Supermarket Proximity and Risk of Hypertension, Diabetes, and CKD: A Retrospective Cohort Study

Abstract

Rationale & Objective:

Living in environments with low access to food may increase the risk of chronic diseases. We investigated the association of household distance to the nearest supermarket (as a measure of food access) with the incidence of hypertension, diabetes, and chronic kidney disease (CKD) in a metropolitan area of the United States.

Study Design:

Retrospective cohort study.

Setting & Participants:

777,994 individuals without hypertension, diabetes, or CKD at baseline within the HealthLNK Data Repository, which contains electronic health records from 7 health care institutions in Chicago, Illinois.

Exposure:

Zip code–level average distance between households and nearest supermarket.

Outcome:

Incidence of hypertension, diabetes, and CKD based on presence of ICD-9 code and/or blood pressure ≥ 140/90 mm Hg, hemoglobin A1c ≥ 6.5%, and eGFR < 60 mL/min/1.73m², respectively.

Analytical Approach:

Average distance to nearest supermarket was aggregated from street-level metrics for 56 Chicagoland zip codes. The cumulative incidence of hypertension, diabetes, and CKD from 2007–2012 was calculated for each zip code in patients free of these diseases in 2006. Spatial analysis of food access and disease incidence was performed using bivariate local indicator of spatial association (BiLISA) maps and bivariate local Moran I statistics. The relationship between supermarket access and outcomes was analyzed using logistic regression.

Results:

Of 777,994 participants, 408,608 developed hypertension, 51,380 developed diabetes, and 56,365 developed CKD. There was significant spatial overlap between average distance to supermarket and incidence of hypertension and diabetes but not CKD. Zip codes with large average supermarket distances and high incidence of hypertension and diabetes were clustered in southern and western neighborhoods. Models adjusted only for neighborhood factors (zip code–level racial composition, access to vehicles, median income) revealed significant associations between zip code–level average distance to supermarket and chronic disease incidence. Relative to tertile 1 (shortest distance), ORs in tertiles 2 and 3, respectively, were 1.27 (95% CI, 1.23–1.30) and 1.38 (95% CI, 1.33–1.43) for diabetes, 1.03 (95% CI, 1.02–1.05) and 1.04 (95% CI, 1.02–1.06) for hypertension, and 1.18 (95% CI, 1.15–1.21) and 1.33 (95% CI, 1.29–1.37) for CKD. Models adjusted for demographic factors and health insurance showed significant and positive association with greater odds of incident diabetes (tertile 2: 1.29 [95% CI, 1.26–1.33]; tertile 3: 1.35 [95% CI, 1.31–1.39]) but lesser odds of hypertension (tertile 2: 0.95 [95% CI, 0.94–0.97]; tertile 3: 0.91 [95% CI, 0.89–0.92]) and CKD (tertile 2: 0.80 [95% CI, 0.78–0.82]; tertile 3: 0.73 [95% CI, 0.72–0.76]). After adjusting for both neighborhood and individual covariates, supermarket distance remained significantly associated with greater odds of diabetes and lesser odds of hypertension, but there was no significant association with CKD.

Limitations:

Unmeasured neighborhood and social confounding variables, zip code–level analysis, and limited individual-level information.

Conclusions:

There are significant disparities in supermarket proximity and incidence of hypertension, diabetes, and CKD in Chicago, Illinois. The relationship between supermarket access and chronic disease is largely explained by individual- and neighborhood-level factors.

Hypertension, diabetes, and chronic kidney disease (CKD) are leading contributors to excess health care costs, morbidity, and premature mortality in the United States.^1–3 Substantial research has explored the complex biological and sociocontextual factors that contribute to these diseases and their associated clinical outcomes. Important factors like nutrition and dietary patterns interplay with income, education, neighborhood racial composition, neighborhood economic investment, and other social factors to influence disease development.^4–12

To explore the connections between diet and chronic disease, efforts are underway to understand the effects of the built environment on access to healthy, nutritious food. The US Department of Agriculture (USDA) defines low food access urban areas as census tracts in which at least 500 people or 33% of the population live greater than one-half mile from a supermarket, super center, or large grocery store.¹³ These communities tend to have lower levels of income and education, higher proportions of Black and Latinx/Hispanic residents, and greater access to independent grocery stores and convenience stores than to chain and nonchain supermarkets. Relative to supermarkets, independent grocery stores often have higher prices, fewer and lower quality fruits and vegetables, and more energy-dense food options.^14–17 Years of living in communities characterized by deprivation of resources and nutrition may increase residents’ risk of obesity, cardiovascular disease, and malignancy, and may reduce their life expectancy.^18–20 However, few studies have investigated nutritious food access and incident chronic illness using electronic health data.

Merging electronic health record data from health centers across the city of Chicago with zip code–level grid mapping of supermarket access, we investigated associations between neighborhood food environments and the risk of incident hypertension, diabetes, and CKD. We hypothesized that zip codes with larger average distances to the nearest supermarket would have higher incidence of hypertension, diabetes, and CKD compared with zip codes with shorter average distances to the nearest supermarket.

Methods

Data Acquisition

We used data from the HealthLNK Data Repository (HDR), which contains electronic health records from 7 health care institutions in Chicago, Illinois: 4 large academic medical centers (Northwestern Medicine, Rush University Medical Center, University of Illinois at Chicago Medical Center, and University of Chicago Hospitals and Clinic), 1 large academic safety net health care system (Cook County Health and Hospitals System), and a network of community health centers with multiple care outlets (the Alliance of Chicago). The HDR uses a secure, Health Insurance Portability and Accountability Act–compliant, Secure Hash Algorithm-512 hashing algorithm and a probability-weighted, deterministic matching algorithm to create deduplicated, deidentified patient-level data linked across health care sites. The HDR includes data on patient demographics, insurance status, and zip code (collected at time of most recent visit), and longitudinal clinical data on approximately 5 million unique patients aged 18–89 years old residing in Chicago and its surrounding suburbs. Additional details regarding the creation of the HDR have been published.²¹ Institutional review board approval (no. STU00053183) was obtained at each participating site. Because deidentified data was used for the analysis, the study was conducted under a waiver of informed consent. The HDR currently contains records for encounters in 2006–2016. Our study used all data from 2006 to 2012 except for those from the Alliance of Chicago, which were unavailable at the time of our data acquisition.

From the HDR, we obtained individual-level demographic data from 2006–2007 including age, sex, race, ethnicity, insurance status (categorized by the most recent insurance type as Medicare, Medicaid, private insurance, self-pay, no charge, and other), and zip code of residence. We also retrieved the total number of patients in the HDR who resided in each zip code. We extracted data on hemoglobin A1c, systolic blood pressure (SBP), diastolic blood pressure (DBP), height, weight, body mass index (BMI), and estimated glomerular filtration rate (eGFR) from medical records from 2006–2012. <monospace>International Classification of Diseases, Ninth Revision</monospace> (ICD-9) codes from 2006–2012 were also obtained. The complete HDR data request is shown in Table S1. For each zip code, we obtained median household income.²² Neighborhood racial and ethnic composition and access to a vehicle were obtained from 2010 US census data.

Quantification of Supermarket Access

The primary measure of food accessibility was the average distance to the nearest supermarket. A supermarket was defined as a full-service store that (1) carried a diverse line of groceries including fresh produce, fresh meats, and a deli in addition to packaged and dry goods; and (2) contained 5 or more checkout lanes. The supermarket classification methodology was based on prior systematic surveys of all available food stores in Chicagoland communities in 2007 and 2011.^23,24 In a subsequent study, we calculated the distance in feet between a residence’s street address and the street address of the nearest validated supermarket, including supermarkets in adjacent census tracts.^25,26 We then reaggregated the street-level distances to calculate the average distance to the nearest supermarket for each Chicago zip code, using zip code as opposed to census tract to align with the HDR data.

Disease Determination and Incidence Calculation

Hypertension was defined as having a single systolic blood pressure recording greater than 140 mm Hg, a single diastolic blood pressure recording greater than 90 mm Hg, or having an ICD-9 code of 401.xx, 402.xx, 403.xx, or 404.xx. Diabetes was defined as having any hemoglobin A1c ≥ 6.5% or having an ICD-9 code of 250.xx. CKD was defined by a single eGFR of <60 mL/min/1.73 m² (calculated using the 2009 CKD-EPI creatinine equation²⁷) or an ICD-9 code of 585.xx or 586.00. A full list of the requested ICD-9 codes can be found in Table S1. To limit our analysis to incident cases of disease, we removed all participants from the individual analysis if they met criteria for hypertension, diabetes, or CKD at any point during 2006 and assessed new disease onset between 2007 and 2012.

Spatial and Statistical Analysis

We mapped the average distance to the nearest supermarket in 2007 for each Chicago zip code. We calculated means, SD, and percentages to describe participant baseline characteristics by tertiles of zip code–level supermarket distance. We calculated the cumulative incidence of hypertension, diabetes, and CKD by zip code from 2007 to 2012.

To investigate spatial relationships between food proximity and disease incidence, we constructed bivariate local indicator of spatial association (BiLISA) maps based on bivariate local Moran I statistics for average zip code distance to nearest supermarket and incident hypertension, diabetes, and CKD.²⁸ The local Moran I statistic assessed whether a positive or negative spatial association between supermarket proximity and incidence of hypertension, diabetes, or CKD exists, with a value greater than 0 reflecting that zip codes with large or small average distances to the nearest supermarket were neighbored by zip codes with a similarly large or small incidence of hypertension, diabetes, or CKD. We defined a neighboring zip code as one that shares a border or vertex.

The BiLISA maps produced a visual analysis of zip code clustering based on the local Moran I statistic and determined whether the clustering was statistically significant. Statistically significant zip code clustering was established at a level of P < 0.05, based on 999 permutations of randomly reshuffling zip code supermarket access and zip code incidence of hypertension, diabetes, and CKD. This was based on the rationale that some clustering is possible with complete spatial randomness, but statistical significance occurs when clustering is greater than 95% of clustering found in 999 permutations of random zip code shuffling.²⁸

The BiLISA maps identified statistically significant, “high-high” clustered zip codes in which zip codes with longer average distances to the nearest supermarket were neighbored by zip codes with a higher average incidence of disease. “Low-low” regions, conversely, were statistically significant clusters of zip codes with shorter distances to the nearest supermarket and a lower average incidence of disease. “High-low” outlier zip codes had a larger average supermarket distance and is bordered by zip codes with low incidence of the disease. By contrast, “low-high” outlier zip codes had a shorter average supermarket distance but is surrounded by zip codes with greater incidence of disease. To determine whether the differences in supermarket proximity and disease incidence between the “high-high” and other zip codes were statistically significant, we performed an analysis of variance and established statistical significance using P < 0.05. Spatial statistical analysis and mapping were performed in the open source software GeoDa.²⁹

We determined the association between supermarket access and incidence of hypertension, diabetes, and CKD using logistic regression in which we adjusted for individual-level variables (race, ethnicity, age, sex, and insurance status) and zip code–level variables (median income, racial and ethnic composition, and vehicle access). Incident hypertension and diabetes were further adjusted in the CKD model. We modeled supermarket access as a continuous distance and in tertiles of distance, with the lowest tertile serving as the reference group. All analyses were performed using Stata 16 (StataCorp). A 2-tailed P < 0.05 was considered statistically significant.

Results

Participant Characteristics

The study included a total population of 777,994 unique participants obtained from the HDR who were free of hypertension, diabetes, and CKD at baseline (2006) and resided in 56 Chicago zip codes. We used the average distance to a supermarket in each zip code to categorize the zip codes into tertiles (with tertile 1 having the shortest distance); zip codes in tertiles 2 and 3 had average supermarket distances of 0.73 miles and 1.23 miles, respectively, which met the USDA criteria for low-access urban environments. Table 1 shows the baseline characteristics overall and by supermarket distance tertile. Compared with individuals in tertile 1 zip codes, a greater proportion of individuals in the upper tertiles were female and had Medicare or Medicaid. In terms of race and ethnicity, across the tertiles there was a relatively equal distribution of participants who were Hispanic, most of those in the highest tertiles were Black, and the majority of those in tertile 1 were White. Across tertiles, access to vehicles was relatively equal.

Table 1.

Participant Characteristics by Tertile of Supermarket Distance

	All Participants (N= 777,994)	Tertile 1 (n = 268,786)	Tertile 2 (n = 263,373)	Tertile 3 (n = 245,835)
Distance to supermarket, miles	0.79 ± 0.37	0.44 ± 0.99	0.73 ± 0.11	1.23 ± 0.29
Individual Characteristics
Age, y	47.02 ± 16.77	45.35 ± 15.57	47.52 ± 16.99	48.32 ± 1 7.64
Male sex	42.30%	44.65%	42.76%	39.07%
Racea
White	37.71%	58.41%	34.28%	18.77%
Black	37.70%	10.87%	39.37%	65.26%
Hispanic	7.90%	9.51%	8.91 %	5.04%
Asian	3.21%	4.45%	3.02%	2.05%
Other	1.08%	1.49%	0.97%	0.74%
Unknown	12.40%	15.27%	13.45%	8.13%
Ethnicityb
Hispanic	14.66%	16.73%	16.42%	10.51 %
Non-Hispanic	85.34%	83.27%	83.58%	89.49%
Insurance status
Medicaid or Medicare	24.50%	14.35%	25.66%	33.28%
Private insurance	43.70%	54.84%	39.89%	35.61 %
No insurance	31.80%	30.81%	34.45%	31.11%
Neighborhood Characteristics
Median incomec
<$30,000	11.19%	0	7.73%	27.14%
$30,000–50,000	45.73%	1 7.99%	66.67%	53.62%
$50,000–75,000	28.24%	45.83%	18.70%	19.24%
>$75,000	14.85%	36.18%	6.91 %	0
Racial and ethnic compositiond
White	49.18%	85.38%	46.77%	12.18%
Black	34.46%	0	31.31%	75.50%
Hispanic	15.96%	22.15%	18.45%	6.55%
No vehicle accessd	28.19%	28.80%	27.23%	28.55%

Distance refers to the average distance between a household and the nearest supermarket for each zip code.

^aOptions in the HealthLNK Data Repository race field were American Indian/Alaskan Native, Asian, Black or African American, Hispanic or Latino, Native Hawaiian or Other Pacific Islander, White, Declined, Other.

^bOptions in the HealthLNK Data Repository ethnicity field were Hispanic/Latino or non-Hispanic/Latino.

^cObtained from Michigan Population Studies Center, expressed in US dollars.²²

^dObtained from the 2010 American Community Survey.

Spatial Analysis

The map in Figure 1 displays the zip code boundaries, with shading indicating the tertiles of zip code–average distance to the nearest supermarket in 2007. Zip codes in tertile 1 (the shortest distance to supermarkets) had an average distance of 0.27 to 0.60 miles and were in the northern parts of Chicago. Areas with the lowest access, in some places almost 2 miles to the nearest supermarket, were located primarily in southern and western neighborhoods.

Figure 1.

Map of Chicago zip codes with shading showing tertile of average distance to the nearest supermarket. Zip codes with longer supermarket distance have darker shading. Abbreviation: Dist, distance.

Food Access and Hypertension

The bivariate Moran I was significant and demonstrated a positive spatial autocorrelation for low food access and hypertension, indicating zip codes with large average distances to the nearest supermarket had neighboring zip codes with high average incidence of hypertension (Moran I = 0.45; z value = 6.21; pseudo P value = 0.001). Of the 56 zip codes, 10 had a significantly elevated incidence of hypertension and large average distances to the supermarket, noted as “high-high” zip codes (Fig 2). Table 2 shows that, compared with the other zip codes, the “high-high” zip codes had longer average distances to the nearest supermarket (1.26 ± 0.39 miles vs 0.73 ± 0.33 miles; P < 0.001) and a significantly higher incidence of hypertension (45 ± 4 vs 32 ± 6 per 100 participants; P < 0.001).

Figure 2.

Spatial analysis of supermarket proximity and disease incidence. The upper maps show the geographic distribution of incident hypertension, diabetes, and CKD by Chicago zip codes from 2007–2012. The lower maps show BiLISA cluster maps. High-high zip code clusters reflect zip codes with longer average supermarket distance and higher average incidence of disease. Low-low clusters are those with shorter average supermarket distance and lower average incidence of disease. High-low clusters have longer supermarket distance and lower disease incidence, while low-high clusters indicate the opposite: shorter supermarket distance and higher incidence of disease. Abbreviations: BiLISA, bivariate local indicator of spatial association; CKD, chronic kidney disease.

Table 2.

Differences in Supermarket Distance and Disease Incidence Between “High-High” Zip Codes and Other Zip Codes

	Distance in 2007 to Nearest Supermarket, Miles	Disease Incidence 2007–2012, per 100 Participants
Hypertension
High-high zip codes (n = 10)	1.26 ± 0.39	45 ± 4
Other zip codes (n = 46)	0.73 ± 0.33	32 ± 6
P value	<0.001	<0.001
Diabetes
High-high zip code (n = 9)	1.31 ± 0.40	10±1
Other zip codes (n = 47)	0.73 ± 0.33	6 ± 3
P value	<0.001	<0.001
CKD
High-high zip code (n = 2)	1.17 ± 0.4	11 ± 0
Other zip codes (n = 54)	0.81 ± 0.4	7 ± 2
P value	0.2	0.01

Values given as mean ± SD. Based on analysis of variance, which determined if differences in supermarket proximity and disease incidence were statistically significant between zip codes in “high-high” clusters compared with other zip codes. High-high zip code clusters reflect zip codes with farther average supermarket distance. Analysis performed in GeoDa.

Abbreviation: CKD, chronic kidney disease.

There were 12 “low-low” zip codes in which low incidence of hypertension and short distances to supermarket overlapped. The “low-low” zip codes were clustered in northern parts of the city whereas “high-high” zip codes were concentrated in the southern and western parts (Fig 2).

Food Access and Diabetes

Similar to the hypertension analysis, a significant positive spatial autocorrelation indicated clustering of zip codes with large average supermarket distances and a high incidence of diabetes (Moran I = 0.51; z value = 6.21; pseudo P value = 0.001). Of the 56 zip codes, 9 had a significantly elevated incidence of diabetes and large average distances to the nearest supermarket, noted as “high-high” zip codes (Fig 2). Compared with the other zip codes, the “high-high” zip codes had longer average distances to the nearest supermarket (1.31 ± 0.40 miles vs 0.73 ± 0.33 miles; P < 0.001) and a higher incidence of diabetes (10 ± 1 vs 6 ± 3 per 100 participants; P < 0.001) (Table 2). By contrast, there were 16 “low-low” zip codes in which short average supermarket distances and a low incidence of diabetes overlapped. Similar to hypertension, the “low-low” zip codes were concentrated in the northern parts of the city whereas “high-high” zip codes were clustered in the southern and western neighborhoods (Fig 2).

Food Access and CKD

There was no spatial overlap between supermarket access and incidence of CKD (Moran I =−0.04; z value =−0.57; pseudo P value = 0.3). There were 2 “high-high” zip codes (Fig 2), which had long average supermarket distances and a high incidence of CKD. The “high-high” zip codes had a significantly higher incidence of CKD compared with the other zip codes (11 ± 0 vs 7 ± 2 per 100 participants; P = 0.01) but no significant difference in distance to nearest supermarket (1.17 ± 0.4 miles vs 0.81 ± 0.4 miles; P = 0.2) (Table 2). Conversely, there were 5 “low-low” zip codes (Fig 2).

Average Supermarket Distance and Odds of Incident Disease

Out of 777,994 participants without hypertension, diabetes, or CKD at baseline (2006), 408,608 developed hypertension, 51,380 developed diabetes, and 56,365 developed CKD during 2007–2012.

In the unadjusted model (Table 3), compared with tertile 1 the zip codes in tertiles 2 and 3, respectively, were associated with higher incidence of hypertension (odds ratios [ORs] of 1.22 [95% CI, 1.20–1.23] and 1.49 [95% CI, 1.47–1.50]), prediabetes (ORs of 2.27 [95% CI, 2.03–2.52] and 3.14 [95% CI, 2.83–3.48]), and diabetes (ORs of 1.94 [95% CI, 1.89–1.99] and 2.55 [95% CI, 2.49–2.61]). Zip codes in tertile 2 were associated with a lower incidence of CKD (OR, 0.95 [95% CI, 0.93–0.97]), but for those in tertile 3 there was no association with CKD incidence (OR, 1.00 [95% CI, 0.98–1.02]).

Table 3.

Association of Supermarket Distance With Incidence of Hypertension, Diabetes, and CKD

Outcome	Unadjusted Model	Adjusted
		Model 1	Model 2	Model 3
Hypertension
Distance, per 0.5-mile greater	1.23 (1.22–1.24)a	1.02 (1.01–1.024)b	0.97 (0.96–0.97)a	0.96 (0.95–0.97)a
Categorical
Distance tertile 1	1.00 (reference)	1.00 (reference)	1.00 (reference)	1.00 (reference)
Distance tertile 2	1.22 (1.20–1.23)a	1.03 (1.02–1.05)a	0.95 (0.94–0.97)a	0.93 (0.92–0.95)a
Distance tertile 3	1.49 (1.47–1.50)a	1.04 (1.02–1.06)a	0.91 (0.89–0.92)a	0.87 (0.86–0.89)a
Prediabetes
Distance, per 0.5-mile greater	1.53 (1.47–1.59)a	1.13 (1.07–1.20)a	1.08 (1.02–1.13)b	1.08 (1.02–1.15)b
Categorical
Distance tertile 1	1.00 (reference)	1.00 (reference)	1.00 (reference)	1.00 (reference)
Distance tertile 2	2.27 (2.03–2.52)a	1.51 (1.33–1.71)a	1.33 (1.18–1.50)a	1.18 (1.04–1.34)b
Distance tertile 3	3.14 (2.83–3.48)a	1.78 (1.55–2.05)a	1.37 (1.21–1.55)a	1.30 (1.13–1.49)a
Diabetes
Distance, per 0.5-mile greater	1.48 (1.47–1.49)	1.12 (1.09–1.13)a	1.13 (1.12–1.14)a	1.06 (1.05–1.08)a
Categorical
Distance tertile 1	1.00 (reference)	1.00 (reference)	1.00 (reference)	1.00 (reference)
Distance tertile 2	1.94 (1.89–1.99)a	1.27 (1.23–1.30)a	1.29 (1.26–1.33)a	1.05 (1.02–1.09)b
Distance tertile 3	2.55 (2.49–2.61)a	1.38 (1.33–1.43)a	1.35 (1.31–1.39)a	1.09 (1.05–1.13)a
CKD
Distance, per 0.5-mile greater	0.95 (0.94–0.96)b	1.04 (1.03–1.06)b	0.83 (0.82–0.84)b	0.95 (0.94–0.98)b
Categorical
Distance tertile 1	1.00 (reference)	1.00 (reference)	1.00 (reference)	1.00 (reference)
Distance tertile 2	0.95 (0.93–0.97)b	1.18 (1.15–1.21)b	0.80 (0.78–0.82)b	0.98 (0.96–1.01)
Distance tertile 3	1.00 (0.98–1.02)	1.33 (1.29–1.37)b	0.73 (0.72–0.76)b	1.00 (0.97–1.04)

Logistic regression analysis; values shown are OR (95% CI). Model 1 includes neighborhood factors (ie, zip code-level racial and ethnic composition, vehicle access, median income), Model 2 includes individual factors (ie, race, ethnicity, age in 2012, sex, insurance status), Model 3 includes neighborhood and individual factors. For the CKD outcome, model 3 is further adjusted for status of hypertension and diabetes. Abbreviations: CKD, chronic kidney disease; OR, odds ratio.

^aP < 0.001.

^bP < 0.01.

Adjusting for neighborhood factors (zip code–level median household income, racial and ethnic composition, and vehicle access; model 1 of Table 3) revealed significant associations between supermarket distance tertile and incidence of hypertension (tertile 2: OR, 1.03 [95% CI, 1.02–1.05]; tertile 3: OR, 1.04 [95% CI, 1.02–1.06]), prediabetes (tertile 2: OR, 1.51 [95% CI, 1.33–1.71]; tertile 3: OR, 1.78 [95% CI, 1.55–2.05]), diabetes (tertile 2: OR, 1.27 [95% CI, 1.23–1.30]; tertile 3: OR, 1.38 [95% CI, 1.33–1.43]), and CKD (tertile 2: OR, 1.18 [95% CI, 1.15–1.21]; tertile 3: OR, 1.33 [95% CI, 1.29–1.37]). Similar results were observed when distance to the nearest supermarket was included as a continuous variable.

By contrast, models adjusted solely for individual factors (age, sex, race, ethnicity, and health insurance; model 2 of Table 3) showed significant associations of tertiles 2 and 3 with greater odds of prediabetes and diabetes but lesser odds of hypertension and CKD (prediabetes: OR, 1.33 [95% CI, 1.18–1.50] and 1.37 [95% CI, 1.21–1.55]; diabetes: OR, 1.29 [95% CI, 1.26–1.33] and 1.35 [95% CI, 1.31–1.39]; hypertension: OR, 0.95 [95% CI, 0.94–0.97] and 0.91 [95% CI, 0.89–0.92]; and CKD: OR, 0.80 [95% CI, 0.78–0.82] and 0.73 [95% CI, 0.72–0.76]).

After adjusting for both neighborhood- and individual-level covariates (model 3 in Table 3), tertiles 2 and 3 of supermarket distance were associated with a higher incidence of diabetes (OR, 1.05 [95% CI, 1.02–1.09] and 1.09 [95% CI, 1.05–1.13]). Conversely, zip codes in the second and third tertiles of supermarket distance were associated with a lower incidence of hypertension (OR, 0.93 [95% CI, 0.92–0.95] and 0.87 [95% CI, 0.86–0.89]). For tertiles 2 and 3, there was no significant association with incidence of CKD (OR, 0.98 [95% CI, 0.96–1.01] and 1.00 [95% CI, 0.97–1.04]). There was a statistically significant association of incident hypertension (OR, 4.05 [95% CI, 3.94–4.16]; P < 0.001) and incident diabetes (OR, 3.13 [95% CI, 3.05–3.21]; P < 0.001) with incident CKD.

Discussion

In this retrospective cohort study, we found substantial disparities in food access paralleled by disparities in rates of hypertension, diabetes, and CKD across Chicago zip codes. These findings were mirrored by notable residential socioeconomic and sociodemographic trends. The initial spatial analysis revealed clustering of zip codes in northern Chicago neighborhoods with shorter average distances to the nearest supermarket and lower incidence of hypertension and diabetes. These zip codes comprised predominantly White, affluent residents. Conversely, zip codes in southern and western Chicago neighborhoods were associated with almost 3 times farther average distances to the nearest supermarket and a higher incidence of hypertension, diabetes, and CKD. Individuals residing in these zip codes tended to be low income, Black, and female. These findings corroborate existing literature on the neighborhood characteristics of urban environments with low food access.^{17,23,30–33}

We found associations between farther supermarket distance and incidence of hypertension, diabetes, and CKD in models adjusted for neighborhood factors. However, after accounting for numerous individual and neighborhood level covariates, the associations were greatly attenuated for diabetes, became negative with for hypertension, and were no longer significant for CKD.

Although supermarket proximity may be an independent predictor of incident diabetes, for diseases such as hypertension and CKD it may also be a proxy measure for individual-level variables (eg, race, sex) and neighborhood-level variables (eg, neighborhood segregation, income, wealth, and education). Indeed, in the models that adjusted for only individual-level factors and models that adjusted for both individual and neighborhood factors, there was a high degree of collinearity, with a variance inflation factor of >10 for the individual-level factors. Our findings are consistent with the existing literature that has shown individual and neighborhood factors accounting for associations between chronic disease outcomes and food access, especially diabetes, hypertension, and late-stage CKD.^34–39

Our research adds to the existing literature regarding the potential role of the built environment in health outcomes. Prior analyses using complex neighborhood-level deprivation indexes including a variety of factors such as neighborhood-level education, employment, income, and racial composition have shown clear associations between the built environment and disease outcomes, even after accounting for individual-level factors.^40–43 The vast majority of individuals residing in such environments are racial and ethnic minorities from low-income backgrounds. This demographic makeup of neighborhoods in the United States reflects, in part, racialized policy influences such as historical redlining policies and present day housing and lending discrimination^44–46 rather than the residential preferences of racial and ethnic minorities themselves.^47–50 As a clear demonstration of structural racism, which has been well described in neighborhood-health literature, residential sociodemographic characteristics, such as the racial makeup of a neighborhood, often determine the availability of health-promoting resources and infrastructure.

As neighborhood characteristics such as supermarket proximity are modifiable risk factors in disease development, initiatives to increase supermarket access have been proposed but have yielded mixed results. In some studies, enhanced supermarket access in urban environments increased awareness of healthy food options and consumption of fruits and vegetables by minority groups, reduced participation in SNAP (Supplemental Nutrition Assistance Program, ie, government nutrition assistance), reduced levels of obesity and high cholesterol, provided a safe environment for positive social interaction and social support among shoppers, and was a potential source for neighborhood investment.^51–55 Other studies reported more modest effects of increasing supermarket access alone, including concerns that despite providing better access and reduced prices, supermarket introduction did not increase healthy food and vegetable intake, did not alter BMI, and increased exposure to unhealthy items.^54,56,57

The mixed results yielded by supermarket research underscores the complexity of the problem at hand. Significant reduction in health disparities may require a greater breadth and depth of policies and interventions to improve more than just physical access to supermarkets, but also critical health-influencing social contexts like housing, education, capital investment, and job creation.

The strengths of our study include merging large-scale electronic health data from 777,994 participants in the HDR with information about supermarket access in Chicago based on a published method. The use of a diverse cohort of electronic patient data offers an alternative method for studying disease than traditional survey-based approaches. The use of visual maps provides greater context to the data and raises additional questions on why such geographic segregation of access, opportunity, and disease exists.

Our study also has limitations. We were not able to capture many complex neighborhood and social dynamics involving poverty, social cohesion, housing, education, gentrification, and discriminatory public policies that may also affect health outcomes. Examining a specific neighborhood characteristic like food access oversimplifies this larger context.^58,59 We were also unable to capture important individual-level information such as individual income, access to automobiles or public transportation, purchasing power, food purchasing patterns, or information regarding health outcomes for family members. Zip code–level income, vehicle access, and racial composition may not be sufficiently granular surrogates.

The data source for our study was a limited dataset, and we did not have individual-level residence address to calculate the distance between each individual residence and the nearest supermarket. Instead, we used the average distance from households to the nearest supermarket for each zip code. The study method assumes that participants reside in one zip code, solely shop for food within their zip code of residence, and are not traveling to other zip codes with more abundant options. We are also limited by not knowing specific nutrition-related behaviors, including the particular items participants purchase or consume.

Although we assessed medical history, using a 1-year lag is short to establish that a participant did not have hypertension, diabetes, or CKD before 2007 when calculating the incidence of these diseases in 2007–2012.⁶⁰ Similarly, though the HDR covers many large hospital systems in Chicago, it is possible that participants may have been diagnosed with diabetes, hypertension, or CKD at other health care institutions before 2007. Although we have made sure the participants were free of hypertension, diabetes, and CKD based on ICD codes or laboratory values, baseline blood pressure, eGFR, and HbA1C values were not available for the majority of our participants. We may be overestimating the rates of hypertension and CKD by using just 1 measurement of SBP >140 mm Hg, DBP >90 mm Hg, and eGFR < 60 mL/min/1.73 m² to define these diseases. Finally, causality cannot be established, and these results may not be generalizable to other communities or neighborhoods.

In conclusion, we observed notable disparities in rates of hypertension, diabetes, and CKD across Chicago zip codes. Except for diabetes, the association of supermarket proximity with hypertension and CKD was largely explained by individual- and neighborhood-level factors. Additional studies using neighborhood deprivation indexes and hierarchical models may further elucidate the respective individual- and neighborhood-level contribution to disease outcomes.

PLAIN-LANGUAGE SUMMARY.

Residing in an environment with limited access to healthy food may be a modifiable risk factor for the development of hypertension, diabetes, and kidney disease. We used a household’s distance to the closest supermarket as a measure of food access and calculated the average distance to the nearest supermarket for 56 Chicago zip codes. We also calculated the incidence of hypertension, diabetes, and kidney disease for each of those zip codes using electronic health data. We found that supermarket distance is a confounding marker for a number of other important individual factors like race, sex, and age and neighborhood factors like income, racial composition, and access to a vehicle. Changing disparities in chronic diseases will require broad, multifaceted solutions.