/small town (82.6%), and rural (82.7%) areas.
TABLE 1.
Individual-level and neighborhood-level characteristics by community type
| All community types n/mean (%/SD) |
High density urban n/mean (%/SD) |
Lower density urban n/mean (%/SD) |
Suburban/small town n/mean (%/SD) |
Rural n/mean (%/SD) |
|
|---|---|---|---|---|---|
| 3,067,627 | 397,916 (13.0) | 1,166,343 (38.0) | 655,971 (21.4) | 847,397 (27.6) | |
| Individual-level variables | |||||
| Age | 58.5 (16.0) | 56.4 (16.2) | 58 (16.5) | 58.4 (16.1) | 60.3 (14.9) |
| Age categories | |||||
| 19-39 | 458863 (15.0) | 71103 (17.9) | 191652 (16.4) | 100724 (15.4) | 95384 (11.3) |
| 40-59 | 926665 (30.2) | 139171 (35.0) | 362630 (31.1) | 195593 (29.8) | 229271 (27.1) |
| 60-79 | 1402894 (45.7) | 155387 (39.1) | 495555 (42.5) | 300475 (45.8) | 451477 (53.3) |
| 80+ | 279205 (9.1) | 32255 (8.1) | 116506 (10.0) | 59179 (9.0) | 71265 (8.4) |
| Gender | |||||
| Male | 2844774 (92.7) | 365978 (92.0) | 1069856 (91.7) | 606590 (92.5) | 802350 (94.7) |
| Female | 222853 (7.3) | 31938 (8.0) | 96487 (8.3) | 49381 (7.5) | 45047 (5.3) |
| Race/eth | |||||
| Non-Hispanic white | 2311981 (75.4) | 213043 (53.5) | 819478 (70.3) | 524188 (79.9) | 755272 (89.1) |
| Non-Hispanic black | 516727 (16.8) | 129836 (32.6) | 237672 (20.4) | 89141 (13.6) | 60078 (7.1) |
| Hispanic | 160520 (5.2) | 38822 (9.8) | 76725 (6.6) | 27613 (4.2) | 17360 (2.1) |
| Non-Hispanic Asian | 28153 (0.9) | 9009 (2.3) | 12758 (1.1) | 4660 (0.7) | 1726 (0.2) |
| Non-Hispanic Native Hawaiian/other Pacific Islander | 25133 (0.8) | 4169 (1.1) | 10844 (0.9) | 5022 (0.8) | 5098 (0.6) |
| Non-Hispanic American Indian/Alaska native | 25113 (0.8) | 3037 (0.8) | 8866 (0.8) | 5347 (0.8) | 7863 (0.9) |
| Marital status | |||||
| Married/living with a partner | 1716936 (56.0) | 151943 (38.2) | 621197 (53.3) | 402827 (61.4) | 540969 (63.8) |
| Single | 1350691 (44.0) | 245973 (61.8) | 545146 (46.7) | 253144 (38.6) | 306428 (36.2) |
| Income/disability | |||||
| Disabled | 1178669 (38.4) | 131863 (33.1) | 455371 (39) | 271002 (41.3) | 320433 (37.8) |
| Low income | 1109805 (36.2) | 191060 (48) | 424484 (36.4) | 207113 (31.6) | 287148 (33.9) |
| None of the above | 779153 (25.4) | 74993 (18.9) | 286488 (24.6) | 177856 (27.1) | 239816 (28.3) |
| BMI continuous | 29.8 (5.6) | 29.3 (5.8) | 29.7 (5.6) | 29.9 (5.6) | 30.0 (5.6) |
| BMI | |||||
| Normal weight (BMI<25) | 573061 (18.7) | 88635 (22.3) | 223438 (19.2) | 114367 (17.4) | 146621 (17.3) |
| Overweight (BMI 25 to <30) | 1178539 (38.4) | 151814 (38.2) | 451486 (38.7) | 252678 (38.5) | 322561 (38.1) |
| Obese (BMI≥30) | 1316027 (42.9) | 157467 (39.6) | 491419 (42.1) | 288926 (44.1) | 378215 (44.6) |
| Prevalence diabetes | 1036641 (33.8) | 136935 (34.4) | 389098 (33.4) | 214383 (32.7) | 296225 (35.0) |
| Neighborhood level variables | |||||
| Relative fast-food restaurants | 0.301 (0.128) | 0.256 (0.135) | 0.314 (0.122) | 0.325 (0.098) | 0.286 (0.145) |
| Relative supermarkets | 0.109 (0.072) | 0.091 (0.075) | 0.103 (0.070) | 0.11 (0.051) | 0.125 (0.081) |
| NSEE continuous | 16.947 (9.838) | 24.809 (13.226) | 14.187 (8.873) | 14.362 (8.46) | 19.054 (7.439) |
| NSEE quartiles | |||||
| 1st quartile (most advantaged) | 609503 (19.9) | 90581 (22.8) | 195318 (16.8) | 125576 (19.1) | 198028 (23.4) |
| 2nd quartile | 815284 (26.6) | 109105 (27.4) | 300479 (25.8) | 178543 (27.2) | 227157 (26.8) |
| 3rd quartile | 883823 (28.8) | 104424 (26.2) | 351187 (30.1) | 193959 (29.6) | 234253 (27.6) |
| 4th quartile (least advantaged) | 759017 (24.7) | 93806 (23.6) | 319359 (27.4) | 157893 (24.1) | 187959 (22.2) |
| Land use environment | 0.06 (0.910) | 0.004 (0.810) | 0.08 (0.906) | −0.006 (0.962) | 0.108 (0.916) |
| Percent Hispanic | 0.103 (0.161) | 0.197 (0.228) | 0.126 (0.168) | 0.08 (0.132) | 0.046 (0.094) |
| Percent NH black | 0.136 (0.221) | 0.269 (0.323) | 0.16 (0.231) | 0.101 (0.165) | 0.069 (0.137) |
NSEE=neighborhood socioeconomic environment
TABLE 2.
Obesity status by individual-level and neighborhood-level characteristics by community type
| All community types | High density urban | Lower density urban | Suburban/small town | Rural | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| n | % (95% CI) | n | % (95% CI) | n | % (95% CI) | n | % (95% CI) | n | % (95% CI) | |
| 1,316,027 | 42.9 (42.8, 43.0) | 157,467 | 39.6 (39.4, 39.7) | 491,419 | 42.1 (42, 42.2) | 288,926 | 44 (43.9, 44.2) | 378,215 | 44.6 (44.5, 44.7) | |
| Individual-level variables | ||||||||||
| Age categories | ||||||||||
| 19-39 | 182,781 | 39.8 (39.7, 40) | 25,338 | 35.6 (35.3, 36) | 75,787 | 39.5 (39.3, 39.8) | 41,788 | 41.5 (41.2, 41.8) | 39,868 | 41.8 (41.5, 42.1) |
| 40-59 | 449,247 | 48.5 (48.4, 48.6) | 61,727 | 44.4 (44.1, 44.6) | 174,178 | 48 (47.9, 48.2) | 98,310 | 50.3 (50, 50.5) | 115,032 | 50.2 (50, 50.4) |
| 60-79 | 618,693 | 44.1 (44, 44.2) | 63,185 | 40.7 (40.4, 40.9) | 215,209 | 43.4 (43.3, 43.6) | 134,683 | 44.8 (44.6, 45) | 205,616 | 45.5 (45.4, 45.7) |
| 80+ | 65,306 | 23.4 (23.2, 23.5) | 7,217 | 22.4 (21.9, 22.8) | 26,245 | 22.5 (22.3, 22.8) | 14,145 | 23.9 (23.6, 24.2) | 17,699 | 24.8 (24.5, 25.2) |
| Gender | ||||||||||
| Male | 1,222,415 | 43 (42.9, 43) | 144,484 | 39.5 (39.3, 39.6) | 451,055 | 42.2 (42.1, 42.3) | 268,137 | 44.2 (44.1, 44.3) | 358,739 | 44.7 (44.6, 44.8) |
| Female | 93,612 | 42 (41.8, 42.2) | 12,983 | 40.7 (40.1, 41.2) | 40,364 | 41.8 (41.5, 42.1) | 20,789 | 42.1 (41.7, 42.5) | 19,476 | 43.2 (42.8, 43.7) |
| Race/eth | ||||||||||
| Non-Hispanic white | 979,120 | 42.3 (42.3, 42.4) | 83,034 | 39 (38.8, 39.2) | 335,224 | 40.9 (40.8, 41) | 225,846 | 43.1 (43, 43.2) | 335,016 | 44.4 (44.2, 44.5) |
| Non-Hispanic black | 231,578 | 44.8 (44.7, 45) | 52,506 | 40.4 (40.2, 40.7) | 107,907 | 45.4 (45.2, 45.6) | 43,277 | 48.5 (48.2, 48.9) | 27,888 | 46.4 (46, 46.8) |
| Hispanic | 75,130 | 46.8 (46.6, 47) | 16,873 | 43.5 (43, 44) | 36,092 | 47 (46.7, 47.4) | 13,662 | 49.5 (48.9, 50.1) | 8,503 | 49 (48.2, 49.7) |
| Non-Hispanic Asian | 7,008 | 24.9 (24.4, 25.4) | 1,971 | 21.9 (21, 22.7) | 3,193 | 25 (24.3, 25.8) | 1,292 | 27.7 (26.4, 29) | 552 | 32 (29.8, 34.2) |
| Non-Hispanic Native Hawaiian/other Pacific Islander | 11,176 | 44.5 (43.9, 45.1) | 1,710 | 41 (39.5, 42.5) | 4,837 | 44.6 (43.7, 45.5) | 2,275 | 45.3 (43.9, 46.7) | 2,354 | 46.2 (44.8, 47.5) |
| Non-Hispanic American Indian/Alaska native | 12,015 | 47.8 (47.2, 48.5) | 1,373 | 45.2 (43.4, 47) | 4,166 | 47 (45.9, 48) | 2,574 | 48.1 (46.8, 49.5) | 3,902 | 49.6 (48.5, 50.7) |
| Marital status | ||||||||||
| Married/living with a partner | 782,175 | 45.6 (45.5, 45.6) | 65,760 | 43.3 (43, 43.5) | 277,749 | 44.7 (44.6, 44.8) | 186,107 | 46.2 (46, 46.4) | 252,559 | 46.7 (46.6, 46.8) |
| Single | 533,852 | 39.5 (39.4, 39.6) | 91,707 | 37.3 (37.1, 37.5) | 213,670 | 39.2 (39.1, 39.3) | 102,819 | 40.6 (40.4, 40.8) | 125,656 | 41 (40.8, 41.2) |
| Income/disability | ||||||||||
| Disabled | 558,333 | 47.4 (47.3, 47.5) | 57,243 | 43.4 (43.1, 43.7) | 212,158 | 46.6 (46.4, 46.7) | 130,956 | 48.3 (48.1, 48.5) | 157,976 | 49.3 (49.1, 49.5) |
| Low income | 439,936 | 39.6 (39.5, 39.7) | 71,056 | 37.2 (37, 37.4) | 166,235 | 39.2 (39, 39.3) | 84,644 | 40.9 (40.7, 41.1) | 118,001 | 41.1 (40.9, 41.3) |
| None of the above | 317,758 | 40.8 (40.7, 40.9) | 29,168 | 38.9 (38.5, 39.2) | 113,026 | 39.5 (39.3, 39.6) | 73,326 | 41.2 (41, 41.5) | 102,238 | 42.6 (42.4, 42.8) |
| Prevalence diabetes | ||||||||||
| Diabetes | 608,144 | 58.7 (58.6, 58.8) | 75,168 | 54.9 (54.6, 55.2) | 224,422 | 57.7 (57.5, 57.8) | 128,696 | 60 (59.8, 60.2) | 179,858 | 60.7 (60.5, 60.9) |
| No diabetes | 707,883 | 34.9 (34.8, 34.9) | 82,299 | 31.5 (31.4, 31.7) | 266,997 | 34.4 (34.2, 34.5) | 160,230 | 36.3 (36.1, 36.4) | 198,357 | 36 (35.9, 36.1) |
| Neighborhood level | ||||||||||
| NSEE quartiles | ||||||||||
| 1st quartile (most advantaged) | 257,104 | 42.2 (42.1, 42.3) | 35,200 | 38.9 (38.5, 39.2) | 79,307 | 40.6 (40.4, 40.8) | 53,948 | 43 (42.7, 43.2) | 88,649 | 44.8 (44.5, 45) |
| 2nd quartile | 351,431 | 43.1 (43, 43.2) | 43,920 | 40.3 (40, 40.5) | 126,462 | 42.1 (41.9, 42.3) | 78,621 | 44 (43.8, 44.3) | 102,428 | 45.1 (44.9, 45.3) |
| 3rd quartile | 383,745 | 43.4 (43.3, 43.5) | 41,971 | 40.2 (39.9, 40.5) | 150,908 | 43 (42.8, 43.1) | 86,554 | 44.6 (44.4, 44.8) | 104,312 | 44.5 (44.3, 44.7) |
| 4th quartile (least advantaged) | 323,747 | 42.7 (42.5, 42.8) | 36,376 | 38.8 (38.5, 39.1) | 134,742 | 42.2 (42, 42.4) | 69,803 | 44.2 (44, 44.5) | 82,826 | 44.1 (43.8, 44.3) |
NSEE=neighborhood socioeconomic environment
The relative availability of fast food restaurants and supermarkets in the total sample was 30.1% (12.8%) and 10.9% (7.2%), respectively (Table S1), with no notable differences by community type (Figures S1, S2). We observed a positive association between the relative availability of fast food restaurants and odds of obesity in all community types, including high-density urban (OR=1.033; 95% CI: 1.028, 1.037), low-density urban (OR=1.041; 95% CI: 1.038, 1.044), suburban/small town (OR=1.033; 95% CI: 1.027, 1.039), and rural (OR=1.010; 95% CI: 1.007, 1.013) areas (Table S1). The interaction between the relative availability of fast food restaurants and time was not significant in high-density urban areas (p=0.83), which is reflected by the lack of divergence of observed associations over time in Figure 1. In contrast, the interaction with year was statistically significant in the other community types (p’s<0.001), with an increase in the odds of obesity in each subsequent year of follow-up. For example, the predicted probability of obesity for participants residing in suburban/small towns with no fast food restaurants (β=0.472; 95% CI: 0.468, 0.477) and those residing in suburban/small towns with 20% fast food restaurant availability within the community-specific buffer (β=0.475; 95% CI: 0.471, 0.478) were not different at baseline; whereas, we observed statistically significantly different predicted probabilities between no fast food restaurants (β=0.456; 95% CI: 0.452, 0.463) and 20% fast food restaurant availability (β=0.476; 95% CI: 0.472, 0.480) at the end of the follow-up period (Table S2).
Figure 1. Model-based associationsa of the relative availability of fast food restaurants and obesityb by year and community type.
NOTE: p-value reflect statistical significance (α=0.05) of interaction term for time (year since baseline) and food environment exposure variables.
aAdjusting for baseline age, sex, race/ethnicity, marital status, income/disability flag, land use environment, neighborhood socioeconomic environment quartiles, percentage Hispanic population and percent non-Hispanic Black population of participants’ Census tracts; and interaction with year since baseline (i.e., cohort entry date).
bObesity defined as body mass index of >30.0 kg/m2.
In models without an interaction term for time, we observed a null association between the relative availability of supermarkets and odds of obesity in high-density urban and low-density urban areas (Table S1), whereas we observed a weak association between the relative availability of supermarkets and odds of obesity in suburban/small (OR=1.033; 95% CI: 1.027, 1.039) and rural (OR=1.007; 95% CI: 1.002, 1.012) areas. The interaction between the relative availability of supermarkets and time was not significant in any community type, indicating that null and weak associations between the relative availability of supermarkets and obesity did not differ over time, which is demonstrated graphically by the close overlap in confidence intervals of the associations depicted in Figure 2.
Figure 2. Model-based associationsa of the relative availability of supermarkets and obesityb by year and community type.
NOTE: p-value reflect statistical significance (α=0.05) of interaction term for time (year since baseline) and food environment exposure variables.
aAdjusting for baseline age, sex, race/ethnicity, marital status, income/disability flag, land use environment, neighborhood socioeconomic environment quartiles, percentage Hispanic population and percent non-Hispanic Black population of participants’ Census tracts; and interaction with year since baseline (i.e., cohort entry date).
bObesity defined as body mass index of >30.0 kg/m2.
The Pearson correlation coefficient for the association between participants’ BMI value and the time (years) between BMI values recorded at clinic visits was −0.02, which indicates a very weak correlation, suggesting frequency of clinic visits did not bias model results.
DISCUSSION
Using 10 years of objective height and weight data from a cohort of over five million U.S. veterans, we found that the relative availability of supermarkets had a weak or no association with odds of obesity across community types, with no significant variation in the magnitude of the associations over time. In contrast, our results suggest that higher relative availability of fast food restaurants was associated with higher odds of obesity in all community types, and, with the exception of veterans in high-density urban areas, the strength of this association increased over time with odds of obesity growing in each subsequent year of follow up, suggesting accumulating effects over time. Taken together, these results suggest that we tailor our obesity prevention strategies to food outlet type, potentially by reducing the availability of fast food restaurants relative to other types of restaurants (e.g., zoning restrictions) and/or supporting healthy choices within fast food restaurants (e.g., added sugar warning labels). The latter may be more feasible, though, especially in areas where changes to public infrastructure are less politically and economically viable. The results also suggest we tailor our approach to community type, given how we see increases in the odds of obesity in low-density urban, suburban/small, and rural areas over time.
There are several potential explanations for the observed effect modification by community type, which may inform intervention and policy solutions. First, many different domains of area-level measures cluster at the community level (i.e., multidimensionality),(32) and the clustering of area-level measures may differ by community type.(32) For example, access to public transportation and restaurant density are typically higher in high-density urban areas relative to other community types, and greater access to public transportation may promote greater access to a mix of restaurant types in high-density urban areas. Furthermore, clustering of area-level measures and the way clustering influences individuals’ diet-related behaviors may change over time, with potential differences across community type. Recent work, for example, shows how the relative availability of sit-down restaurants grew over time in urban areas compared to suburban areas in the Midwest region,(35) which may differentially mitigate risk of weight gain from eating out in fast food restaurants in urban areas. It is also possible that veterans living in non-urban areas are less likely than their peers in urban areas to leave their local food environment for alternative options due to a lack of willingness or ability to drive, or other reasons (e.g., time savings); and these differences in food purchasing behaviors may widen over time, especially given how our cohort of veterans is older and has greater health burdens than the civilian population.(36, 37)
Our findings may also be driven by how food environment measures have different meanings in different community types (a form of measurement non-invariance, or differential item functioning).(38, 39) Distributional differences in how veterans access fast food restaurants (e.g., drive-through versus eat-in), the frequency with which veterans visit these establishments, the types of fast food restaurants themselves (e.g., fast causal versus quick service), and the items the fast food restaurants sell (e.g., beverages and desserts with high levels of added sugars) may all inform the differences we observed by community type; and, similarly, temporal changes in the distributional differences in these factors across community types may differentially influence obesity risk over time. For example, it’s possible that the veterans in our sample visited fast food restaurants more frequently over time (relative to their cohort entry date), reflecting nationwide increases in food-away-from-home expenditures during this period.(40) Evidence also suggests that restaurants have shifted to selling less healthy food and beverage items over time.(15, 41)
We recently published a longitudinal study of the association between the neighborhood food environment and T2D incidence using data from this cohort of veterans,(42) and we found that the relative availability of supermarkets was associated with lower T2D risk in suburban and rural communities, and the relative availability of fast food restaurants was associated with higher risk of T2D in all community types. These results approximately mirror the results for odds of obesity in the current study, though the previous study did not assess for potential interactions with time, which may have masked meaningful temporal changes in diabetes risk across community types over the follow-up period. Our study is also similar to an analysis of veterans in the Weight and Veterans’ Environments Study (WAVES), a retrospective longitudinal cohort study of those who received health care services from the VA between 2009 and 2014. The authors examined the relationship between the relative accessibility of supermarkets – defined as the percentage of supermarkets out of the number of supermarkets and fast food restaurants – and BMI using a model with person fixed effects, and found no meaningful association.(9) It’s difficult to make a direct comparison to our findings, though, given the differences in the outcome (BMI vs. obesity status) and differences in modeling approaches. The inclusion of person fixed effects, while a robust strategy for mitigating time-invariant confounding bias, makes comparison especially challenging given how person-level changes in the food environment are only due to either migration to a new address or food outlet openings and closings, which does not reflect associations for non-movers and those for whom the food environment does not change.
Our study had several limitations. The study design was observational, and we lacked key individual-level data due to the use of EHR data, including dietary consumption behaviors and household income. We did, however, have access to objective measures of height and weight, and we used a hierarchical variable of low income and/or disability status as a proxy for socioeconomic status. We lacked data on in-store offerings (e.g., type, prices) or what, if anything, participants purchased at their neighborhood food establishments. We also did not examine the impact of other aspects of the neighborhood built environment on obesity. There was variation in the frequency of follow-up time among cohort participants, though our analyses suggested that the frequency of clinic visits did not bias model results. And we linked neighborhood exposures and covariates to participants’ baseline addresses, which overlooks potential changes due to migration. Furthermore, our results may not be generalizable to non-veteran populations, which differ from veterans across a spectrum of characteristics, including key differences in financial, health, and social factors.(36, 37) Our primary strength, however, included our consideration of community type, which is unique relative to other national studies of the food environment and weight-related outcomes. We also leveraged a large sample size and time-varying data, with millions of participant observations, including a sizable number of women and non-white participants.
CONCLUSION
In this study, we found that the relative availability of supermarkets had a weak or no association with odds of obesity, whereas higher relative availability of fast food restaurants was associated with higher odds of obesity in all community types, with an increase in the odds over 10 years of follow-up in low-density urban, suburban/small town, and rural areas. To date, many healthy eating policies (e.g., healthy checkout policies, sugary drink taxes) have passed in urban areas, though our results suggest that such policies might also mitigate obesity risk among veterans residing in non-urban areas. Our results also suggest that initiatives to increase access to supermarkets may not be effective in reducing obesity risk in any community type, and other retail strategies may be needed to change diet-related behaviors (e.g., financial incentives for fruits and vegetables). In the future, it will be important to replicate our results in a different study sample, given the unique demographic characteristics of veterans.
Supplementary Material
STUDY IMPORTANCE.
The relationship between the food environment and obesity is not clear, in part due to a lack of studies stratifying by community type (e.g., urban, suburban, rural). This is important to understand potential moderation by place-based factors, potentially due to differences in the meaning of food availability across community type.
The relative availability of supermarkets had a weak or no association with odds of obesity, whereas higher relative availability of fast food restaurants was associated with higher odds of obesity in all community types, with an increase in the odds over 10 years of follow-up in low-density urban, suburban/small town, and rural areas.
Many healthy eating policies have been enacted in urban areas, though our results suggest such policies might also mitigate obesity risk in non-urban areas, especially those focused on restaurant settings.
FUNDING:
This study was supported by the Center for Disease Control and Prevention (5 U01DP006299-02-00; PI: LET) and the National Institute of Diabetes and Digestive and Kidney Diseases (R01DK124400).
Footnotes
DISCLOSURE: The authors declared no conflict of interest
REFERENCES
- 1.Hales CM, Carroll MD, Fryar CD, Ogden CL. Prevalence of obesity among adults and youth: United States, 2015–2016. 2017. [PubMed] [Google Scholar]
- 2.Lundeen EA, Park S, Pan L, O'Toole T, Matthews K, Blanck HM. Obesity Prevalence Among Adults Living in Metropolitan and Nonmetropolitan Counties - United States, 2016. MMWR Morbidity and mortality weekly report. 2018;67(23):653–8. Epub 2018/06/15. doi: 10.15585/mmwr.mm6723a1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Abrams LR, Myrskylä M, Mehta NK. The growing rural-urban divide in US life expectancy: contribution of cardiovascular disease and other major causes of death. International journal of epidemiology. 2022;50(6):1970–8. Epub 2022/01/10. doi: 10.1093/ije/dyab158. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Cosby AG, McDoom-Echebiri MM, James W, Khandekar H, Brown W, Hanna HL. Growth and Persistence of Place-Based Mortality in the United States: The Rural Mortality Penalty. American journal of public health. 2019;109(1):155–62. Epub 2018/11/30. doi: 10.2105/ajph.2018.304787. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.McAlexander TP, Malla G, Uddin J, Lee DC, Schwartz BS, Rolka DB, et al. Urban and rural differences in new onset type 2 diabetes: Comparisons across national and regional samples in the diabetes LEAD network. SSM - population health. 2022;19:101161. Epub 2022/08/23. doi: 10.1016/j.ssmph.2022.101161. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Trivedi T, Liu J, Probst J, Merchant A, Jhones S, Martin AB. Obesity and obesity-related behaviors among rural and urban adults in the USA. Rural and remote health. 2015;15(4):3267. Epub 2015/10/16. [PubMed] [Google Scholar]
- 7.Cobb LK, Appel LJ, Franco M, Jones-Smith JC, Nur A, Anderson CA. The relationship of the local food environment with obesity: A systematic review of methods, study quality, and results. Obesity (Silver Spring, Md). 2015;23(7):1331–44. Epub 2015/06/23. doi: 10.1002/oby.21118. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Gao X, Engeda J, Moore LV, Auchincloss AH, Moore K, Mujahid MS. Longitudinal associations between objective and perceived healthy food environment and diet: The Multi-Ethnic Study of Atherosclerosis. Social science & medicine (1982). 2022;292:114542. Epub 2021/11/23. doi: 10.1016/j.socscimed.2021.114542. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Zenk SN, Tarlov E, Wing C, Matthews SA, Jones K, Tong H, et al. Geographic Accessibility Of Food Outlets Not Associated With Body Mass Index Change Among Veterans, 2009-14. Health affairs (Project Hope). 2017;36(8):1433–42. Epub 2017/08/09. doi: 10.1377/hlthaff.2017.0122. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Bahadoran Z, Mirmiran P, Azizi F. Fast Food Pattern and Cardiometabolic Disorders: A Review of Current Studies. Health promotion perspectives. 2015;5(4):231–40. Epub 2015/01/01. doi: 10.15171/hpp.2015.028. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.van Erpecum CL, van Zon SKR, Bültmann U, Smidt N. The association between the presence of fast-food outlets and BMI: the role of neighbourhood socio-economic status, healthy food outlets, and dietary factors. BMC public health. 2022;22(1):1432. Epub 2022/07/28. doi: 10.1186/s12889-022-13826-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Harbers MC, Beulens JWJ, Boer JM, Karssenberg D, Mackenbach JD, Rutters F, et al. Residential exposure to fast-food restaurants and its association with diet quality, overweight and obesity in the Netherlands: a cross-sectional analysis in the EPIC-NL cohort. Nutrition journal. 2021;20(1):56. Epub 2021/06/18. doi: 10.1186/s12937-021-00713-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Hobbs M, Green M, Roberts K, Griffiths C, McKenna J. Reconsidering the relationship between fast-food outlets, area-level deprivation, diet quality and body mass index: an exploratory structural equation modelling approach. Journal of epidemiology and community health. 2019;73(9):861–6. Epub 2019/06/07. doi: 10.1136/jech-2018-211798. [DOI] [PubMed] [Google Scholar]
- 14.Alexander E, Rutkow L, Gudzune KA, Cohen JE, McGinty EE. Trends in the healthiness of U.S. fast food meals, 2008-2017. European journal of clinical nutrition. 2021;75(5):775–81. Epub 2020/10/29. doi: 10.1038/s41430-020-00788-z. [DOI] [PubMed] [Google Scholar]
- 15.McCrory MA, Harbaugh AG, Appeadu S, Roberts SB. Fast-Food Offerings in the United States in 1986, 1991, and 2016 Show Large Increases in Food Variety, Portion Size, Dietary Energy, and Selected Micronutrients. Journal of the Academy of Nutrition and Dietetics. 2019;119(6):923–33. Epub 2019/03/04. doi: 10.1016/j.jand.2018.12.004. [DOI] [PubMed] [Google Scholar]
- 16.McKerchar C, Smith M, Gage R, Williman J, Abel G, Lacey C, et al. Kids in a Candy Store: An Objective Analysis of Children's Interactions with Food in Convenience Stores. Nutrients. 2020;12(7). Epub 2020/07/28. doi: 10.3390/nu12072143. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.An R. Fast-food and full-service restaurant consumption and daily energy and nutrient intakes in US adults. European journal of clinical nutrition. 2016;70(1):97–103. Epub 2015/07/02. doi: 10.1038/ejcn.2015.104. [DOI] [PubMed] [Google Scholar]
- 18.Banks J, Fitzgibbon ML, Schiffer LA, Campbell RT, Antonic MA, Braunschweig CL, et al. Relationship Between Grocery Shopping Frequency and Home- and Individual-Level Diet Quality Among Low-Income Racial or Ethnic Minority Households With Preschool-Aged Children. Journal of the Academy of Nutrition and Dietetics. 2020;120(10):1706–14.e1. Epub 2020/08/24. doi: 10.1016/j.jand.2020.06.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Bivoltsis A, Cervigni E, Trapp G, Knuiman M, Hooper P, Ambrosini GL. Food environments and dietary intakes among adults: does the type of spatial exposure measurement matter? A systematic review. International journal of health geographics. 2018;17(1):19. Epub 2018/06/11. doi: 10.1186/s12942-018-0139-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Luan H, Law J, Quick M. Identifying food deserts and swamps based on relative healthy food access: a spatio-temporal Bayesian approach. International journal of health geographics. 2015;14:37. Epub 2015/12/31. doi: 10.1186/s12942-015-0030-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Rummo PE, Algur Y, McAlexander T, Judd SE, Lopez PM, Adhikari S, et al. Comparing competing geospatial measures to capture the relationship between the neighborhood food environment and diet. Annals of epidemiology. 2021;61:1–7. Epub 2021/05/30. doi: 10.1016/j.annepidem.2021.05.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Arthur KN, Knutsen SF, Spencer-Hwang R, Shavlik D, Montgomery S. Health-Predictive Social-Environmental Stressors and Social Buffers Are Place Based: A Multilevel Example From San Bernardino Communities. Journal of primary care & community health. 2019;10:2150132719835627. Epub 2019/03/22. doi: 10.1177/2150132719835627. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Chen M, Creger T, Howard V, Judd SE, Harrington KF, Fontaine KR. Association of community food environment and obesity among US adults: a geographical information system analysis. Journal of epidemiology and community health. 2019;73(2):148–55. Epub 2018/11/07. doi: 10.1136/jech-2018-210838. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Lenardson JD, Hansen AY, Hartley D. Rural and Remote Food Environments and Obesity. Current obesity reports. 2015;4(1):46–53. Epub 2015/12/03. doi: 10.1007/s13679-014-0136-5. [DOI] [PubMed] [Google Scholar]
- 25.McAlexander TP, Algur Y, Schwartz BS, Rummo PE, Lee DC, Siegel KR, et al. Categorizing community type for epidemiologic evaluation of community factors and chronic disease across the United States. Social sciences & humanities open. 2022;5(1). Epub 2022/04/05. doi: 10.1016/j.ssaho.2022.100250. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Avramovic S, Alemi F, Kanchi R, Lopez PM, Hayes RB, Thorpe LE, et al. US veterans administration diabetes risk (VADR) national cohort: cohort profile. BMJ open. 2020;10(12):e039489. Epub 2020/12/06. doi: 10.1136/bmjopen-2020-039489. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.ArcGIS by Esri. https://www.esri.com/en-us/home.
- 28.Drexel University Urban Health Collaborative. The Retail Environment and Cardiovascular Disease (RECVD) Project. https://sites.google.com/view/recvd-team-project-site. [Google Scholar]
- 29.American Community Survey (ACS). https://www.census.gov/programs-surveys/acs. [Google Scholar]
- 30.Hirsch AG, Carson AP, Lee NL, McAlexander T, Mercado C, Siegel K, et al. The Diabetes Location, Environmental Attributes, and Disparities Network: Protocol for Nested Case Control and Cohort Studies, Rationale, and Baseline Characteristics. JMIR research protocols. 2020;9(10):e21377. Epub 2020/10/20. doi: 10.2196/21377. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Hirsch JA, Moore KA, Cahill J, Quinn J, Zhao Y, Bayer FJ, et al. Business Data Categorization and Refinement for Application in Longitudinal Neighborhood Health Research: a Methodology. Journal of urban health : bulletin of the New York Academy of Medicine. 2021;98(2):271–84. Epub 2020/10/03. doi: 10.1007/s11524-020-00482-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Messer LC, Laraia BA, Kaufman JS, Eyster J, Holzman C, Culhane J, et al. The development of a standardized neighborhood deprivation index. Journal of urban health : bulletin of the New York Academy of Medicine. 2006;83(6):1041–62. Epub 2006/10/13. doi: 10.1007/s11524-006-9094-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Meeker MA, Schwartz BS, Bandeen-Roche K, Hirsch AG, De Silva SSA, McAlexander TP, et al. Assessing Measurement Invariance of a Land Use Environment Construct Across Levels of Urbanicity. GeoHealth. 2022;6(10):e2022GH000667. Epub 2022/10/21. doi: 10.1029/2022gh000667. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.National Household Food Acquisition and Purchase Survey (FoodAPS). Economic Research Service (ERS), U.S. Department of Agriculture (USDA). https://www.ers.usda.gov/foodaps. [Google Scholar]
- 35.Peng K, Rodriguez DA, Hirsch JA, Gordon-Larsen P. A method for estimating neighborhood characterization in studies of the association with availability of sit-down restaurants and supermarkets. International journal of health geographics. 2021;20(1):15. Epub 2021/03/27. doi: 10.1186/s12942-020-00257-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Liu Y, Sayam S, Shao X, Wang K, Zheng S, Li Y, et al. Prevalence of and Trends in Diabetes Among Veterans, United States, 2005-2014. Preventing chronic disease. 2017;14:E135. Epub 2017/12/15. doi: 10.5888/pcd14.170230. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Schult TM, Schmunk SK, Marzolf JR, Mohr DC. The Health Status of Veteran Employees Compared to Civilian Employees in Veterans Health Administration. Military medicine. 2019;184(7-8):e218–e24. Epub 2019/02/23. doi: 10.1093/milmed/usy410. [DOI] [PubMed] [Google Scholar]
- 38.Jones RN. Differential item functioning and its relevance to epidemiology. Current epidemiology reports. 2019;6:174–83. Epub 2019/12/17. doi: 10.1007/s40471-019-00194-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Sapawi R, Said I. Constructing indices representing physical attributes for walking in urban neighborhood area. Procedia-Social Behavioral Sciences. 2012;50:179–91. [Google Scholar]
- 40.Okrent AM, Elitzak H, Park T, Rehkamp S. Measuring the value of the US food system: revisions to the food expenditure series. 2018. [Google Scholar]
- 41.Frelier JM, Moran AJ, Vercammen KA, Jarlenski MP, Bleich SN. Trends in Calories and Nutrients of Beverages in U.S. Chain Restaurants, 2012-2017. American Journal of Preventive Medicine. 2019;57(2):231–40. doi: 10.1016/j.amepre.2019.03.023. [DOI] [PubMed] [Google Scholar]
- 42.Kanchi R, Lopez P, Rummo PE, Lee DC, Adhikari S, Schwartz MD, et al. Longitudinal Analysis of Neighborhood Food Environment and Diabetes Risk in the Veterans Administration Diabetes Risk Cohort. JAMA network open. 2021;4(10):e2130789. Epub 2021/10/30. doi: 10.1001/jamanetworkopen.2021.30789. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.