‘Source-specific’ air pollution and risk of stroke in Denmark

Abstract

Background

Long-term air pollution is a risk factor for stroke. Which types and sources of air pollution contribute most to stroke in populations is unknown. We investigated whether risk of stroke differed by type and source of air pollution.

Methods

We selected all persons aged >50 years and living in Denmark in the period 2005–17. We estimated running 5-year mean residential air-pollution concentrations of particulate matter <2.5 µm (PM_2.5), ultrafine particles (UFP), elemental carbon (EC) and nitrogen dioxide (NO₂). Pollutants were modelled as total air pollution from all emission sources, as well as apportioned into contributions from non-traffic and traffic sources. Hazard ratios (HRs) and CIs were estimated by using Cox proportional hazards models, adjusting for area-level and personal demographic and socio-economic covariates. We identified all primary strokes from hospital and mortality registers.

Results

The cohort numbered 2 million people and 94 256 cases of stroke. Interquartile ranges (IQR) of air pollution were associated with risk of stroke with HRs of 1.077 (95% CI: 1.061–1.094, IQR: 1.85 µg/m³) for PM_2.5, 1.039 (1.026–1.052, IQR: 4248 particles/cm³) for UFP, 1.009 (1.001–1.018, IQR: 0.28 µg/m³) for EC and 1.028 (1.017–1.040, IQR: 7.15 µg/m³) for NO₂. Traffic sources contributed little to the total exposure. HRs associated with air pollution from traffic were close to the null, whereas non-traffic sources tended to be associated with HRs higher than those for total air pollution, e.g. for non-traffic PM_2.5, the HR was 1.091 (1.074–1.108).

Conclusions

Air pollution, including UFP, was associated with risk of stroke. The risk appeared attributable mainly to air pollution from non-traffic sources.

Introduction

Air pollution is an established risk factor for cardiovascular disease,¹ with both experimental and human studies demonstrating that air pollution causes systemic and local inflammation and increases oxidative stress and plaque formation,² which are known risk factors for stroke. A meta-analysis of cohort studies found that 5 µg/m³ higher exposure to particulate matter of <2.5 µg (PM_2.5) was associated with a 6.3% (95% CI: 5.4–6.8%) higher risk of stroke.³ A later study, pooling six European cohorts, found a 10% increase in risk for stroke in association with the same increase in PM_2.5 exposure.⁴

A recent meta-analysis of nitrogen dioxide (NO₂) indicated a positive association with stroke mortality though the evidence was inadequate to conclude a causal relationship.⁵ Similarly, a meta-analysis of incidence/hospitalization suggested a weak positive association with NO₂, with heterogeneity between studies.⁶ The few studies on elemental carbon (EC)/black carbon (BC) or PM_2.5 absorbance have indicated weak or no association with stroke.^4,7–10

Ultrafine particles (UFP), PM < 0.1 µm, contribute little to PM_2.5 mass but their large number and small size provide a large reactive surface area and allow penetration beyond the respiratory tract, potentially causing health effects disproportionate to their mass.^11–13 Short-term UFP exposure has been associated with elevated blood pressure, systemic inflammation, autonomic tone and stroke, and studies have demonstrated that UFP can pass the blood–brain barrier and may also reach the brain via the olfactory nerve.^12–15 Studies on cardiovascular effects of long-term exposure to UFP have found associations with hypertension,¹⁶ myocardial infarction^10,17 and ischaemic heart disease.^18,19 The few studies on UFP and stroke have been hampered by small population size or limited quality of data on exposure and outcome. Two studies have found a positive association between short-term exposure to UFP and stroke incidence^14,20 and two studies have found an association with long-term residential exposure.^10,18

In 2021, the World Health Organization (WHO) specified new air quality guideline values for PM_2.5 (5 µg/m³) and NO₂ (10 µg/m³). For UFP and EC/BC, the WHO abstained from stating guideline values due to lack of studies.²¹ Also of relevance for regulation, the physical and chemical properties of air pollution, as well as the relative contribution to total exposure, differ between sources such as traffic and non-traffic. Quantification of the stroke burden of different sources will therefore facilitate more efficacious regulatory measures.²² Few large-scale studies have, however, undertaken such subdivisions of air-pollution sources.

We investigated total and source-specific UFP, PM_2.5, EC and NO₂, assessed by using a state-of-the-art model and stroke incidence in a nationwide cohort with detailed information on socio-demographic covariates.

Methods

All Danish citizens have since 1968 received a unique personal identification number. The number enables all citizens to be followed in all health and administrative registers.²³ We established complete residential histories for all persons living in Denmark or born in Denmark any time after 1979. Address histories ended when people left Denmark or had >14 consecutive days of incomplete address data. Eligible for our cohort were all (n = 2 048 282) born after 1920 (educational data are not available for those born before), living in Denmark on 1 January 2005 and who were ≥50 years of age any time between this date and 31 December 2017. We excluded 54 416 persons diagnosed with stroke before baseline and 22 620 with missing covariate information and followed up the cohort from 2005 until 2017 (Supplementary Figure S1, available as Supplementary data at IJE online).

By Danish law, entirely register-based studies do not require ethics approval.

Outcome

From the Danish national patient register²⁴ and the register of cause of death,²⁵ we identified all stroke cases (ICD8: 431–434, 436, ICD10 I61–I64) recorded as primary cause of death or admission in the period 1977–2017. We excluded prevalent cases at baseline and only counted first ever events identified from either register as cases.

Exposure

We geocoded all Danish addresses in the years 1995–2017.²⁶ We modelled air-pollution contributions at three scales using the Danish Eulerian Hemispheric Model (DEHM)/Urban Background Model (UBM)/AirGIS modelling system^27,28 including (i) the DEHM, covering the northern hemisphere, for the long-range transported regional background;²⁹ (ii) the UBM³⁰ for the local background at a resolution of 1×1 km² for the whole of Denmark, calculated from Danish emissions of air pollution;³¹ and (iii) the Operational Street Pollution Model (OSPM^®), modelling air pollution from traffic in streets carrying >500 vehicles/day, using detailed input parameters including traffic load, composition and emission factors, street and building configuration, and meteorology.^27,28 Contributions from the three scales were summed, thus providing a final address-level geographical resolution. We modelled PM_2.5, EC and NO₂ mass concentration and in a new addition to the modelling system, described in detail elsewhere,^32,33 we modelled particle number concentration, in the present paper denoted as UFP as these quantities are highly correlated. Using high-quality Danish emission inventories,³¹ we modelled air pollution both with and without the emissions from Danish road traffic to the local background (UBM) and the traffic on the address street. Subtracting the two modelling results from each other enabled apportioning total pollutant levels into the contribution emitted by Danish road traffic (in the present paper denoted as ‘traffic’) and the contribution emitted by non-traffic sources (including also contributions from road traffic in neighbouring countries). Contributions from road traffic to secondary particles (e.g. nitrate) are therefore not included in the road-traffic estimate. Road-traffic contributions to PM_2.5 were further apportioned as tailpipe and non-tailpipe contributions. Monthly mean air-pollutant levels were aggregated from modelled hourly concentrations. Combining these monthly data with individual cohort member address histories, we calculated running 5-year time-weighted average exposures. We used time-weighted averages to reflect the exact number of days lived at each address during the running 5-year period. The 5-year time-weighted average exposure period was our a priori main exposure metric but we also applied 1- and 10-year time-weighted average exposure periods.

The Danish DEHM/UBM/AirGIS modelling system generally compares well to measured air-pollution concentrations for Denmark. For example, at a street measurement station, the relative difference between the modelled and observed concentrations was –10% for NO₂, +11% for PM_2.5 and 6% for PNC (≈UFP),^33,34 and the correlation coefficients between the modelled and measured annual UFP concentrations were 0.95 at the street level³³ (for further details on models and validation, see Supplementary material, available as Supplementary data at IJE online).

Covariates

Potential confounders were selected a priori and annually updated data were extracted from Statistics Denmark. We included civil status (married/cohabiting, other), highest attained educational level (mandatory, secondary/vocational, medium/long), occupational status (high-level white-collar, low-level white-collar, blue-collar, unemployed, retired), country of origin (‘Danish origin’, having Danish citizenship or having at least one parent who has), personal income and household income (sex and calendar-year specific quintiles). Statistics Denmark provided annually updated information for all Danish parishes (2160 parishes in year 2017, mean area 16.2 km², median population 1032 persons) about the proportion of inhabitants with only basic education, with manual labour, with income in the lowest quartile, living in social housing, living in single-parent households, with a criminal record and with non-Western background. All persons with missing covariates were excluded.

Statistical methods

We used Cox proportional hazards models with age as the time axis to calculate hazard ratios (HRs) and 95% CIs for 5-year time-weighted average air pollution and stroke.

We followed cohort members from 50 years of age or 1 January 2005, whichever came last, until stroke, >14 consecutive days of unknown address, emigration, death or 31 December 2017, whichever came first. Associations were evaluated linearly per interquartile range (IQR) of exposure and per fixed increment, and categorically by percentiles of exposure (<10/reference, 10 to <25, 25 to <50, 50 to <75, 75 to <90, 90 to <95 and 95–100).

We analysed three models. Model 1 only adjusted for age, sex and calendar-year (in 2-year categories). Model 2 additionally adjusted for personal covariates: educational level, occupational status, civil status, country of origin and personal and household income. Finally, our main model (Model 3) had additional adjustment for area-level/contextual factors: proportion of parish inhabitants living in single-parent households, with only basic education, with manual labour, with income in lowest quartile, with non-Western background, living in social housing and with a criminal record.

Except for sex and country of origin, all variables were modelled time-dependently. That means that at any point in time during follow-up, the variables reflected exactly the conditions pertaining to that specific situation.

In sensitivity analyses, we evaluated effect modification by age. We also evaluated 1-year and 10-year time-weighted average periods. Statistical analysis was performed by using SAS 9.4 (SAS Institute Inc., Cary, NC, USA).

Results

The final cohort included 1 971 246 persons with 18 344 976 years of follow-up and 94 256 stroke events.

The cohort in total and above and below median PM_2.5 concentrations is described in Table 1 and air-pollution concentrations are provided in Table 2. High concentrations of PM_2.5 were associated with not being a blue-collar worker, being of non-Danish origin, being female, having older age, being retired and not living with a spouse (Table 1). Higher residential concentrations were also associated with living in a parish with more social housing and non-Western immigrants. The same pattern was observed for UFP (Supplementary Table S1, available as Supplementary data at IJE online).

Spearman rank correlations between PM_2.5, UFP, EC and NO₂ ranged from 0.71 to 0.93 (Supplementary Table S2, available as Supplementary data at IJE online). Concentrations averaged over 1, 5 and 10 years were highly correlated (all R > 0.85) (Supplementary Table S3, available as Supplementary data at IJE online). Sources other than traffic contributed most of the air pollution (Supplementary Figure S2, available as Supplementary data at IJE online). The correlation between PM_2.5 from tailpipe and non-tailpipe traffic sources was high (r = 0.94) (Supplementary Table S2, available as Supplementary data at IJE online).

Table 3 shows fully adjusted HRs for stroke per IQR of 1.077 (95% CI: 1.061–1.094) for PM_2.5, 1.039 (1.026–1.052) for UFP, 1.009 (1.001–1.018) for EC and 1.028 (1.017–1.040) for NO₂. Adjustment for covariates influenced these HRs only a little. Associations were most convincing for the air-pollution contributions from non-traffic sources (Table 3, Supplementary Figure S3a–d, available as Supplementary data at IJE online). The HRs for air pollution from traffic were only slightly higher than 1.00 with all 95% CIs spanning 1.00 (Table 3, Supplementary Figure S4a–d, available as Supplementary data at IJE online). HRs for PM_2.5 from tailpipe and non-tailpipe traffic sources differed little (Table 3). Figure 1 shows exposure–response associations without indication of lower thresholds and with levelling-off in the upper end of the exposure distributions of all four pollutants. Table 4 shows stronger associations in the older age groups.

Averaging air-pollution concentrations over 1 or 10 years produced similar risk estimates as the main analysis (Supplementary Table S5, available as Supplementary data at IJE online).

Discussion

In a cohort of 2 million Danes, UFP, PM_2.5, EC and NO₂ were associated with higher risk of incident stroke with the most convincing associations for the PM air-pollution contributions from non-traffic sources. The results indicated no lower exposure thresholds for associations and associations were strongest among the elderly. We found little evidence of associations with air pollution from traffic.

A meta-analysis of 20 studies showed a HR for stroke per 5 µg/m³ increase in PM_2.5 of 1.06 (95% CI: 1.05–1.07)³ and the international ELAPSE project showed a HR of 1.10 (95% CI: 1.01–1.21).⁴ For the same increase in exposure, we found a HR of 1.223 (95% CI: 1.175–1.273), which is higher than those of the previous studies, although CIs are overlapping with the ELAPSE study. There are several potential explanations for the higher risk estimates of the present study. First, as early-age strokes are relatively rare and may have distinct physiological causes,³⁵ we employed a lower age limit of 50 years. Many previous studies included also younger age groups, which may have lowered their risk estimates as we, in accordance with some previous studies, found stronger associations at older ages.^4,6,8 Second, data from our and other studies^4,36 suggested a stronger exposure–response association at lower exposure levels and since air-pollution levels in Denmark are comparatively low, this may have produced higher HRs in the present study. Third, if our air-pollution model provided less exposure misclassification than those of previous studies, e.g. due to a finer spatiotemporal resolution, this could have led to higher HRs.

We found UFP to be associated with stroke. A small Dutch cohort found risk estimates for UFP near identical to those of the present study (HR: 1.11, 95% CI: 0.88–1.41) per 10 000 particles/m³.¹⁰ The small German Heinz Nixdorf Recall Study found accumulation mode particle number count to be associated with risk of stroke.³⁷ A study using highway proximity as a proxy for UFP exposure indicated an association with self-reported stroke or myocardial infarction.¹⁸ Small size and crude exposure assessment, however, limit the value of these studies. The small particle diameters of UFP allow penetration beyond the respiratory tract, even reaching the brain,^38,39 and the small size combined with a large number provides a large reactive surface area, which may render them more potent toxins.⁴⁰ The present evidence suggests a positive association of stroke with long-term exposure to UFP, but more evidence is needed.

Previous studies on long-term exposure to the closely related entities EC/BC or PM_2.5-absorbance have not demonstrated a clear association with stroke. Some studies found no association with stroke,^7–10,41 whereas the ELAPSE study, in accordance with our results, showed an association with stroke.⁴ Further studies are needed to resolve whether EC is associated with stroke.

We found an HR for the association between NO₂ and stroke of 1.039 (95% CI: 1.023–1.056) per 10 µg/m³, which is compatible with the results of a meta-analysis of 12 studies on NO₂ and incidence of cerebrovascular disease providing a HR per 10 µg/m³ of 1.05 (95% CI: 1.00–1.11)⁶ and the results from pooled European cohort studies (ELAPSE project), which reported a HR for stroke incidence of 1.08 (95% CI: 1.04–1.10) per 10 µg/m³. A few studies have found higher risk estimates than ours.^42,43 Altogether, the literature supports an association between NO₂ and risk of stroke but the association could be due to correlated air pollutants or other risk factors. We found an association with NO₂ from non-traffic sources but no association with NO₂ from Danish traffic, which speaks against NO₂ per se as the causal factor.

For EC and even more so for NO_2, the association with stroke risk appeared weaker at the highest level of exposure. A possible explanation could be that these factors are only proxies for the true risk factors and the correlation with such true risk factors might be weaker in high-exposure settings such as central urban areas where air-pollution composition may vary greatly over short distances.

We found no association between primary emitted air pollutants from national road traffic and stroke. A review of health effects of air pollution from different sources found some evidence that particulate matter from traffic and from coal-powered power plants may be particularly harmful.⁴⁴ The overall conclusion was, however, that at present no clear hierarchy of harmfulness could be established between PM from different sources. This review combined short- and long-term exposure studies on all health end points. Different end points could, however, be associated with different attributes of air pollution. For example, we have recently, in the same cohort as the present, found that traffic-related air pollution was most closely associated with risk of diabetes⁴⁵ whereas we found that non-traffic PM was most closely related to risk of stroke. A review specifically on stroke found that the evidence linking traffic-related air pollution and stroke was of low to moderate quality.⁴⁶ Prospective studies looking specifically at source-specific long-term exposure to both traffic- and non-traffic-related air pollution and risk of stroke are few. A small (1391 incident cases) Swedish study of two urban cohorts found similar associations between stroke and air pollution from different sources (e.g. traffic, residential heating, shipping and industry).⁹ In a later expansion and update (3119 incident cases), BC but not PM_2.5 was associated with stroke risk.⁷ In two-pollutant models with local traffic exhaust and residential heating, the association was entirely due to traffic exhaust. The source-specific analysis focused on locally emitted particles and did not include long-distance transported PM, which contributed the majority of particles. Finally, in the German Heinz Nixdorf Recall study (118 cases), situated in the highly industrialized Ruhr district, traffic-specific PM was more strongly associated with stroke than was industry-specific PM.³⁷ The results of these studies appear to contradict our observation. They are, however, all situated in urban environments and cover limited geographical areas, which means that the variation in long-distance transported air pollutants will be comparatively small and therefore a larger part of the variation in exposure may be ascribed to traffic compared with our study, which covers an entire country, including also rural areas. Also, the composition of air pollution may differ between studies. We found that the association with myocardial infarction was primarily due to non-traffic sources, the proportion and composition of which may differ substantially between locations. Additionally, neither they nor we accounted for road-traffic noise, which is a risk factor for stroke;⁴⁷ the confounding potential may, however, be greater in these entirely urban studies showing an association with traffic-specific air pollution, which of course shares sources with road-traffic noise. Further, the chemical composition of non-traffic emissions differs considerably by source (e.g. inorganic components from agriculture vs organic components from combustion processes) and may thus lead to different associations with health end points. Finally, both the temporal and spatial precision of exposure modelling in these studies is lower than in our study and the small size could also increase the potential for chance findings. Our study indicated that air pollution from non-traffic sources could be the target of focused prevention strategies, but further large studies are required to confirm our findings.

In accordance with some previous studies,^4,6,8,37 we found stronger associations between air pollution and risk of stroke among the elderly. This could be caused by higher lung deposition rates due to lower respiratory function among the elderly and possibly a lower degree of exposure misclassification due to more time spent at the home address.⁴⁸ Additionally co-morbidity is likely to be higher in the elderly and several studies have documented a stronger association between air pollution and risk of stroke among people with higher levels of co-morbidity.^6,49 Co-morbidity and old-age frailty might be associated with a higher vulnerability to air pollution due to biological changes but are also likely to involve more contacts with the healthcare system, which might increase the likelihood of milder strokes being recorded. This could have influenced our results if it occurs more in highly exposed urban areas.

Strengths and limitations

A major strength of our study was the nationwide prospective cohort design. Another strength was the state-of-the-art modelling of time-varying, source-specific air-pollutant concentrations including UFP for an entire country. Additionally, we benefitted from reliable and detailed information on outcome, residential history and a wide array of covariates from near complete public registers.^23–25,50 However, assessing exposure to air pollution by modelling at the home addresses inevitably entails some exposure misclassification due to modelling uncertainty and lack of information on non-residential exposures. This will likely be independent of case status and a mixture of classical and Berkson error, and may affect the size and precision of risk estimates. In the same vein, it is unknown which exposure time window is most relevant for development of stroke. Applying a suboptimal time window could reduce the estimated risk. However, we found similar risk estimates when applying exposure time windows of 1, 5 and 10 years. Our modelling system only allowed separate estimation of primarily emitted national road-traffic pollution and we can thus not exclude that long-distance transported secondary traffic species contribute to stroke risk. Another limitation was the lack of information on lifestyle factors such as body mass index (BMI) and smoking habits. However, a large cohort study of 246 766 Danes on associations between air pollution and risk of stroke showed that in models with exactly the same exposure data and adjustment for register-based socio-demographic factors as in the present study, additional adjustment for lifestyle factors (smoking status and intensity, intake of fruit, intake of vegetables, intake of red meat, physical activity and BMI) did not appreciably alter the risk estimates.⁵¹ We identified cases from the Danish National Patient Register, where the positive predictive value of a stroke diagnosis for inpatients is good (83.5%).⁵² Lower accuracy for stroke subtype classification and a large proportion of cases being recorded as unspecified stroke precluded reliable division of cases by type. We excluded only 1% of cohort persons with missing data on covariates, primarily education. There were only small differences between the cohort and those excluded and since they only constituted 1% their exclusion is unlikely to have had appreciable impact (Supplementary Table S6, available as Supplementary data at IJE online).

We believe our results to be applicable to other Western populations of similar age composition. However, differences in sources and composition of air pollution should be considered before generalizing our results, as should exposure levels as our data indicate that the exposure–response association is stronger at low exposure levels.

Conclusion

Our study provides evidence that air pollution, including UFP, is associated with risk of stroke. The risk appeared attributable mainly to air pollution from sources other than local road traffic.

Ethics approval

According to Danish law, ethics permission or informed consent are not required for entirely register-based studies.

Data availability

The data that support the findings of this study are available from Statistics Denmark (and only at a secure server at Statistics Denmark). However, restrictions apply to the availability of these data, which were used under licence for the current study and so are not publicly available. Access to data requires permission from Statistics Denmark and the Danish Cancer Society.

Supplementary data

Supplementary data are available at IJE online.

Author contributions

A.H.P., M.S. and O.R.N. contributed to the study concept and design. M.S., A.H.P., U.A.H., L.M.F., M.K., J.H.C., J.B. and U.I. obtained, generated and/or cleaned data important for the analyses. M.S. did the statistical analyses and drafted the paper. All authors contributed to a critical revision of the manuscript and final approval of the version to be published.

Funding

This work was supported by the Health Effects Institute (HEI) (Assistance Award No. R-82811201). HEI is an organization jointly funded by the US Environmental Protection Agency (EPA) and certain motor vehicle and engine manufacturers. The contents of this article do not necessarily reflect the views of HEI or its sponsors, nor do they necessarily reflect the views and policies of the EPA or motor vehicle and engine manufacturers. The study funder was not involved in the design of the study; the collection, analysis and interpretation of the data; or writing the paper; and did not impose any restrictions regarding the publication of the paper.

Conflict of interest

None declared.

References