The risk for diabetes was weakly positively associated with increasing mean levels of traffic-related air pollution at the residence. The risk was highest in nonsmokers and physically active people.
This is the first study to relate air pollution to incidence of diabetes assessed objectively from a nationwide register. Previous studies used self-reports of diabetes incidence (16
), whereas diabetes prevalence was based on administrative databases of physician billing and hospital discharges (14
) or on a national survey (15
). The onset of diabetes based on self-reports is loosely defined and subject to recall bias (16
). Thus, the use of objective measures of diabetes onset based on the nationwide register, NDR, is appealing and convenient, since the entire population is covered by uniform inclusion criteria and the dropout rate is zero (19
). The disadvantages of the NDR include inability to distinguish between diabetes types, although the majority of new diabetes cases in this age-group is most likely of type 2 diabetes (1
). The date of inclusion in the NDR is only a proxy for the date of formal clinical diagnosis, which was likely made some time before inclusion in the register. No information on glucose or other clinical measurements used at diagnosis are available. Finally, the NDR likely underestimates the actual diabetes burden, since people without clinical diagnoses are not included.
“All diabetes” (n
= 4,040) definition of incidence was previously validated (19
) by a study comparing register-identified patients with their general practitioners, finding sensitivity of 86% and positive predictive value of 90%. Of four inclusion criteria, hospital discharge diagnoses, diabetes medication records, and chiropody all reflect highly likely confirmed diabetes cases. In a “confirmed diabetes” definition (n
= 2,877), we excluded the 1,163 diabetes cases included solely because of blood glucose tests, because without available results of these tests, or other records in the NDR, it is likely that many of these people did not have diabetes. Because of increasing awareness among physicians on detecting undiagnosed diabetes, it is not uncommon for elderly healthy people to have five blood glucose tests per year.
Incidence rates in this cohort, as observed in all of Denmark, increased over the period 1995–2004 and showed a tendency to decline after 2004 (19
), which we adjusted for by modeling the calendar year using restricted cubic splines.
Our findings are generally consistent with the limited epidemiological evidence linking diabetes prevalence (14
) and the incidence (16
) to traffic-related air pollution. The magnitude of reported associations is smaller than that of other studies (15
), but these were based on women only. Moreover, proximity of the residence to a road was significantly associated with risk for diabetes only in a cohort of women (Nurses’ Health Study), where no significant associations were detected with air pollution in a cohort of men (Health Professionals Follow-Up Study) (17
). Our results also suggest that the risk for diabetes associated with air pollution may be limited to women (). Sex-related differences in susceptibility to air pollution could be associated with physiological differences in inflammatory responses or with lifestyle and activity patterns. Brook et al. (15
) documented that women spent more time in and around the home and typically worked closer to home in Canada, contributing to smaller exposure misclassification and higher air pollution estimates, whereas no data in Denmark exist to support this.
Assigning individual exposure to air pollution with high spatial (address-specific) and temporal (residential addresses) resolution, historically back to 1971, is novel and a major strength of this study. We detected identical associations with long (mean levels since 1971 and 1991) and short (1-year mean at follow-up) exposure windows to NO2
. Puett et al. (17
) on the other hand failed to detect associations with the 1-year mean particulate pollution levels before diabetes incidence. In our data, associations with baseline 1-year mean NO2
levels and more “naïve” proxies based on traffic density around the residence were weaker, likely because of exposure misclassification in individuals who moved since baseline.
The association between air pollution and diabetes incidence in his cohort was strongest in nonsmokers (). Smoking and environmental tobacco smoke are confirmed risk factors for type 2 diabetes (21
), as corroborated by our () and previous data (16
). Inhalation of tobacco smoke triggers similar responses in the lung as air pollution, and these two related exposures share a plausible biological mechanism leading to glucose intolerance via systemic inflammation and inflammation in adipose tissue (22
). It has been hypothesized that preexisting low-level inflammation seen in smokers would further enhance the effects of air pollution (2
). In contrast, our results suggest a marginal contribution of air pollution to the development of diabetes in smokers and ex-smokers.
Furthermore, in contrast to existing evidence suggesting that individuals with preexisting cardiovascular disease would be more susceptible to the adverse effects of air pollution (2
), we found no statistically significant effect modification of the effects of air pollution by MI, hypertension, or hypercholesterolemia (). This finding contradicts the study of air pollution–triggered diabetes deaths, which reported effects only in individuals with preexisting cardiovascular disease (11
). Whereas diabetic patients have been found to be at higher risk of developing cardiovascular disease due to air pollution than nondiabetic people (7
), our results do not support the idea that cardiovascular disease enhances the susceptibility to air pollution and the associated risk of diabetes. Similarly, physical inactivity was found to be the expected risk factor for diabetes, whereas we only found association with air pollution among individuals who were physically active (). This observation may reflect smaller air pollution exposure misclassification in physically active individuals, because of more time spent outdoors and higher relative exposures and exaggerated air exchange rates during physical activity. It is possible that the association between diabetes and air pollution is only detected among individuals with a low prior risk, i.e., people who are physically active, representing a limited absolute risk, whereas this cannot be detected among individuals with a priori increased risk, suggesting additive patterns (23
). Among people with a low level of education, we found the expected increased risk for diabetes as well as enhanced effects of air pollution, although without significant interaction.
We have earlier reported significant positive association between hospital admissions for COPD and asthma and long-term exposure to NO2
in this cohort (24
). It can thus be argued that our results may be affected by diagnosis bias, due to more frequent screening for disease, including diabetes, among chronic respiratory disease patients, who are also more likely to live in areas with higher air pollution levels than the rest of the cohort (24
). However, this is unlikely, since we found only a marginally higher prevalence of asthma and COPD in diabetic people than in nondiabetic people (), and history of asthma or COPD hospitalizations was not significantly associated with increased risk of diabetes (results not shown). Furthermore, comorbidity with asthma or COPD did not statistically significantly modify the effect of air pollution on diabetes ().
Effects of air pollution were slightly enhanced in people with a high waist-to-hip (), corroborating the evidence from animal models, where exposure to air pollution led to glucose intolerance only in obese rats (7
). However, enhanced air pollution effects were not seen with increasing BMI. This is the first study of air pollution and diabetes to adjust for waist-to-hip ratio, which is an independent predictor of diabetes next to BMI. Excluding waist-to-hip ratio from the full model as others did (14
) resulted in even stronger associations with air pollution (HR 1.05 [95% CI 1.02–1.10]) as did exclusion of BMI (1.08 [1.04–1.12]). Mean levels of NO2
were weakly correlated with BMI (correlation coefficient r
= 0.03) and waist-to-hip ratio (r
= 0.02). Obesity, the most important risk factor for diabetes (1
), has been linked to air pollution (22
) and may be on the biological pathway between air pollution and diabetes. Our results remained unchanged when adding total energy intake, including energy due to alcohol intake, into the model.
The strengths of this study include the large prospective cohort with an objective assessment of diabetes incidence and well-defined confounders, with minimal possibility of recall and information bias. The main limitation is the exposure misclassification in modeled air pollution concentrations, since these are only proxies of personal exposure. This cohort lacked information on indoor exposures, use of air purifier and air conditioning, work address, working time, as well as commuting habits and personal outdoor activity patterns. However, lack of these parameters and resulting exposure misclassification is likely to be nondifferential with respect to diabetes diagnoses. The air pollution models used to assess NO2
levels were successfully validated (26
) and applied in epidemiological studies (24
). Another limitation is the lack of results from blood glucose tests from the National Health Insurance Registry, which precludes better ascertainment of diabetes among 1,163 people included in NDR solely due to blood glucose tests and more specific overall diabetes definition. Finally, there is little evidence coming from original research that supports biological plausibility of our findings (8
), and more studies will be needed to confirm or reject the link between air pollution and diabetes.
We detect weak positive associations between diabetes incidence and traffic-related air pollution at residence, and we add to the body of evidence that air pollution may be a risk factor for diabetes. We offer several novel findings that need to be reproduced. The effects of air pollution were strongest in nonsmokers and physically active people. Facing the emerging challenges in controlling the diabetes epidemic, a possibility that air pollution contributes to the diabetes burden may have a huge public health impact.