Background

Key risk factors for type 2 diabetes mellitus (T2DM) are lack of physical activity, an unhealthy diet and lack of sleep [1, 2]. Real-life T2DM prevention programmes aimed at changing people’s lifestyle and behaviour have often been ineffective in the long term [3]. An important reason for this may be the focus on individual-level determinants of these lifestyle behaviours, such as motivation and ability, whereas they are also determined by more upstream drivers, such as the availability and accessibility of healthy options in an individual’s environment. In terms of changing and sustaining healthy lifestyle behaviours, the built environment is of importance [4,5,6,7].

Urbanisation is one example of an upstream driver. Urbanisation is associated with lower total physical activity and increased consumption of processed foods, which are high in fat, added sugars, animal products and refined carbohydrates [4, 8]. However, urbanisation has also been linked to higher total walking and cycling for transportation [4]. Built environmental characteristics, such as higher walkability, access to parks, and access to shops and services, are consistently associated with higher physical activity [4, 5]. Food built environmental characteristics, such as the perceived availability of healthy foods, are also associated with higher diet quality. In addition, greater availability of fast-food outlets has been associated with lower fruit and vegetable consumption [9, 10]. Other built environmental characteristics have been associated with higher stress and lack of sleep through residential noise, e.g. noise due to road and air traffic [11, 12].

By influencing physical activity, diet and sleep, these built environmental characteristics may also affect the risk/prevalence of T2DM. Indeed, the diabetes atlas showed higher T2DM prevalence in urban vs. rural areas [8], and a recent systematic meta-analysis reported similar results for South East Asia [13]. Two other systematic reviews addressed the association between specific built environmental characteristics and T2DM [14, 15]. However, one review only included German studies [14], while the second review included a broad range of cardiovascular disease outcomes, but only one study was included that considered T2DM as an outcome [15]. A recent meta-analysis showed that higher residential noise was associated with higher T2DM risk [16].

A comprehensive systematic review and meta-analysis of the current international evidence is, thus, lacking. This study aims to review systematically the evidence on the association between built environmental characteristics related to lifestyle behaviours and T2DM risk or prevalence, worldwide. Since characteristics of the built environment may vary with the country-specific income level, we stratified our analyses by this factor when possible. Meta-analyses were performed when three or more studies investigated the same exposure and outcome.

Methods

Data sources and searches

A literature search was performed based on the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement (www.prisma-statement.org). We systematically searched the bibliographic databases PubMed, EMBASE.com and Web of Science Core Collection from their inception to 6 June 2017 (NdB and LS). Search terms included indexed terms from MeSH in PubMed, EMtree in EMBASE.com, as well as free-text terms. We used free-text terms only in Web of Science. Search terms expressing ‘diabetes’ were used in combination with search terms comprising ‘environment’. Bibliographies of the identified articles were hand-searched for relevant publications. Duplicate articles were excluded. The full search strategies for all databases can be found in Additional file 1. The protocol and search strategy used were uploaded to PROSPERO prior to the study being carried out (CRD42016035663).

Study selection

Two reviewers independently screened titles, abstracts and full-text articles for eligibility (NdB and JL, or JWJB). Studies were included if they: (i) studied a population of adults, 18 years or older; (ii) had T2DM incidence or prevalence, or the glycaemic markers HbA1c, glucose or insulin sensitivity as outcomes; (iii) included independent variables covering built environmental characteristics that potentially influence the risk of T2DM via lifestyle behaviours, physical activity, diet and sleep; and (iv) were written in English, Dutch or German. We excluded studies if they: (i) were not conducted in the general population, but in specific subsamples, like pregnant women, or T2DM patients; (ii) focused on built environmental characteristics that directly affect the cardiovascular system (i.e. not via lifestyle behaviours), such as exposure to particulates due to roadway proximity; (iii) performed prediction analyses or (iv) were specific publication types that do not report original scientific research (editorials, letters, legal cases and interviews). As in the general population, the vast majority of diabetes cases are T2DM (>90%), studies were included if they did not specify the type of diabetes (type 1 diabetes mellitus or T2DM). Inconsistencies in study selection were resolved through consensus with a third reviewer (JL or JWJB).

Data extraction

One reviewer (NdB) performed data extraction, according to a standard protocol, including measures of study design, outcome, outcome assessment and exposure assessment, demographics, and prevalence or effect measure. Data extraction was appraised by a second reviewer (JL) for a random subsample of the included studies.

Quality assessment

Two reviewers (NdB and JWJB, or JL) independently evaluated the methodological quality of the full-text papers using the Quality Assessment Tool for Quantitative Studies, as described earlier by Mackenbach et al. [17]. This tool provides a quality score based on study design, representativeness at baseline (selection bias) and follow-up (withdrawals and drop-outs), confounders, data collection, data analysis and reporting. Each domain received a weak, moderate or strong score, resulting in seven scores. A study was rated as strong when it received four strong ratings and no weak ratings. A study was rated as moderate if it received one weak rating and less than four strong ratings. Finally, a study was rated weak if it received two or more weak ratings. Study quality was assessed in terms of the reported association between the relevant built environmental characteristic and T2DM, even if this was not the primary analysis presented in the study. Studies with a weak rating (n = 23) are presented in Additional file 2 and were included in sensitivity analyses, but excluded from the main analyses.

Data synthesis

Study characteristics were described in a systematic manner, according to the built environmental characteristics under investigation. These categories were made as homogeneous as possible, based on the lifestyle behaviours. Findings were further described according to country-level income, based on the World Bank list of economies, 2016 [18].

Studies were meta-analysed when three or more studies investigated the same exposure and outcome variables. In addition, the studies had to provide at least age and sex adjusted or standardised risk ratios or prevalence, and have a moderate or strong quality rating. If reported ratios were stratified and could not be pooled with the information provided in the publication, the study’s authors were contacted and asked to provide the pooled-risk ratio [19,20,21,22,23]. Reference categories were harmonised by taking the inverse of the risk ratio and 95% confidence interval (CI). If a risk ratio for a continuous variable was reported, we transformed this to a categorical risk ratio based on the methods of Danesh et al. [24]. Forest plots and random-effects meta-analysis models were fitted to relative risks or odds ratios. Plots and models were stratified for country income level and study quality, where permitted. In the sensitivity analyses, the studies with weak quality ratings were added to the models. Heterogeneity was tested using I2. Analyses were performed in R version 3.2.5 using the Metafor package.

Results

From the 11,279 identified references, 299 full articles were screened, and 109 of these studies were included in our review, of which 23 were not included in our main analyses due to a weak quality rating (Fig. 1 and Additional file 2). Included studies were categorised according to the built environmental characteristic investigated (Tables 1 and 2), and built environments were subdivided by physical activity environment, food environment and residential noise (Table 2).

Fig. 1
figure 1

Flow chart of study inclusion

Table 1 Study characteristics and results of studies investigating the association between urban and rural built environments and diabetes mellitus
Table 2 Study characteristics of studies investigating physical activity environment, food environment, residential noise and diabetes mellitus

Sixty studies compared T2DM risk/prevalence in urban vs. rural environments (Table 1 and Additional file 2). The studies rated weak (n = 16) did not differ in terms of country income levels from the other studies [25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40].

Of the remaining 44 studies, 25 (57%) of them found a higher risk or prevalence of T2DM in urban areas compared to rural areas. Altogether, 19 studies were eligible for the meta-analysis, which revealed a significantly higher risk/prevalence of T2DM in urban areas vs. rural areas (1.40; 95% CI, 1.22–1.61) (Fig. 2). This association was stronger in studies with strong quality ratings (1.44; 95% CI, 1.18–1.75), compared to those with moderate quality ratings (1.38; 95% CI, 1.11–1.70). After stratifying for country income level, one study was excluded [41] because the subgroup contained fewer than three studies. Associations were not different for upper-middle income countries (1.49; 95% CI, 1.16–1.92) and lower-middle income countries (1.45; 95% CI, 1.20–1.74), but were non-significant for high-income countries (1.16; 95% CI, 0.70–1.89).

Fig. 2
figure 2

Forest plots of meta-analysis of the association between built environmental characteristics and T2DM risk/prevalence. a Urban vs. rural environments, stratified for study quality. b Urban vs. rural environments, stratified for country income level. c Walkability. d Green space. e Grocery stores. f Noise. T2DM type 2 diabetes mellitus. RE model random effects model

Sensitivity analyses that included studies with weak quality ratings [33, 40] did not significantly change the results (Additional file 3).

Thirty studies investigated physical activity environment [19,20,21,22, 42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64] (Fig. 1, Table 2 and Additional file 2). All studies were performed in high-income level countries, except for one, which was performed in an upper-middle-level-income country [49].

Ten studies investigated the association between neighbourhood walkability and T2DM risk/prevalence. Six studies received a strong quality rating [20, 48, 57, 58, 62, 65]. Six studies observed that highly walkable neighbourhoods were associated with a lower T2DM risk/prevalence [19,20,21,22, 45, 54, 65]. In the meta-analyses of six studies, a pooled-risk ratio of 0.79 (95% CI, 0.72–0.87) was found, with an I2 for heterogeneity of 91.9%.

Six studies investigated the association between facilities for physical activity and T2DM risk/prevalence. Three studies received a strong quality rating [48, 49, 61]. Four studies did not observe an association between density of facilities and T2DM risk/prevalence [46, 48, 49, 61]. In two other studies, the higher availability of neighbourhood resources for physical activity was associated with lower T2DM risk [47, 63].

Seven studies investigated the association between green space and T2DM risk/prevalence. Two studies received a strong quality rating [44, 59]. Four studies observed that a higher availability of green space was associated with lower T2DM risk/prevalence [44, 54, 59, 64, 66]. One study observed that living closer to parks was significantly associated with higher prevalence of T2DM [64]. Aanother study observed a non-significant lower risk [42]. In meta-analyses of six studies, more green space tended to be associated with lower T2DM risk/prevalence with a pooled-risk ratio of 0.90 (95% CI, 0.79–1.03) with I2 for heterogeneity of 95.1%.

Four studies investigated infrastructure in relation to T2DM risk/prevalence. Two studies received a strong quality rating [49, 67]. Four studies did not observe an association between connectivity, infrastructure and road quality and T2DM risk/prevalence [49, 56, 68]. One study observed that a better transportation infrastructure, defined as more paved roads, was associated with higher T2DM prevalence [67].

Four studies investigated the association between safety and T2DM risk/prevalence. One study received a strong rating [49]. None of the studies showed an association between either traffic safety or safety from crime and T2DM risk/prevalence [49, 50, 56].

Twenty studies investigated characteristics of the food environment [46,47,48, 51,52,53,54,55, 60, 61, 63, 69,70,71,72,73,74,75,76,77] (Fig. 1, Table 2 and Additional file 2). All studies were performed in high-income-level countries.

Eight studies investigated the association between supermarkets and grocery stores and T2DM risk/prevalence. Two studies received a strong quality rating [61, 69]. One study observed that greater availability of grocery stores was associated with lower T2DM prevalence and that a higher percentage of households without a car located far from a supermarket was associated with higher T2DM prevalence [46]. A second study observed an unadjusted correlation between a greater distance to markets and lower fasting glucose levels [53]. Five studies did not observe a significant association between availability of supermarkets/grocery stores and T2DM prevalence [60, 61, 63, 69, 71, 75]. In a meta-analysis of three studies [48, 60, 61], a higher density of grocery stores was not associated with T2DM risk/prevalence (1.01; 95% CI, 0.98–1.05; I2 = 0%).

Seven studies investigated the association between availability of fast-food outlets and convenience stores and T2DM risk/prevalence. Three studies received a strong quality rating [61, 69, 72]. Four studies did not observe an association between availability of fast-food outlets/convenience stores and T2DM prevalence [61, 63, 69, 71, 75]. A higher availability of fast-food outlets and convenience stores was associated with higher T2DM prevalence in two studies [46, 72]. Studies could not be meta-analysed because the studies did not investigate consistent outcomes (T2DM risk vs. markers).

Four studies investigated the healthiness of the food environment subjectively or as an index and the association with T2DM risk/prevalence. One study received a strong quality rating [48]. Two studies focused on the perceived availability of healthy foods, rather than objectively measured availability. One study observed greater self-reported availability of healthy food resources to be associated with lower T2DM risk [47]. The second study assessed perceived availability, objective availability and a combination of the two, of which only perceived availability was associated with a lower T2DM risk [48]. Another study found no association between the presence of food deserts and T2DM prevalence [78].

Three studies used a ratio of unhealthful food stores to more healthful food stores, such as the Relative Food Environment Index (RFEI), with a higher value indicating an unhealthier food environment. One study received a strong quality rating [70]. This study observed that a higher ratio, i.e. a relatively unhealthier food environment, was associated with a higher risk of T2DM. Two studies did not observe consistent associations between RFEI and T2DM risk [54, 74].

Six studies used composite measures of physical activity and food-related built environmental characteristics (Tables 2 and 3, and Additional file 4). One study received a strong quality rating [79]. A summary score indicating the presence of more healthy food resources and physical activity resources was associated with lower T2DM incidence [47]. Furthermore, residing in a neighbourhood with physical and social-environmental disadvantages was associated with higher T2DM prevalence [79]. Clusters of large metropolitan counties, characterised by low population density, median income, low socioeconomic status index and greater access to food observed less T2DM [73]. Finally, no association was observed between vibrancy index, density and obesogenicity clusters and T2DM risk/prevalence [68, 80, 81].

Table 3 Study results of studies investigating physical activity environment, food environment, residential noise and diabetes mellitus

Four studies investigated the association between residential noise and T2DM risk/prevalence. One study received a strong quality rating [82]. All studies observed that higher exposure to residential noise was associated with increased T2DM risk/prevalence [82,83,84,85]. In meta-analyses of four studies [83,84,85,86], higher exposure to residential noise was not associated with T2DM risk/prevalence (1.49; 95% CI, 0.78–2.82, I2 = 75.8%).

Discussion

This systematic review investigated evidence for the association between built environmental characteristics, related to lifestyle behaviours, and T2DM risk/prevalence, worldwide. The association between built environmental characteristics and T2DM risk/prevalence has been investigated a fair amount, with 84 studies on the subject, although for our review, 23 of these studies were excluded due to their low quality ratings. Urbanisation was associated with a higher T2DM risk/prevalence. The evidence for an association between the physical activity environment and T2DM risk was more consistent than it was for the food environment. Higher neighbourhood walkability was associated with lower T2DM risk and more green space tended to be associated with lower T2DM risk.

First, we observed that residing in urban areas was associated with higher T2DM risk/prevalence, in line with the findings of the IDF diabetes atlas [8] and a recent meta-analysis for South East Asia. Urbanisation is a process in which inhabitants of a particular region increasingly move to more densely populated areas. Urbanisation is a broad operationalisation of the built environment and includes a range of characteristics, such as higher availability of food, facilities, and infrastructure. In general, previous reviews have observed conflicting results for urbanisation [4, 5, 8]. Urbanisation has consistently been associated with less physical activity and unhealthier dietary habits, but also with higher total walking and cycling for transportation [4, 5, 8]. The observed heterogeneity in terms of results might be due to the variety of definitions used to classify an urban area, which is distinct for different countries and studies. To account for this, we stratified our analyses by country income level [18], and the majority of studies (38 out of 60) were conducted in middle-income countries, which reduces the heterogeneity in the studies included. It must be recognised that considerable heterogeneity in definitions of urban vs. rural exists beyond stratification on country income level. Across countries with the same country income level, there is large variety of what urban or rural areas may look like and the populations that reside in these areas. At present, there is no homogeneous and generally accepted definition of urban or rural areas and the majority of studies did not include a definition that was used to make this classification.

Second, the present study provides consistent evidence for an association between the built physical activity environment and T2DM risk/prevalence. Higher walkability and availability of green space were most consistently associated with lower T2DM risk/prevalence. Our results for urbanisation seem contradictory to the lower T2DM risk/prevalence associated with greater neighbourhood walkability, since greater walkability is often observed in more urbanised environments [5]. These seemingly contradictory results could be explained by the underrepresentation of high-income countries in the urban to rural comparison studies, and the overrepresentation of these countries in walkability studies. The (perceived) walkability of urban areas also varies across different parts of the world. So, whereas walkability may be a feature of cities in high-income regions, this may not be the case in cities in lower-income regions. Furthermore, urbanisation is a much broader construct than walkability, and even within one urban area, walkability may differ between or even within neighbourhoods. In addition, other urbanisation-related issues, besides walkability, may be more important, such as other physical activity environment characteristics and the food environment, which counterbalance the effects of walkability in urban areas. These results would suggest that certain aspects of the built food environment were associated with a higher T2DM risk, but we could not find consistent evidence of this in our review.

An association between the built food environment and T2DM risk/prevalence was not consistently observed. The availability of fast-food and convenience stores and the perceived healthiness of the food environment tended to be associated with higher T2DM risk/prevalence and lower T2DM risk/prevalence, respectively. However, due to heterogeneity in the studies, insufficient studies were available for meta-analysis, thus preventing us from drawing solid conclusions. The only possible meta-analyses were three studies including the density of grocery stores, but this confirmed that no significant associations could be observed. Also by reviewing the evidence, supermarkets and grocery stores and the RFEI were not associated with T2DM risk/prevalence. These findings are consistent with an earlier systematic review that reported that perceived availability was associated with healthy dietary behaviours [9], whereas objective measures of accessibility and availability of food environment yielded mixed results [9]. The association between the perceived environment and a healthier diet can be explained by not limiting the concept of environment to specific shops or locations, but rather to the participant’s resources for healthy food, e.g. gardens and markets. On the other hand, perceptions may also reflect an individual’s intentions and motivations rather than location alone. A difficulty with regard to establishing useful diet measures is that they are very heterogeneous and difficult to define. For instance, access to a supermarket is often seen as contributing to a healthy food environment, even though they are also sources of unhealthy products [9]. Establishing a comprehensive definition is further complicated because food can be bought in a variety of shops and locations that are not directly associated with food, e.g. at the counter of a pharmacy. The same heterogeneity was observed to a lesser extent in the built physical activity environment. For instance, infrastructure includes drivers for active transportation (sidewalks and cycling lanes) as well as for passive transportation (public transport and roads) [87]. We conclude that the heterogeneity in exposure assessment associated with built environmental variables made the examination of the associations with T2DM risk/prevalence more difficult.

Finally, although higher exposure to residential noise was consistently associated with higher T2DM risk/prevalence in individual studies, this was not confirmed in our meta-analysis, in contrast with an earlier meta-analysis [16]. This difference could be explained by the inclusion of only confounder adjusted risk ratios in our study.

A strength of this study is the comprehensive overview of the literature on the association between built environmental characteristics and T2DM risk/prevalence, in which we included worldwide evidence. We assessed study quality and took country income levels into account. However, certain limitations of this study need to be addressed.

A weakness of any systematic review and meta-analysis is that its quality is dependent on the quality of the studies included. For instance, not all studies that were included distinguished between T2DM and type 1 diabetes mellitus. However, the majority of all people with diabetes have T2DM so the evidence provided in our review was very likely applicable to T2DM risk/prevalence [1]. Secondly, because most studies in the present review were cross-sectional, our review cannot provide the foundation for causal inferences. Finally, publication bias could influence our findings, but our search turned out a relatively high number of null findings, suggesting publication bias an unlikely limitation. Finally, residential self-selection is an important issue that should be included in studies investigating the associations between built environment and disease. Self-selection occurs when residents choose a residence based on socioeconomic or other circumstances, or lifestyle preferences. Evidently, such selections may influence our results, as for instance higher socioeconomic status neighbourhoods may contain more green space, as well as more highly educated and health-conscious residents. However, the true effect of residential self-selection on these associations has often not been accounted for in the included studies and is difficult to investigate. One narrative review observed that studies using various approaches to identify self-selection (i.e. a questionnaire or statistical methods) explained only a minor part of the associations between built environment and travel behaviours [88]. Two studies included in the present review observed that residential relocation, as an indicator of self-selection, resulted in inconsistent effects on associations with health outcomes [57, 58]. It is, therefore, hard to conclude on the effect of self-selection bias on our results, based on the current evidence.

Despite the limitations of our study, our results may be relevant for infrastructure planning. For example, in addition to other positive consequences of walkability and access to green space, these environmental characteristics may also contribute to T2DM prevention. Future research should focus on developing a more homogeneous definition of environmental characteristics, particularly in relation to the food environment. Also, more in-depth explorations are necessary of the pathways through which environments affect diabetes risk, while taking the potential confounding variables into account.

Conclusions

In conclusion, urbanisation is associated with higher T2DM risk/prevalence. The built physical activity environment - walkability and access to green space, in particular - was consistently associated with reduced T2DM risk/prevalence, while no consistent evidence was found for an association between the built food environment and T2DM risk/prevalence. These conclusions have implications in terms of urban planning and the inclusion of walkable and green cities.