Higher Neighborhood Drivability Is Associated With a Higher Diabetes Risk in Younger Adults: A Population-Based Cohort Study in Toronto, Canada Free

The global rise in diabetes has been linked to increasing levels of urbanization (1,2), partly because of losses in opportunities for physical activity and an increase in sedentary behaviors. In many cities, the urban landscape has shifted over time from that of compact, pedestrian-oriented communities to sprawling car-dependent neighborhoods. As a result, car use has become the most common mode of transportation in many regions of the world. However, in large longitudinal studies, car travel has been associated with greater exposure to air pollution, noise, and stress and a higher likelihood of physical inactivity and weight gain than other, more active, modes of transportation (3–7). Each of these risk factors contributes to the development of type 2 diabetes. In fact, some studies demonstrated a higher rate of diabetes among people who spend more time in cars compared with those who primarily walk or cycle (8–10).

Neighborhood environments play a critical role in adverse health outcomes and whether individuals choose to drive as their primary mode of transportation (11,12). Evidence suggests that neighborhoods that enable residents to engage in active forms of transportation have lower rates of obesity and type 2 diabetes (2,13). A meta-analysis indicated a 21% lower risk of diabetes in such neighborhoods (2). How such associations vary across age-groups is not completely understood. In general, older adults are less likely to change their choice of transportation (14), and the association between the built environment and physical activity appears to be smaller in adults aged ≥65 years (15,16). In fact, prior studies suggest that working age adults are more susceptible to the positive health effects of living in a supportive environment – (and of engaging in active forms of transportation – than older adults (6,17). Thus, younger adults might benefit most from neighborhood-level interventions to reduce car use. Younger adults have experienced the greatest rise in obesity in recent decades, with even modest changes in body weight tipping the balance toward diabetes development in genetically susceptible individuals (18,19). By generating built environments that reduce reliance on cars for travel, cities could help to mitigate the rise in obesity-related conditions such as diabetes, especially among younger adults (20).

To date, no studies have comprehensively investigated the health impact of built environment exposures on health outcomes in different age-groups. This study investigates whether residing in a car-dependent neighborhood (high drivability) is associated with an increased risk of developing diabetes compared with less car-dependent areas (low drivability) and whether younger adults are indeed more susceptible to these effects. We explored this phenomenon in a large population-based sample of working age adults from Toronto, Canada, using a novel, validated index that predicts car use (12).

This population-based cohort study examined the association between neighborhood drivability and diabetes development in Toronto, Ontario, Canada. Toronto is one of the largest cities in North America, with a population of nearly 2.8 million in 2021 (12). The study population was constructed using administrative health care data from the province of Ontario, including comprehensive information on hospitalizations, physician visits, laboratory testing, and other services covered by the province’s universal health care system. These data are housed at ICES (formerly known as the Institute for Clinical Evaluative Sciences), an independent and nonprofit research institute affiliated with the University of Toronto. ICES’s legal status under Ontario’s personal health information privacy legislation allows it to collect and analyze health care and demographic data without consent for health system evaluation and improvement. Records for individuals were linked across data sets using a unique encoded identifier to retain anonymity.

The Registered Persons Database was used to identify all adults aged 20–64 years who were living in Toronto on 1 April 2011 (index date) and were permanent residents of Ontario at the time and therefore eligible for health care coverage under the Ontario Health Insurance Plan (OHIP). The Registered Persons Database includes demographic and residential information and vital status of all individuals registered under OHIP. We used records from the Ontario Diabetes Database (described below) to exclude individuals who had preexisting diabetes at baseline. We also excluded anyone with missing neighborhood-level income or drivability scores (as described below), anyone who did not have health care coverage under OHIP a minimum of 3 years prior to the index dates, or anyone living in a long-term-care facility (i.e., a nursing home or home for elderly people) on the index date. The study protocol was approved by the ICES Privacy and Legal Office.

A neighborhood drivability index was developed for the City of Toronto using factor analysis and validated against individual-level travel survey data, as described previously (12). Briefly, the drivability index was derived for each neighborhood, known as a dissemination area (DA). A DA is a small geographical unit defined by Statistics Canada as having a median population of 537, an area of 1 km², and fairly homogeneous socioeconomic characteristics of its residents. Our final drivability index is a composite score based on three factors: urban sprawl (lower densities of residents, jobs, facilities, and street intersections), pedestrian facilities (lower number of pedestrian crossings, public transit options, and walkable destinations), and parking facilities (more parking facilities and lower land use mix). Scores were created using open data sources, including the 2016 Canadian Census, Canadian Active Living Environments database, DMTI Spatial, and the Municipal Open Data Catalogue for the City of Toronto (12). Scores for each of the three factors were combined into a single composite score by taking the sum, which was then divided into quintiles from lowest (quintile 1 [Q1], least car dependent) to highest (Q5, most car dependent). Prior validation demonstrated that the neighborhood drivability index predicted car use (area under the curve 0.73). Highest neighborhood drivability was associated with an 80% higher overall rate of car travel and a nearly threefold higher rate of short car trips (12). In this study, drivability quintiles were assigned to the residential DAs of cohort members at baseline. To assess the impact of residential relocation during follow-up, we reported the number of participants who relocated and whether this relocation resulted in changes in drivability quintile. We performed a sensitivity analysis whereby we reran our main models after excluding individuals who moved out of Toronto (drivability could not be assigned).

Individuals were followed from our baseline date of 1 April 2011 until a maximum follow-up date of 31 March 2018. Diabetes onset was derived from the Ontario Diabetes Database, which is based on two validated algorithms using hospital records and physicians’ services claims from the Canadian Institute for Health Information and OHIP claims. The algorithms do not discriminate between type 1 and type 2 diabetes, but >95% of total cases are estimated to be type 2 diabetes. The two algorithms used in the main analysis indicated a diagnosis of diabetes with a high level of sensitivity (94%) and specificity (92%) (21). In a sensitivity analysis, we used an algorithm with a lower sensitivity (84%) but higher specificity (99%) (21) to assess the robustness of the results.

We included several covariates in our analyses, including baseline demographic information, such as age-group (20–34, 35–44, 45–54, and 55–64 years), sex (male/female), immigration status (recent immigrant yes/no), and ethnicity. Immigration status was derived from the Immigration, Refugees and Citizenship Canada Permanent Resident database, which included information on all immigrants accepted as permanent residents in Canada since 1985. Those included in this database were classified as recent immigrants to Canada; all others were considered to be long-term Canadian residents (22). Ethnicity was defined according to a validated surname algorithm that categorizes individuals into three categories: South Asian, Chinese, and general population (not South Asian or Chinese) (23). Using established algorithms based on administrative health databases (24–26), we retrieved information regarding prior comorbidities, including asthma, chronic obstructive pulmonary disorder, stroke, myocardial infarction, congestive heart failure, and rheumatoid arthritis.

We included several neighborhood characteristics. Area-level income was assigned to each participant using the median household income level of their residential neighborhood (DA) based on the 2011 Canadian census and divided into quintiles. We also assigned neighborhood walkability to residential DA using a validated index consisting of four components: population density, dwelling density, street connectivity, and number of destinations (e.g., banks, grocery stores) within an 800-m walking distance from the center of each neighborhood (27).

Baseline characteristics were summarized by drivability quintile as mean ± SD or median and interquartile range, as appropriate for continuous variables, and number and percentage for categorical variables. For our main analyses, Cox proportional hazards modeling was used to estimate the effect of neighborhood drivability quintile on diabetes development. Person-time for each individual was calculated from baseline until the time of diabetes onset, death, loss to follow-up, admission to a long-term-care facility, or end-of-study date. We report hazard ratios (HRs) and 95% CIs for each quintile of drivability (reference category, Q1) after adjusting for age and sex (model 1) and additionally adjusting for area-level income, ethnicity, and immigration status (model 2) and for baseline comorbidities (model 3). We tested for clustering of individuals by neighborhood by adding a random intercept at the DA level in model 1 and by calculating the variance partition coefficient (28). A random intercept was added to the model if the variance partition coefficient was >0.05. Otherwise, Cox regression models without random intercept are presented.

On the basis of our initial hypotheses, we examined whether there was an interaction between drivability and age-group. We also assessed interactions among other key demographic variables, including sex, neighborhood-level income, and recent immigration status, by including a multiplicative interaction term. Findings were then stratified by both age and one of the above variables if the estimates for the association between drivability and diabetes incidence in each subcategory of the variable had 95% CIs that were not overlapping.

As a sensitivity analysis, we investigated the combined effect of neighborhood drivability and walkability on diabetes incidence in each age-group. Therefore, we included interaction terms for each level of drivability and walkability tertiles. To quantify the additive interaction, we calculated a relative excess risk due to interaction (29). To understand whether the two variables were measuring the same construct, we mapped the spatial distribution of walkability and drivability tertiles. All analyses were performed at ICES using SAS Enterprise 7.1 statistical software (SAS Institute, Cary, NC).

This study included all working age residents who lived in Toronto on 1 April 2011 (N = 1,739,694). We excluded those with preexisting diabetes (n = 126,832, 7.2%) and those lacking OHIP coverage (n = 124,671, 7.2%), those residing in a long-term-care facility (n = 683, 0.04%), and those with missing DA-level income (n = 8,153, 0.5%) or DA-level drivability (n = 5,361, 0.3%). Our final study population included 1,473,994 individuals (mean age 40.9 ± 12.2 years, 48.5% males). We found that those living in highly drivable neighborhoods were slightly older, more often female, and somewhat less likely to have chronic respiratory conditions at baseline than those living in the lowest drivability quintile. Neighborhoods with the lowest drivability scores had a greater concentration of high-income households and a lower proportion of recent immigrants (Table 1). Exploration of relocation illustrated that 50% of the sample did not move residence, and 22% moved to a similar (same or adjacent) quintile (Supplementary File 1). Only 8% changed more than one quintile in drivability index, and 20% moved outside of Toronto.

Overall, 77,835 (5.3%) individuals in our study population developed diabetes over a median follow-up of 7.0 years, including 18,393 (6.2%) individuals living in high drivability areas (Q5) and 10,835 (3.7%) individuals living in low drivability areas (Q1) (Table 1). Residents living in the most drivable neighborhoods had a 41% higher risk of diabetes after accounting for covariates (HR 1.41, 95% CI 1.37–1.44) compared with those living in the least drivable neighborhoods, and this risk rose as the drivability quintiles increased (Table 2). The variance partition coefficient for clustering at the neighborhood level was 0.05; therefore, we used models that were unadjusted for clustering. Supplementary File 2 shows the comparison of the main model with and without accounting for clustering.

There was a significant interaction between age-group and drivability (P < 0.001). Among younger adults, living in highly drivable neighborhoods was most associated with a 50–70% increased diabetes incidence compared with those residing in the least drivable neighborhoods (HR_{20–34 years} 1.57 [95% CI 1.4–1.68], HR_{35–44 years} 1.61 [95% CI 1.52–1.69]) (Fig. 1). In the older age-groups, diabetes risk was also increased in highly drivable neighborhoods, but the magnitude of the association was smaller than for younger age-groups (HR_{45–54 years} 1.33 [95% CI 1.28–1.39], HR_{55–64 years} 1.31 [95% CI 1.26–1.37]). Furthermore, we observed effect modification by neighborhood-level income (P < 0.001 for interaction) such that the strongest associations were found among middle-income groups and the weakest associations among the lowest (Q1) and highest (Q5) income groups. Differences in risk across the strata of sex and immigration status were not distinct over each drivability index quintile (Supplementary File 3).

Because of this additional interaction by neighborhood-level income, we performed a double stratification for the associations between drivability and diabetes risk, including both age-group and neighborhood-level income quintile (Fig. 2). In all age-groups, the strongest associations between drivability and diabetes risk were observed among the middle-income populations. This finding was most pronounced in the youngest age-group. Among the young, middle-income population, the risk of diabetes was almost twofold higher in the highest drivability quintile compared with the lowest drivability quintile (middle income 1.96 [95% CI 1.64–2.33], low income 1.31 [95% CI 1.16–1.48], high income 1.29 [95% CI 1.04–1.60]). Among the older age-groups, patterns were similar, with associations between drivability and diabetes incidence greatest among the middle-income group; however, the magnitude of these associations was smaller (middle income 1.46 [95% CI 1.32–1.62], low income 1.20 [95% CI 1.10–1.29], high income 1.23 [95% CI 1.11–1.36]). Excluding people who moved out of Toronto over follow-up (20.1%) did not meaningfully change the results. The effect estimates became somewhat larger, albeit CIs were somewhat wider in the remaining sample (n = 1,182,317) (Supplementary File 4). Findings from all interaction analyses are shown in Supplementary File 5. Models that used a high-specificity algorithm to assign new diabetes cases yielded similar results (Supplementary File 6).

In sensitivity analyses, we observed that the risk of developing diabetes was lower when drivability decreased and walkability increased (Supplementary File 7). Even though these two constructs were correlated (r = 0.75), the relative excess risk due to interaction was 0.28, indicating the difference in magnitude of the combined effect of both drivability and walkability together minus the sum of each individual effect alone. In other words, excess risk associated with high drivability and low walkability (i.e., the two variables combined) was greater than that associated with each variable alone. The spatial distribution of walkability and drivability tertiles are mapped in Supplementary File 8.

This study shows that working age adults who reside in highly drivable neighborhoods were 41% more likely to develop diabetes than those living in low drivability areas. The strongest associations were observed among middle-income adults aged <45 years, among whom those living in the most car-dependent neighborhoods had an almost twofold increase in diabetes incidence. These findings suggest that urban designs enabling active forms of transportation have the potential to enhance diabetes prevention efforts on a large scale, especially in younger adults.

These results are in line with earlier evidence on the role of active commuting in disease risk. Previous studies have shown a direct association between the time people spend in cars and the degree of weight gain and risk of obesity (4–6). Several longitudinal studies also highlighted the positive health effects of active commuting as an alternative to car use with respect to BMI and metabolic outcomes (4–7). Moreover, neighborhoods that are conducive to walking are associated with lower levels of obesity and diabetes (2,13). The effect sizes observed for neighborhood drivability in this study are, however, much larger than those shown in earlier studies on walkability. A recent meta-analysis indicated that high walkability was associated with a 21% reduced risk of diabetes (2), while in the current study, high drivability was associated with a 40–96% increase in diabetes development in young and middle-income populations. One reason for this difference could be that elements of our drivability index, including access to public transit, are more closely tied to one’s modal choice than walkability alone (12). Alternatively, our study may have captured other related exposures or intermediate pathways toward diabetes incidence, such as traffic-related air pollution and stress caused by traffic congestion, which could further magnify the adverse health effects of car dependency (3,6). These explanations may point to why the combination of both drivability and walkability, although mutually correlated, was more strongly associated with diabetes risk than each construct alone. Thus, to optimize their impact, elements of both indices may need to be targeted to optimize the impact of healthy urban designs (7). Together with other studies, our research suggests that shifts in modal choice are a promising avenue for diabetes prevention. A policy evaluation study in Ottawa, Canada, estimated that shifting from car use during peak commuting hours to public transit, walking, or cycling could prevent ∼1,600 diabetes cases over 10 years in a working age population (30).

To our knowledge, this study is the first to show that younger and middle-income adults are particularly susceptible to the health impacts of car dependency. Modal choice resulting in healthier or unhealthier lifestyle behaviors could be a more impactful modifiable risk factor for diabetes in younger than older adults. Diabetes at a younger age results in a longer disease trajectory and has a greater impact on overall life expectancy, but lifestyle changes that can prevent the onset or conversion from prediabetes to diabetes at this early stage will have a long-lasting impact on an otherwise long disease trajectory (31). In older adults, the aging process, which results in declining β-cell function, is a major driver of diabetes development, and therefore, the impact of obesity and lifestyle behaviors could be less pronounced. Moreover, younger adults often commute to work or school each day, and thus, may be more strongly influenced by environmental factors associated with car use. In addition, older adults have accumulated more environmental exposures over their lifetime than younger adults. Although the drivability of a given neighborhood is likely to have remained stable over time, the social characteristics of residential areas can change substantially as a result of gentrification (27). Even though older adults have accumulated more environmental exposure over their lifetime, in terms of diabetes risk, the relative contribution of cumulative versus recent exposures is not well understood and could not be fully evaluated in our study.

In our study of working age adults, high- and low-income groups appeared less susceptible to the benefits of low neighborhood drivability than middle-income populations. There may be various reasons for this observation, including the extent to which populations depend on active transportation for their physical activity requirements. Time spent driving adds to an extended period of sedentary behavior in work settings, particularly for those in office-related occupations in middle- as well as high-income populations (32). High-income groups may be more health conscious and less reliant on transportation as an opportunity for physical activity because of greater resources to support leisure time physical activity through access to gym memberships or personal trainers (33). In contrast, middle-income adults may lack these resources to counter the negative health impacts of car use yet may still be able to buy or lease a car. While low-income groups may lack the resources to own a car altogether, the association between environmental factors and car use might be less pronounced in this group. Moreover, walking for leisure may not be promoted in low-income neighborhoods because of the lack of a well-maintained infrastructure, safety, and esthetics (34). Aside from car use, other risk factors may be more important determinants of diabetes risk among low-income populations (e.g., job insecurity, stress, food insecurity), thus mitigating differences in diabetes risk caused by drivability (35). Other studies have reported higher rates of active transportation in populations living in deprived areas (36,37). Furthermore, an Australian study observed a U-shaped association between neighborhood disadvantage and cycling for transportation, whereby middle-income neighborhoods had the lowest levels of cycling (37). Housing affordability also shapes commuting patterns and health impacts, as in some cities, residence away from the city center tends to be more affordable, resulting in longer commute times (38). In turn, the latter is associated with higher rates of obesity, mediated by car use (38), as well as an increased likelihood of smoking, shorter sleep duration, and increased psychological stress, which have all been linked to diabetes risk (39).

Strengths of the current study include the large, representative, and longitudinal population-based cohort, which enabled subgroup analyses. Availability of data on the novel drivability index as well as walkability index also provided unique insights into the field of transportation and diabetes risk.

Some limitations should be taken into account. First, we assessed drivability at baseline, assuming stability over follow-up. Since built environment characteristics related to transportation are rather static, we argue that one-time measurement is a fair indicator of exposure to neighborhood drivability. Other changes in exposure to the built environment could have emerged because of residential relocation, but we found that excluding relocators did not affect our results. Second, we focused on the drivability of residential areas only, where one’s entire activity space is substantially larger. Earlier studies showed that work environment is an important determinant of car use, which was not included in the current study, but it is likely that the inclusion of such environments would have magnified the associations we observed (11,32). Third, we could not test the causative pathway of neighborhood drivability, car use, physical activity, and the consequent development of diabetes, and, like any observational study, there may be residual confounding that could not be accounted for in our analysis. It remains important to explore causality in the association between built environment and health outcomes, for instance through statistical methods such as mediation analysis. Mediation analysis should be used to examine the underlying mechanisms of the association, including mediators such as physical activity, sedentary behavior, increased exposure to air pollution, stress in congested traffic, and lower access to health care. Finally, other social determinants of health may have been insufficiently adjusted for, such as neighborhoods with a high drivability having been shown to have a higher proportion of ethnic minorities who are at higher risk of diabetes (35).

This study provides implications for future research and practice. Future studies could investigate the interaction between drivability and other risk factors (e.g., the microbiome, epigenetic markers) as part of the broader exposome by linking environmental data to clinical or biomedical markers of diabetes risk. In clinical practice, evidence-based counseling regarding health behaviors can be personalized by considering built environment characteristics (e.g., drivability) and other social determinants of health. As such, a precision medicine approach can be applied to the built environment in which patients live; for example, if limited opportunities for active transport exist, tailored advice may be offered to increase leisure time physical activity instead (40).

Car dependency is a risk factor for diabetes development in our population, particularly for younger adults. These findings confirm the importance of health-promoting environments in altering disease risk and suggest a role for neighborhood drivability as an important target for population-level interventions to prevent diabetes. Our research indicates that middle-income neighborhoods may be an ideal target for built environment interventions that reduce dependency on cars. However, a healthy built environment may be necessary but not sufficient to mitigate diabetes risk in lower socioeconomic status areas. Discriminatory housing and urban planning practices, including historical policies such as redlining in the U.S., reinforce the adversity faced by low-income and racialized communities. Beyond drivability, broader public policies are needed to gain equity in creating healthy living environments. A key aim of the World Health Organization global action plan for noncommunicable disease prevention is to “reduce modifiable risk factors for NCDs and underlying social determinants through creation of health-promoting environments” (20). As societies recover from the coronavirus 2019 pandemic, consideration should be taken to identify opportunities for people to (re)engage with public spaces and to promote active forms of transportation while doing so.

Funding. Financial support for this research was given by an Amsterdam Public Health Research Institute travel grant; a European Foundation for the Study of Diabetes Albert Renold Travel Fellowship; Canadian Institutes of Health Research operating funds; EXPOSOME-NL, funded through the Gravitation program of the Dutch Ministry of Education, Culture, and Science and the Netherlands Organization for Scientific Research (NWO grant 024.004.017); and EXPANSE, funded from the European Union’s Horizon 2020 research and innovation program under grant agreement 874627.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. N.R.d.B. initiated the study, designed the analysis plan, interpreted results, and drafted the manuscript. J.W.J.B., G.S.F., P.G., N.A.H., J.L., J.S.M., F.R., J.B., and G.L.B. contributed to the conceptualization of the study, interpretation of the results, and major revisions of the manuscript. C.F.W. conducted all analyses. P.G. performed geographical analyses and support. R.M. provided advice for the statistical analyses. All authors critically read and approved the final submitted version of the manuscript. G.L.B. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.