Beyond Access: Characteristics of the Food Environment and Risk of Diabetes

Abstract

Characteristics of the built environment, including access to unhealthy food outlets, are hypothesized to contribute to type 2 diabetes mellitus (T2D). Swedish nationwide registry data on 4,718,583 adults aged 35–80 years living in 9,353 neighborhoods, each with at least 1 food outlet, were geocoded and linked to commercial registers (e.g., restaurants and grocery stores). Multilevel logistic regression was used to examine the prospective relationship between characteristics of the food environment and T2D from 2005 to 2010. Relative access to health-harming food outlets was associated with greater likelihood of both prevalent and incident T2D in a curvilinear manner, with the highest risk being observed for environments in which one-third of outlets were health-harming. Relative to individuals whose food environment did not change, those who moved into areas with more health-harming food outlets had higher odds of developing T2D (odds ratio = 3.67, 95% confidence interval: 2.14, 6.30). Among those who did not move, living in an area that gained relative access to health-harming food outlets was also associated with higher odds of T2D (odds ratio = 1.72, 95% confidence interval: 1.27, 2.33). These results suggest that local food environment, including changes that result in greater access to unhealthy food outlets, is associated with T2D.

The incidence of type 2 diabetes mellitus (T2D) has increased substantially in the past 30 years, and T2D currently affects more than 170 million adults worldwide (1). During this period, socioeconomic disparities in T2D have widened, in terms of both total prevalence and clinical detection and treatment (2). While these disparities are not due to any single factor, characteristics of the built environment (e.g., access to healthy food, green space, and neighborhood walkability) are hypothesized to contribute to social disparities in obesity-related health conditions, including T2D (3–5). Several studies have found an association between access to recreational areas or access to healthy food outlets and T2D onset or risk factors (e.g., obesity or insulin resistance) (6–10). However, evidence that characteristics of the food environment explain socioeconomic disparities in T2D is mixed (11). In this study, we aimed to clarify the relationship between characteristics of the food environment and T2D in a nationwide sample.

The processes that link aspects of the food environment to socioeconomic disparities in T2D risk are complex, and it is likely that the relevance of any particular process is influenced by context (12). For example, proponents of the “deprivation amplification” hypothesis argue that environmental or community factors that harm health, or at least do not promote health, act synergistically to compound socioeconomic disadvantage and contribute to health disparities (13). However, data are mixed as to whether individuals in low-socioeconomic status (SES) areas actually have less access to resources than their affluent neighbors (14). For example, in a nationwide study in Sweden, Kawakami et al. (15) reported that more deprived areas had higher availability of both health-promoting (e.g., physical activity centers and health-care facilities) and health-harming (e.g., fast-food outlets, bars, or pubs) resources compared with areas of lower deprivation. This indicates that focusing solely on availability of health-harming food outlets is insufficient because it does not adequately index the net food environment.

Related to the notion of deprivation amplification, “food deserts” are generally defined as populated areas with poor access to nutritious and affordable food (16). In a recent systematic review, Beaulac et al. (17) found evidence to support the idea of food deserts in the United States but mixed evidence for other developed economies. For example, in a study of fast-food restaurants in New Orleans, Louisiana, Block et al. (18) found that fast-food restaurants are geographically associated with predominately black and low-income neighborhoods. However, studies from several different European nations have found that all types of food outlets (health-promoting and health-harming) are more common in deprived areas (15, 19) or that there is a nonlinear relationship between neighborhood deprivation and access to food outlets (20).

The goals of this study were to examine the relationship between characteristics of the food environment and risk of T2D and to investigate whether differential access to health-harming food outlets contributes to socioeconomic disparities in T2D. We also sought to clarify whether the observed associations between the food environment and T2D risk reflect processes related to causation (e.g., lack of access to healthy food outlets increases T2D risk) or selection (e.g., individuals who have elevated risk of T2D select into unhealthy food environments) (21), and thus we also examined whether the relationship between characteristics of the food environment and T2D risk differed as a function of residential mobility. We examined these research questions using a prospective study design and data drawn from nationwide registries in Sweden.

METHODS

Sample

Data came from a linkage of several nationwide registries performed at the Center for Primary Health Care Research at Lund University in Malmö, Sweden (22, 23). Information on demographic characteristics (e.g., age, sex, educational attainment, and household income) was derived from census data. All residential and commercial addresses in Sweden have been geocoded to small geographic units that have boundaries defined by homogeneous types of buildings. These neighborhood areas, called small area market statistics (SAMS), have an average of 1,000 people (2,000 in the Stockholm area) and were used as a proxy for neighborhoods (24). Finally, data from the nationwide Swedish Prescribed Drug Register, which records all medication prescribed and dispensed, was used to index T2D status (25). These databases were linked using the individual national identification number that is assigned to each person in Sweden for their lifetime, which was replaced by a serial number to anonymize the data of individual participants.

The analytical sample was restricted to adults aged 35–80 years in 2005 who lived in SAMS with a minimum of 200 residents and where adjacent houses are no more than 200 m apart (9,353 SAMS). This restriction was made because estimates of neighborhood deprivation and the food environment are less reliable in SAMS in small, sparsely populated areas. The analytical population size in 2005 was 4,718,583 (including 365,222 individuals with prevalent T2D). The sample size for specific analyses varies because each analysis was limited to individuals who lived in a SAMS that had at least 1 of the types of food outlets being analyzed (e.g., individuals living in a SAMS with no grocery stores were excluded from the analysis of that food outlet).

This study was approved by the Institutional Review Board at Lund University.

Measures

Food environment

We measured several different characteristics of the food environment using geographic information systems. Each characteristic was indexed both as a count of the number of each type of outlet and as density (number of outlets per km²), and the characteristics were examined as both continuous variables and categorical variables (5 quantiles). For the continuous formulation of each characteristic, we included a squared term as indicated by visual inspection of the relationship. We indexed number and density of fast-food restaurants and convenience stores, sit-down restaurants, and food stores (supermarkets, grocery stores, and farmers markets), each assessed within a 1,000-m buffer of each person. These categories were derived from a classification made by a business information company (Bisnode, formerly known as Teleadress, Solna, Sweden). Results obtained using these 2 indicators of access (number and density) were similar, and thus we present only the results derived from use of the density measure. Densities of these indicators were highly correlated: for fast-food outlets and restaurants, r² = 0.915 (P < 0.001); for fast-food outlets and grocery stores, r² = 0.942 (P < 0.001); and for grocery stores and restaurants, r² = 0.924 (P < 0.001).

To quantify the “net-negative” food environment, we created a variable indicating the ratio of the number of health-harming food outlets to the total number of food outlets within a 1,000-m buffer of each person:

$$\frac{\text{Fast - food outlets and}\text{convenience}\text{stores}}{\text{Sit - down}\text{restaurants}+\text{grocery}\text{stores}+\text{fast - food}\text{and}\text{convenience}\text{stores}}.$$

There were 2 rationales for the 1,000-m buffer. First, this distance is a reasonable approximation of walking distance (approximately 15 minutes to travel by foot), and second, this buffer size is commonly used in studies that have applied geographic information systems methods to quantify neighborhood characteristics (26, 27). Thus, this buffer size allowed us to more directly compare our findings with prior work. However, to assess whether our results were consistent across different scales of measurement, we repeated these analyses using buffers of 2,000 m and 4,000 m.

Neighborhood SES

As described in detail elsewhere (28), neighborhood SES was indexed using a composite of characteristics of the population within each SAMS, including percentage of residents with low educational attainment, percentage with low household income, percentage unemployed, and percentage receiving social welfare assistance. Higher values on the index indicate more deprived (lower-SES) neighborhoods. Index scores were then categorized for analysis as low deprivation (>1 standard deviation (SD) below the mean), moderate deprivation (≤1 SD below and ≤1 SD above the mean) and high deprivation (>1 SD above the mean).

Diabetes status

Cases of T2D from July 1, 2005, to December 31, 2010, were identified using the Swedish Prescribed Drug Register. This register includes data on all hospital, outpatient, and community (primary-care) prescriptions dispensed in Sweden since July 1, 2005. T2D was indexed by a prescription for insulin or insulin analogs (Anatomical Therapeutic Chemical Classification System code A10A) or oral antidiabetic agents (Anatomical Therapeutic Chemical Classification System code A10B or A10X). For the analysis of prevalent cases, the outcome was a diabetes medication prescription received between July 1, 2005, and December 31, 2005. For the analysis of incident T2D identified from 2006 to 2010, prevalent cases of T2D in 2005 were excluded from the analysis.

Other covariates

Individual-level covariates included age (in years), sex, educational attainment (categorized as ≤9 years, 10–11 years, or ≥12 years of schooling), and household income (categorized into quartiles).

Analysis

We first examined the relationship between the food environment and T2D by focusing on exposures measured in 2005. We examined the relationship between characteristics of the food environment and both prevalent T2D in 2005 and incident T2D from 2006 to 2010 (excluding prevalent cases of T2D and individuals who migrated or died). For both outcomes, random-intercept multilevel logistic regression was used to account for the geographic clustering of observations within SAMS, and we assessed whether the relationship was nonlinear using squared terms. Results from these models were adjusted for individual-level characteristics, including age, sex, household income, and education. We estimated the intraclass correlation coefficient, the proportion of variance in the outcome attributable to differences between individuals in different SAMS as opposed to differences between individuals within the same SAMS (29, 30). The intraclass correlation coefficient ranges from 0 to 1, with higher values indicating that individuals within the same SAMS are more highly correlated than individuals in different SAMS. We also conducted a sensitivity analysis by repeating this analysis while restricting the sample to residents of the lowest-SES neighborhoods (SAMS in the bottom quintile of the SES index). Finally, we repeated the analysis within geographic strata (southern Sweden vs. northern Sweden), and our results were consistent across these areas.

Next, to determine the degree to which individual residential mobility (self-selection) may have confounded the relationship between the food environment and T2D risk, we categorized individuals into “stayers” (lived in the same SAMS in 2000 and 2005) and “movers” (lived in different SAMS in 2000 and 2005). We used incident T2D as an outcome in 2006–2010, and we examined associations in movers and stayers.

Analyses were completed using ArcGIS software (version 10; ESRI, Redlands, California), SAS (version 9.3; SAS Institute, Inc., Cary, North Carolina), and MLwiN (version 2.27; Bristol, United Kingdom). All P values refer to 2-tailed tests.

RESULTS

The prevalence of T2D in 2005 was 7.7%, and during the follow-up period (2006–2010) there were 143,187 incident cases of T2D (cumulative incidence: 3.2%). Table 1 shows the distribution of T2D cases in 2005 by neighborhood SES and characteristics of the food environment. As expected, T2D was more prevalent in high-deprivation areas. For all indicators of the food environment (both health-harming and health-promoting), the distribution of T2D cases was curvilinear, with the lowest prevalence being observed in the areas with the highest and lowest concentrations of these outlets. We confirmed the curvilinear relationship using a categorical dummy variable (e.g., groupings of 0%–19%, 20%–39.9%, 40%–59.9%, etc.) and results were consistent with those derived using the continuous measure.

Table 1.

Distribution of Study Participants According to Neighborhood Deprivation, Characteristics of the Food Environment, and Type 2 Diabetes Mellitus, Sweden, 2005

Characteristic	All Individuals		SAMS		Prevalent T2D		P Valueb
	No.a	%	No.	%	No.	%
Neighborhood deprivationc							<0.001
Low	990,921	21.0	1,706	18.2	54,762	5.5
Moderate	2,943,245	62.4	6,467	69.2	228,016	7.7
High	784,417	16.6	1,180	12.6	82,444	10.5
Quintile of fast-food and convenience store densityd							<0.001
1	773,578	26.2			58,157	7.5
2	493,154	16.7			39,781	8.1
3	604,854	20.5			52,146	8.6
4	558,923	19.0			48,240	8.6
5	518,342	17.6			36,861	7.1
Quintile of sit-down restaurant densityd							<0.001
1	861,548	25.5			62,884	7.3
2	503,635	14.9			39,519	7.8
3	696,003	20.6			58,977	11.7
4	652,836	19.3			56,574	8.7
5	664,535	19.7			48,027	7.2
Quintile of grocery store densityd							<0.001
1	845,467	25.7			63,031	7.5
2	558,578	17.0			44,051	7.9
3	677,334	20.6			57,940	8.6
4	615,993	18.7			52,921	8.6
5	590,288	18.0			43,273	7.3
Quintile of the ratio of health-harming food outlets to total food outletsd							<0.001
1	667,041	22.6			48,141	7.2
2	525,518	17.8			43,175	8.2
3	586,253	19.9			49,500	8.4
4	655,945	22.2			54,867	8.4
5	514,094	17.4			39,502	7.7

Abbreviations: SAMS, small area market statistics; SD, standard deviation; T2D, type 2 diabetes mellitus.

^a Sample size for each type of food outlet is limited to SAMS with ≥1 outlet of that type.

^b P for trend across levels of neighborhood deprivation or quintiles of food environment.

^c Low, >1 SD below the mean; moderate, ≤1 SD below and ≤1 SD above the mean; and high, >1 SD above the mean.

^d Density of food outlets was measured as number of outlets per km².

Examined as individual exposures, neither the number nor the density of fast-food restaurants (see Web Table 1, available at http://aje.oxfordjournals.org/), sit-down restaurants (Web Table 2), or grocery stores (Web Table 3) was associated with prevalent T2D in 2005, in either crude or adjusted models, nor were any of these indices associated with incident T2D from 2006 to 2010 (Web Tables 4–6). However, food environments that were more net-negative (indicated by the ratio of heath-harming food outlets to total food outlets) were associated with higher relative odds of both prevalent (Table 2 and Figure 1) and incident (Table 3 and Figure 1) T2D in a curvilinear manner. Environments in which one-third of the food outlets were health-harming were associated with the highest risk for T2D (Figure 1). Notably, neighborhood SES remained significantly associated with T2D even after accounting for characteristics of the food environment (and even when characteristics of the food environment were not themselves related to T2D, as shown in Web Tables 1–6). This indicates that these features of the built environment do not explain the SES disparities in T2D shown in Table 1. Findings obtained using the 2,000-m and 4,000-m buffers were consistent with those obtained using the 1,000-m buffer (data not shown).

Table 2.

Relationship Between the Ratio of Number of Health-Harming Food Outlets to Total Number of Food Outlets and Prevalent Type 2 Diabetes Mellitus (N = 2,948,851), Sweden, 2005

	Unadjusteda		Adjustment 1b		Adjustment 2c
	OR	95% CI	OR	95% CI	OR	95% CI
Ratio of health-harming food outlets to total food outlets, per km2	0.88	0.82, 0.95	1.81	1.45, 2.26	1.85	1.51, 2.26
Ratio of health-harming food outlets to total food outlets, per km2 squared			0.32	0.22, 0.46	0.39	0.28, 0.55
Neighborhood deprivationd
Low					1.00	Referent
Moderate					1.26	1.24, 1.28
High					1.71	1.67, 1.75
Age, years
35–44			1.00	Referent	1.00	Referent
45–54			2.24	2.20, 2.28	2.24	2.20, 2.28
55–64			4.23	4.16, 4.30	4.25	4.18, 4.31
65–74			5.71	5.61, 5.82	5.75	5.65, 5.86
75–80			5.02	4.92, 5.12	5.07	4.97, 5.17
Female sex (referent: male)			0.60	0.59, 0.60	0.60	0.59, 0.60
Quartile of annual family incomee
High			1.00	Referent	1.00	Referent
Middle-high			1.14	1.13, 1.16	1.13	1.12, 1.15
Middle-low			1.36	1.34, 1.38	1.34	1.32, 1.36
Low			1.40	1.38, 1.42	1.38	1.36, 1.40
Educational attainment, years
>12			1.00	Referent	1.00	Referent
10–12			1.39	1.37, 1.40	1.38	1.36, 1.40
≤9			1.69	1.66, 1.71	1.67	1.64, 1.69

Abbreviations: CI, confidence interval; ICC, intraclass correlation coefficient; OR, odds ratio; SD, standard deviation; SE, standard error; SES, socioeconomic status; T2D, type 2 diabetes mellitus.

^a Model variance = 0.108 (SE, 0.003); explained variance = 0%; ICC = 0.032.

^b Results were adjusted for age, sex, education, and household income. Model variance = 0.064 (SE, 0.002); explained variance = 41%; ICC = 0.019.

^c Adjusted for age, sex, education, household income, and neighborhood deprivation. Model variance = 0.029 (SE, 0.001); explained variance = 74%; ICC = 0.009.

^d Low, >1 SD below the mean; moderate, ≤1 SD below and ≤1 SD above the mean; and high, >1 SD above the mean.

^e In Swedish kronor. High: >194,600; middle-high: 138,800–194,600; middle-low: 101,000–138,700; low: <101,000.

Figure 1.

Relationship between the ratio of health-harming food outlets to total food outlets and the predicted probability of type 2 diabetes mellitus, Sweden, 2005–2010. The analytical sample totaled 2,948,851 for prevalent type 2 diabetes mellitus in 2005 and 2,805,533 for incident type 2 diabetes mellitus during the period 2006–2010. Estimates were adjusted for age, sex, education, family income, and neighborhood deprivation index. Bars, 95% confidence intervals.

Table 3.

Relationship Between the Ratio of Health-Harming Food Outlets to Total Food Outlets and Incident Type 2 Diabetes Mellitus (N = 2,805,533), Sweden, 2006–2010

	Unadjusteda		Adjustment 1b		Adjustment 2c
	OR	95% CI	OR	95% CI	OR	95% CI
Ratio of health-harming food outlets to total food outlets, per km2	0.98	0.88, 1.09	2.32	1.68, 3.20	2.11	1.57, 2.82
Ratio of health-harming food outlets to total food outlets, per km2 squared			0.20	0.11, 0.34	0.30	0.18, 0.50
Neighborhood deprivationd
Low					1.00	Referent
Moderate					1.27	1.24, 1.30
High					1.78	1.73, 1.83
Age, years
35–44			1.00	Referent	1.00	Referent
45–54			2.15	2.10, 2.21	2.16	2.11, 2.21
55–64			3.47	3.39, 3.55	3.48	3.40, 3.56
65–74			3.85	3.75, 3.95	3.88	3.79, 3.97
75–80			2.76	2.67, 2.85	2.80	2.71, 2.89
Female sex (referent: male)			0.62	0.61, 0.63	0.62	0.61, 0.63
Quartile of annual family incomee
High			1.00	Referent	1.00	Referent
Middle-high			1.10	1.08, 1.12	1.08	1.06, 1.11
Middle-low			1.25	1.23, 1.28	1.22	1.20, 1.25
Low			1.34	1.31, 1.37	1.31	1.28, 1.34
Educational attainment, years
>12			1.00	Referent	1.00	Referent
10–12			1.39	1.36, 1.42	1.37	1.34, 1.40
≤9			1.67	1.64, 1.71	1.63	1.60, 1.67

Abbreviations: CI, confidence interval; ICC, intraclass correlation coefficient; OR, odds ratio; SD, standard deviation; SE, standard error; SES, socioeconomic status; T2D, type 2 diabetes mellitus.

^a Model variance = 0.112 (SE, 0.004); explained variance = 0%; ICC = 0.033.

^b Results were adjusted for age, sex, education, and household income. Model variance = 0.079 (SE, 0.003); explained variance = 30%; ICC = 0.023.

^c Results were adjusted for age, sex, education, household income, and neighborhood deprivation. Model variance = 0.022 (SE, 0.002); explained variance = 69%; ICC = 0.011.

^d Low, >1 SD below the mean; moderate, ≤1 SD below and ≤1 SD above the mean; and high, >1 SD above the mean.

^e In Swedish kronor. High: >194,600; middle-high: 138,800–194,600; middle-low: 101,000–138,700; low: <101,000.

We then examined whether the observed relationship between the ratio of health-harming food outlets to total food outlets and T2D risk was moderated by residential mobility. To do this, we compared the relationship between this exposure and incident T2D among 2 groups of people: individuals who changed SAMS between 2000 and 2005 and individuals who resided in the same SAMS during this period. As shown in Table 4, among persons who did not move, living in an area that gained relative access to health-harming food outlets also had higher odds of T2D (odds ratio = 1.72, 95% confidence interval: 1.27, 2.33). However, this relationship was stronger among persons who moved into areas with greater relative access to health-harming food outlets (odds ratio = 3.67, 95% confidence interval: 2.14, 6.30), suggesting that at least some of the relationship between access to the food environment and incident T2D reflects processes related to selection.

Table 4.

Relationship of Residential Mobility Between SAMS to Incident Type 2 Diabetes Mellitus, Sweden, 2006–2010

Residential Mobility Status	No. of Participants	Unadjusted		Adjustment 1a		Adjustment 2b
		OR	95% CI	OR	95% CI	OR	95% CI
Entire sample	2,805,533	0.98	0.88, 1.09	2.32	1.68, 3.20	2.11	1.57, 2.82
Persons who lived in the same SAMS in 2000 and 2005 (“stayers”)	2,050,956	0.88	0.79, 0.99	2.01	1.42, 2.87	1.72	1.27, 2.33
Persons who lived in different SAMS in 2000 and 2005 (“movers”)	754,577	1.29	1.06, 1.56	6.46	3.49, 11.96	3.67	2.14, 6.30

Abbreviations: CI, confidence interval; OR, odds ratio; SAMS, small area market statistics; T2D, type 2 diabetes mellitus.

^a Adjusted for age, sex, education, and household income.

^b Adjusted for age, sex, education, household income, and neighborhood deprivation.

DISCUSSION

The primary finding from this study is that in this population with at least some access to food outlets, specific characteristics of the local food environment (e.g., number of grocery stores and number of fast-food outlets), in and of themselves, were not related to T2D. Rather, more net-negative local food environments were associated with elevated likelihood of both prevalent and incident T2D in a curvilinear manner. This association persisted after accounting for individual-level characteristics. However, these characteristics of the food environment do not substantially explain socioeconomic disparities in T2D. These findings are consistent with a recent US study that used a similar analytical approach and found little evidence that objective measures of the built environment related to either physical activity or food access were related to T2D risk; however, the investigators did report significant relationships between subjective measures of perceived access to these outlets and T2D (26). To our knowledge, the present study is one of the first to examine how residential mobility moderates the relationship between the food environment and T2D risk, and our analyses indicate that a substantial portion of the association between the local food environment and T2D risk reflects selection processes.

Our finding that the relationship between the net food environment and T2D risk is curvilinear warrants comment. There is a growing appreciation that geographic disparities in health do not reflect monotonic relationships between exposures (e.g., studies of population density indicate that health disparities are present in both rural and urban areas) (31, 32). Our interpretation is that the ratio of health-harming food outlets to total food outlets reflected the fact that areas of limited access (food deserts) and areas of high concentration of health-harming food outlets (“food swamps”) are both associated with T2D. However, the mechanisms at these alternate ends of the distribution are likely different, which potentially explains the curvilinear relationship (33–35). Prior work from Sweden has shown that negative food outlets (e.g., fast-food outlets and convenience stores) are positively correlated with recreational facilities, and that both of these health-harming and health-promoting resources are negatively correlated with neighborhood SES (15). This suggests that the net food environment may also be reflecting other aspects of the environment that, in concert with food access, influence T2D risk. In this analysis, we excluded both the most sparsely populated SAMS and SAMS that had no food outlets of each particular type, thereby excluding outliers that could introduce bias into studies of geographic patterns of health in less densely populated areas (36). Despite these exclusions, we still observed this curvilinear relationship, indicating that neighborhoods where one-third of the food outlets were health-harming were associated with the highest T2D risk. Other neighborhood characteristics that are hypothesized to reduce T2D risk (e.g., walkability) (37) may be positively correlated with this metric of the net food environment and contribute to this curvilinear relationship. The recent report by Christine et al. (26) also points to the relevance of factors related to how individuals perceive their environments.

Public health and policy implications

A growing body of research indicates that there is substantial international and intranational heterogeneity in the relationship between neighborhood SES, food environments, and obesity-related conditions such as T2D (12). The finding that the density of neither health-promoting nor health-harming food outlets, in and of themselves, was associated with T2D is contrary to several prior reports, particularly those from the United States (7–10). This demonstrates that the relationship between neighborhood SES and the built environment is modifiable (11, 12); that is, it is not inevitable that poor urban areas are characterized by either food deserts or high concentrations of health-harming food outlets (31). This suggests that social factors and government policies could influence the relationship between neighborhood poverty and food environments in both positive and negative ways. For example, studies from Sweden have shown that both health-damaging and health-promoting food outlets are more common in poor areas (15). Thus, health-promoting outlets, such as physical activity facilities and green spaces, may counteract the potentially negative influence of, for example, fast-food outlets (15).

Strengths and limitations

These results should be interpreted in the light of our study's limitations and strengths. We recognize that there is debate as to how to best index “good” or “bad” food environments (33). Our measures indexed food access but not food outlet utilization (11); additional information regarding frequency of use of various types of food outlets would clarify these relationships but was not available in our data. Also, while we treated grocery stores as a health-promoting resource, these outlets also provide the opportunity for large purchases of calorie-dense and nutrient-poor foods, and thus access to a supermarket does not necessarily translate into a healthy diet (14). Our assessment of T2D relied on pharmacy records and thus it was limited to only clinically identified cases. Although the vast majority (>85%) of T2D patients are treated with some sort of medication (38), individuals managing their T2D with lifestyle modification alone would not have been captured by this data source. If individuals managing their T2D by lifestyle alone were also more likely to live in low-deprivation neighborhoods, this might have biased our findings. We also lacked data on other important confounders, such as ethnicity and physical activity, but our findings are consistent with those of studies that have been able to account for these variables (26). This study also had a number of strengths, including examination of multiple indicators of food access, the prospective study design, the large sample, and an analytical strategy that allowed us to examine the relationship between changes in the food environment and T2D risk while accounting for self-selection.

Conclusion

The present study indicates that elements of the food environment are associated with T2D risk and supports the adoption of policies aimed at reducing the potentially negative influence of health-harming local food environments. However, these results also indicate that the food environment does not substantially explain socioeconomic disparities in T2D in Sweden—a finding that is consistent with other studies from European nations (15, 19). Therefore, in order to address social disparities in T2D, policies and programs likely need to go beyond altering aspects of the food environment, because these characteristics were not correlated with SES disparities in this nationwide study.