Concomitant exposure to air pollution, green space and noise, and risk of myocardial infarction: a cohort study from Denmark

Abstract

Aims

The three correlated environmental exposures (air pollution, road traffic noise, and green space) have all been associated with the risk of myocardial infarction (MI). The present study aimed to analyse their independent and cumulative association with MI.

Methods and results

In a cohort of all Danes aged 50 or older in the period 2005–17, 5-year time-weighted average exposure to fine particles (PM_2.5), ultrafine particles, elemental carbon, nitrogen dioxide (NO₂), and road traffic noise at the most and least exposed façades of residence was estimated. Green space around residences was estimated from land use maps. Cox proportional hazard models were used to estimate hazard ratios (HRs) and 95% confidence interval (CI), and cumulative risk indices (CRIs) were calculated. All expressed per interquartile range. Models were adjusted for both individual and neighbourhood-level socio-demographic covariates. The cohort included 1 964 702 persons. During follow-up, 71 285 developed MI. In single-exposure models, all exposures were associated with an increased risk of MI. In multi-pollutant analyses, an independent association with risk of MI was observed for PM_2.5 (HR: 1.026; 95% CI: 1.002–1.050), noise at most exposed façade (HR: 1.024; 95% CI: 1.012–1.035), and lack of green space within 150 m of residence (HR: 1.018; 95% CI: 1.010–1.027). All three factors contributed significantly to the CRI (1.089; 95% CI: 1.076–1.101).

Conclusion

In a nationwide cohort study, air pollution, noise, and lack of green space were all independently associated with an increased risk of MI. The air pollutant PM_2.5 was closest associated with MI risk.

Graphical Abstract

Lay Summary

The present study aimed to analyse their independent and cumulative association of the three correlated environmental exposures: air pollution, road traffic noise, and green space with MI.

• Air pollution, noise, and lack of green space were all independently associated with MI.

• Risk estimates for air pollution, noise, and lack of green space were similar, indicating that all may be equally relevant targets for regulatory measures.

Epidemiology, Cohort study, Myocardial infarction, Air pollution, Traffic noise, Green space

See the editorial comment for this article ‘Health and the environment: messages for the cardiologist’, by P.M. Mannucci, https://doi.org/10.1093/eurjpc/zwad343.

Introduction

Poor cardiovascular health has been associated with a lack of green space, air pollution, and noise. The European Environmental Agency has highlighted the two latter factors as the environmental exposures responsible for most negative health effects in the European Union (EU).¹ A large and rapidly increasing proportion of the world’s population lives in urban environments with high levels of these factors, leading to an increased risk of cardiovascular disease, including myocardial infarction (MI), a major cause of morbidity and death.

Air pollution is a complex mixture of substances from many sources that can cause oxidative stress, local and systemic inflammation, endothelial dysfunction, increased platelet activity, and accelerated plaque formation.² Air pollution is a leading cause of cardiovascular disease in both developed and developing countries, and a large number of studies have associated air pollution with MI, with the strongest evidence for particulate matter.³ With few exceptions,^4,5 subsequent studies have also observed an association with particulate matter < 2.5 µm (PM_2.5).^6,7 Ultrafine particles (UFPs), particulate matter < 0.1 µm, constitute a minute fraction of the mass-based PM_2.5 measure, but due to large numbers, they present a large reactive surface area, and their small size allows trans-penetration of the lung epithelium into the bloodstream and beyond causing inflammation of the vasculature and possibly triggering the atherosclerotic process.² They could therefore be particularly harmful. The few epidemiologic studies on long-term exposure to UFP have found an association with MI^8,4,9 and ischaemic heart disease.¹⁰ Elemental carbon (EC) or closely related entities, such as PM_2.5 absorbance, have been linked with coronary heart disease¹¹ and MI.⁸ A Swedish study did, however, not find an association with ischaemic heart disease.¹² Recent studies have provided conflicting results on the association of nitrogen dioxide (NO₂) and MI.^4,5,7,9

The chemical composition and toxicity of air pollution may differ by emission source,¹³ but most studies estimate only total exposure from all sources.

There is increasing evidence linking noise with cardiometabolic disease including MI.¹⁴ Based on experimental and epidemiological evidence, proposed pathways from noise to cardiovascular disease include stress-induced activation of the hypothalamic–pituitary–adrenal axis and sleep disturbance.¹⁴

Green space, i.e. areas covered by vegetation, has been associated with better cardiovascular health and reduced risk of MI.¹⁵ Proposed causal pathways included the promotion of physical activity and reduced stress. Furthermore, the presence of green space is associated with fewer noise and air pollution sources, and wooded areas may, to some extent, filter and bar noise and air pollution.

Mainly, the health effects of air pollution, noise, and green space have been studied separately. However, since they are correlated aspects of urban living, they may act as confounders for each other. They should, therefore, be analysed simultaneously to establish the best possible evidence about independent and combined effects on public health. We, therefore, aimed to investigate this in a nationwide cohort study of long-term exposure to PM_2.5, UFP, EC, and NO₂, noise at most and least exposed façade, and two measures of green space.

Methods

All Danish citizens can be followed in health and administrative registers via a unique personal identification number issued to all Danish citizens at birth since 1968. For all persons ever living in Denmark after 1979, we established address histories until 31 December 2017, emigration from Denmark, or >14 consecutive days of incomplete address data. We then created a cohort of all (n = 2 048 282) living in Denmark on 1 January 2005 and who were ≥50 years of age at any time between this date and 31 December 2017, when the study ended. Due to the availability of educational information, participants were required to be born in 1921 or later. The lower limit of 50 was employed as MI is rare in the young.

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

Outcome

We defined cases as persons with MI (ICD8: 410; ICD10: I21) recorded as the primary cause of death or admission in the Danish Register of Death or the Danish National Patient Register.¹⁶ We excluded persons diagnosed with MI before study entry. Cases diagnosed during the study period were censored at first MI.

Air pollution

Total concentrations of PM_2.5, UFP, EC, and NO₂ and their source-specific contributions [traffic (on roads) and non-traffic] were estimated at the front door of all Danish residential addresses. This was done with the Danish Eulerian Hemispheric Model (DEHM)/Urban Background Model (UBM)/AirGIS modelling system,¹⁷ which consists of three air pollution sub-models: (i) the DEHM—Northern Hemisphere, long-range transported regional background; (ii) the UBM—Danish air pollution emissions at 1 km × 1 km resolution¹⁸; and (iii) the Operational Street Pollution Model (OSPM)—contributions from traffic on the address street, with input on traffic compositions and intensity, emission factors, meteorology, and street and building configurations. The final address level estimates are formed by integrating the three models such that UBM obtains boundary conditions from DEHM and OSPM from UBM. Modelling of UFP by means of particle number concentration, a closely related quantity, has recently been implemented in the modelling system, which has been documented and validated elsewhere.^19,20 The model performs well with high correlations between measured and modelled concentrations at the address level: R: 0.67–0.85 for PM_2.5; 0.77–0.79 for EC; 0.60–0.80 for NO₂^21,22; and 0.86–0.95 for UFP.^19,20

We modelled total air pollution, including contributions from all Danish and international traffic and non-traffic sources. Using the high-quality Danish emission inventories,²³ using the detailed Selected Nomenclature for Air Pollution (SNAP) categorization of sources, we could also exclude all road traffic contributions from the DEHM and OSPM models. This provided an exposure estimate for each address that excluded all contributions from road traffic within 25 km. This quantity is denoted non-traffic in the present paper, although it includes road traffic contributions from further away than 25 km. The exact SNAP categories contributing to total and non-traffic air pollution are tabulated in Supplementary material online, Table S1. Air pollution was modelled as hourly concentrations accounting for short-term variations in weather and was then summarized into 1-, 5-, and 10-year averages based on the person-specific address histories. In an earlier investigation of the present cohort, we found that air pollution from traffic sources only constitutes a small proportion of total air pollution concentrations and that non-traffic emissions were the primary drivers of the association of air pollution and MI. For example, for PM_2.5, the hazard ratio (HR) for MI was 1.051 [1.032, 1.069, interquartile range (IQR) = 1.63 µg/m³] for non-traffic contributions and 1.011 (1.003, 1.018, IQR = 0.37 µg/m³) for traffic contributions. A separate analysis of the traffic contributions is, therefore, not included in the present paper.

Road traffic noise

We estimated road traffic noise at the most (LdenMax) and least (LdenMin) exposed façade of all Danish dwellings. We used the Nordic prediction method to calculate the address-specific A-weighted equivalent sound pressure for day, evening, and night.²⁴ This metric of perceived loudness considers the sensitivity of the human ear at different frequencies. The noise model includes input for all Danish road links on road type, annual average daily traffic load, travel speed, and light/heavy vehicle distribution. Furthermore, the model considers building floor and screening effects from terrain, buildings, and noise barriers. Both 1st- and 2nd-order noise reflections were included. Exposure was modelled for the years 2000, 2005, 2010, and 2015. For all other years between 2000 and 2017, exposure was quantified by linear interpolation. Values < 35 dB were set to 35 dB as noise below this level is unlikely to be discernible above background noise. In a Nordic prediction method validation study, the mean difference between measured and estimated road LdenMax was 0.3 dB.²⁵ As for air pollutants, noise was also summarized into 1-, 5-, and 10-year averages based on the person-specific address histories.

Surrounding green space

BaseMap is a digital map that classifies all land use in Denmark in 2016 at a 10 m × 10 m resolution, based on 36 classes of land use²⁶ (see Supplementary material online, Figure S1 and Table S2). We used these data to define two green space measures: (i) Green1000m: proportion of recreational areas, forests, and open nature areas within 1000 m of the residence and (ii) Green150m: proportion of gardens (BaseMap category ‘low built up’), green areas around multi-story buildings (BaseMap category ‘high built up’), agricultural areas, recreational areas, forests, and open nature areas within 150 m of residence. Green1000m was conceived as an indicator of accessible, attractive green space that might promote physical activity, whereas Green150m was conceived as an indicator of potential stress-reducing green areas near the residence. To express the effect as an increased risk, we used NonGreen1000m (1-Green1000m) and NonGreen150m (1-Green150m) in analyses.

Covariates

Statistics Denmark provided annually updated data on a priori selected potential confounders: highest attained educational levels (mandatory, secondary/vocational, and medium/long), occupational status (high-level white-collar, low-level white collar, blue collar, unemployed, and retired), marital status (married/cohabiting, other), country of origin (‘Danish origin’, i.e. having Danish citizenship or having at least one parent who has vs. ‘other’), personal income, and household income (sex and calendar year-specific quintiles). We also collected data for all parishes in Denmark (2160 parishes in 2017, mean area 16.2 km², and median population 1032 persons). These data included 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 covariate information were excluded.

Statistical analysis

We calculated Spearman correlations between 5-year average pollutants for the entire study period. Cox proportional hazard models with age as time scale were used to calculate HRs and 95% confidence intervals (CIs) for 5-year time-weighted average exposures and MI. We evaluated associations linearly per IQR of exposure. Models were adjusted for individual and area-level factors. Individual factors were age, sex, calendar year (in 2-year categories), educational level, occupational status, civil status, country of origin, personal income, and household income. Area-level factors were proportion of parish inhabitants living in single-parent households, with only basic education, with manual labour, with income in the lowest quartile, with non-Western background, living in social housing, and with a criminal record. Study entry was at 50 years of age or 1 January 2005, whichever came last. Censoring was at diagnosis of MI, >14 consecutive days of unknown address, emigration, death, or 31 December 2017, whichever came first. All variables except sex and country of origin were included as time-varying variables: air pollution and noise as running 5-year averages over all addresses held in the preceding 5 years and green space and neighbourhood- and individual-level indicators reflecting conditions at any given time. In sensitivity analyses, we also investigated 1- and 10-year averaging periods.

For each of the three exposure domains, air pollution (PM_2.5, UFP, EC, and NO₂), noise (LdenMax and LdenMin), and green areas (NonGreen1000m and NonGreen150m), we undertook single, two, and multi-pollutant analyses exploring all combinations of pollutants within each domain. Pollutants showing consistent associations with MI in all these analyses were included in the multi-pollutant analysis combining the three domains.

When investigating total air pollution contributions, the resulting multi-pollutant model included PM_2.5, UFP, LdenMax, LdenMin, NonGreen1000, and NonGreen150. When investigating non-traffic air pollution contributions, the resulting multi-pollutant model included PM_2.5, UFP, NO₂, LdenMax, LdenMin, NonGreen1000, and NonGreen150.

We calculated the cumulative risk index (CRI) per IQR to quantify the cumulative burden of multiple exposures:

$$\mathrm{CRI}=\exp \left[{\displaystyle \sum _{p=1}^p}{\hat{\beta }}_p{x}_p\right]\equiv \exp ({\hat{\beta }}^{\mathrm{\prime }}{x}^{\mathrm{\prime }}),$$

where ${\hat{\beta }}^{\mathrm{\prime }}=({\hat{\beta }}_1,\dots {\hat{\beta }}_p)$ is the log(HR) from a Cox model with p exposures estimated at x_p concentrations.^27,28

Statistical analysis was performed in SAS 9.4 (SAS Institute Inc., Cary, NC, USA).

Results

From the eligible population (n = 2 048 282), we excluded prevalent cases of MI at baseline (n = 60 967) and persons with missing information about one or more covariates (n = 22 613), leaving a final cohort of 1 964 702 persons. During 18 309 318 years of follow-up, the cohort accrued 71 285 cases of MI. The dwellings of the included cohort member were located in 1 306 854 buildings. The cohort is described at baseline in Table 1 and neighbourhood characteristics are provided in Supplementary material online, Table S3. Non-traffic sources contributed most to total concentrations for all air pollutants, and the correlation between total and non-traffic concentrations was high (Rs: 0.82–0.96) (Table 2). The correlations between total air pollutant concentrations were in the range of 0.71–0.93, and between non-traffic contributions correlations were in the range of 0.63–0.85. Air pollution was only weakly to moderately correlated with noise (Rs: 0.07–0.53). Noise and air pollution were weakly correlated with NonGreen150m. NonGreen1000m was very weakly related to other factors.

Table 1

Baseline individual-level socio-demographic characteristics and 5-year exposure levels among the study population of 1 964 702 persons

Baseline characteristics	%	Median (5–95 percentile)
Women	53
Age (years), median		58 (50–79)
Country of origin
Denmark	98
Other	2
Civil status
Married/cohabiting	73
Other	27
Education
Mandatory	36
Secondary or vocational	45
Medium or long	19
Occupational status
White collar, high level	10
White collar, low level	15
Blue collar	30
Unemployed	4
Retired	41
Personal income, quintiles
1st (low)	25
2nd–4th	55
5th (high)	20
Household income, quintiles
1st (low)	21
2nd–4th	54
5th (high)	25
Air pollution levels (5-year mean)
PM2.5 total (µg/m3)		11.2 (8.7–12.6)
PM2.5 non-traffic (µg/m3)		10.9 (8.52–11.7)
UFP total (particles/cm3)		11 099 (7212–17 232)
UFP non-traffic (particles/cm3)		9755 (6821–13 061)
EC total (µg/m3)		0.7 (0.4–1.1)
EC non-traffic (µg/m3)		0.5 (0.4–0.7)
NO2 total (µg/m3)		15.2 (9.3–27.3)
NO2 non-traffic (µg/m3)		11.2 (7.8–14.1)
Road traffic noise (dB)
Most exposed façade (LdenMax)		55 (40–68)
Least exposed façade (LdenMin)		44 (33–56)
Surrounding green space (%)
Green space within 150 ma		58.8 (17.9–87.5)
Publicly accessible green space within 1000 mb		13.1 (2.9–36.3)

CI, confidence interval; EC, elemental carbon; HR, hazard ratio; IQR, interquartile range; NO₂, nitrogen dioxide; UFP, ultrafine particle.

^a(Green150m) Percentage of areas < 150 m from residence, not classified as agricultural areas, household gardens, recreational areas, forests, and open nature areas. For analysis, the complementary quantity NonGreen150m = 1 − Green150m was used.

^b(Green1000m) Percentage of areas < 1000 m from residence that are not publicly accessible green areas (i.e. not recreational areas, forests, and open nature areas). For analysis, the complementary quantity NonGreen1000m = 1 − Green1000m was used.

Table 2

Spearman rank correlations between 5-year time-weighted average air pollution concentration at residences of all cohort members, Denmark, 2005–17

	Total PM2.5	Total UFP	Total EC	Total NO2	Non-traffic PM25	Non-traffic UFP	Non-traffic EC	Non-traffic NO2	Ldenmaxa	Ldenminb	NonGreen150mc	NonGreen1000d
Total PM2.5	1	0.77	0.71	0.76	0.96	0.77	0.69	0.80	0.20	0.26	0.23	0.02
Total UFP	0.77	1	0.91	0.92	0.67	0.96	0.86	0.82	0.25	0.44	0.37	−0.06
Total EC	0.71	0.91	1	0.93	0.56	0.80	0.9	0.69	0.34	0.51	0.40	−0.07
Total NO2	0.76	0.92	0.93	1	0.61	0.80	0.78	0.82	0.39	0.53	0.40	−0.05
Non-traffic PM25	0.96	0.67	0.56	0.61	1	0.72	0.63	0.77	0.07	0.16	0.13	0.03
Non-traffic UFP	0.77	0.96	0.8	0.80	0.72	1	0.85	0.84	0.12	0.32	0.29	−0.05
Non-traffic EC	0.69	0.86	0.90	0.78	0.63	0.85	1	0.71	0.13	0.37	0.29	−0.07
Non-traffic NO2	0.80	0.82	0.69	0.82	0.77	0.84	0.71	1	0.15	0.31	0.26	−0.00
Ldenmaxa	0.20	0.25	0.34	0.39	0.07	0.12	0.13	0.15	1	0.48	0.19	0.05
Ldenminb	0.26	0.44	0.51	0.53	0.16	0.32	0.37	0.31	0.48	1	0.34	−0.08
NonGreen15°c	0.23	0.37	0.40	0.40	0.13	0.29	0.29	0.26	0.19	0.34	1	0.01
NonGreen1000d	0.02	−0.06	−0.07	−0.05	0.03	−0.05	−0.07	−0.00	0.05	−0.08	0.01	1

EC, elemental carbon; NO₂, nitrogen dioxide; UFP, ultrafine particle.

^aLdenMax: Traffic noise at most exposed façade.

^bLdenMin: Traffic noise at least exposed façade.

^cNonGreen150m: Percentage of areas < 150 m from residence, not classified as agricultural areas, household gardens, recreational areas, forests, and open nature areas.

^dNonGreen1000m: Percentage of areas < 1000 m from residence that are not publicly accessible green areas (i.e. not recreational areas, forests, or open nature areas).

In single-pollutant models, adjusted only for age and sex, all air pollutants were associated with a reduced risk of MI (see Supplementary material online, Table S4). In fully adjusted single-pollutant models, all total air pollutants were associated with a higher risk of MI (Table 3). PM_2.5 and UFP were consistently associated with elevated risk in models adjusting for one or more other air pollutants. In models including both UFP and PM_2.5, the latter exhibited the least change compared with the single pollutant model. EC and NO₂ were reduced to or below the null in all models including UFP or PM_2.5.

Table 3

Single and multi-pollutant models of the air pollutants PM_2.5, ultrafine particles, nitrogen dioxide, and elemental carbon in relation to risk of myocardial infarction

Total concentrations of air pollutants	PM2.5 total	UFP total	EC total	NO2 total
	HR (95% CI)a	HR (95% CI)a	HR (95% CI)a	HR (95% CI)a
	per IQR: 1.85 µg/m3	per IQR: 4248 #/cm3	per IQR: 0.28 µg/m3	per IQR: 7.15 µg/m3
Single pollutant models
PM2.5	1.053 (1.035–1.071)
UFP		1.040 (1.025–1.055)
EC			1.009 (1.000–1.019)
NO2				1.027 (1.013–1.040)
Two-pollutant models
PM2.5 + UFP	1.039 (1.015–1.063)	1.017 (0.997–1.037)
PM2.5 + EC	1.072 (1.048–1.096)		0.984 (0.970–0.997)
PM2.5 + NO2	1.055 (1.030–1.080)			0.998 (0.980–1.016)
UFP + EC		1.063 (1.041–1.086)	0.979 (0.964–0.994)
UFP + NO2		1.038 (1.016–1.060)		1.002 (0.983–1.021)
EC + NO2			0.99 (0.974–1.007)	1.036 (1.016–1.056)
Three-pollutant models
PM2.5 + UFP + EC	1.055 (1.030–1.081)	1.044 (1.020–1.069)	0.967 (0.950–0.984)
PM2.5 + UFP + NO2	1.046 (1.020–1.072)	1.024 (1.002–1.047)		0.987 (0.967–1.008)
PM2.5 + EC + NO2	1.066 (1.039–1.093)		0.979 (0.962–0.996)	1.011 (0.989–1.034)
UFP + EC + NO2		1.056 (1.031–1.082)	0.974 (0.957–0.992)	1.014 (0.992–1.036)
Four-pollutant models
PM2.5 + UFP + EC + NO2	1.056 (1.030–1.084)	1.045 (1.020–1.071)	0.968 (0.950–0.986)	0.997 (0.974–1.021)

CI, confidence interval; EC, elemental carbon; HR, hazard ratio; IQR, interquartile range; NO₂, nitrogen dioxide; UFP, ultrafine particle.

^aAll estimates are per interquartile increase in exposure and adjusted for age, sex, calendar year, civil status, individual and family income, country of origin, occupational status, education, and neighbourhood-level percentage of population with low income, with only basic education, who are unemployed, with manual labour, with non-Western background, with criminal record, who are single-parents, and who live in social housing

For non-traffic air pollution, all air pollutants but EC were associated with MI in single-pollutant models, with the highest HRs for NO₂ (HR per IQR: 1.048; 95% CI: 1.034–1.062) and PM_2.5 (HR per IQR: 1.051; 95% CI: 1.032–1.069) (Table 4). In models with two or three air pollutants, NO₂, UFP, and PM_2.5 remained positively associated with MI. In a model including all four air pollutants, PM_2.5, NO₂, and UFP were all associated with MI.

Table 4

Single-, two-, three-, and four-pollutant models of the non-traffic contribution of PM_2.5, ultrafine particles, nitrogen dioxide, and elemental carbon in relation to risk of myocardial infarction

Non-traffic contribution of air pollutants	PM2.5 non-traffic	UFP non-traffic	EC non-traffic	NO2 non-traffic
	HR (95% CI)a	HR (95% CI)a	HR (95% CI)a	HR (95% CI)a
	per IQR: 1.63 µg/m3	per IQR: 2769 #/cm3	per IQR: 0.12 µg/m3	per IQR: 2.68 µg/m3
Single-pollutant models
PM2.5	1.051 (1.032–1.069)
UFP		1.034 (1.022–1.046)
EC			1.001 (0.996–1.007)
NO2				1.048 (1.034–1.062)
Two-pollutant models
PM2.5 + UFP	1.032 (1.010–1.054)	1.022 (1.008–1.037)
PM2.5 + EC	1.062 (1.041–1.083)		0.992 (0.984–1.000)
PM2.5 + NO2	1.018 (0.996–1.041)			1.04 (1.023–1.057)
UFP + EC		1.050 (1.035–1.065)	0.987 (0.979–0.995)
UFP + NO2		1.009 (0.993–1.026)		1.041 (1.023–1.059)
EC + NO2			0.995 (0.988–1.003)	1.051 (1.036–1.065)
Three-pollutant models
PM2.5 + UFP + EC	1.042 (1.020–1.065)	1.038 (1.022–1.055)	0.983 (0.974–0.992)
PM2.5 + UFP + NO2	1.016 (0.993–1.039)	1.007 (0.990–1.023)		1.036 (1.016–1.056)
PM2.5 + EC + NO2	1.028 (1.004–1.053)		0.992 (0.983–1.000)	1.04 (1.023–1.057)
UFP + EC + NO2		1.023 (1.003–1.043)	0.990 (0.981–0.999)	1.036 (1.017–1.055)
Four-pollutant models
PM2.5 + UFP + EC + NO2	1.029 (1.004–1.054)	1.023 (1.003–1.044)	0.986 (0.976–0.996)	1.025 (1.004–1.046)

CI, confidence interval; EC, elemental carbon; HR, hazard ratio; IQR, interquartile range; NO₂, nitrogen dioxide; UFP, ultrafine particle.

^aAll estimates are given per interquartile increase in exposure and adjusted for age, sex, calendar year, civil status, individual and family income, country of origin, occupational status, education, and neighbourhood-level percentage of population with low income, with only basic education, who are unemployed, with manual labour, with non-Western background, with criminal record, who are single-parents, and who live in social housing.

In single-pollutant models, noise at the least and most exposed façade was associated with a higher risk of MI. Mutual adjustment left LdenMax unchanged, whereas LdenMin was attenuated (Table 5). NonGreen150m and NonGreen1000m were both associated with risk of MI in single-pollutant models. NonGreen1000 was slightly attenuated in two-pollutant models, whereas NonGreen150m was unchanged (Table 5).

Table 5

Single- and two-pollutant models of road traffic noise and lack of green space in relation to risk of myocardial infarction

	Noise, most exposed façade (LdenMax)	Noise, least exposed façade (LdenMin)
	HR (95% CI)a	HR (95% CI)a
	per IQR: 10.6 dB	per IQR: 9.5 dB
Single-pollutant models
LdenMax	1.033 (1.023–1.043)
LdenMin		1.027 (1.014–1.039)
Two-pollutant model
LdenMax + LdenMin	1.030 (1.018–1.041)	1.008 (0.994–1.022)
	NonGreen1000mc	NonGreen150mb
	HR (95% CI)a	HR (95% CI)a
	per IQR: 12.6%	per IQR: 18.1%
Single-pollutant models
NonGreen1000md	1.014 (1.004–1.023)
NonGreen150me		1.025 (1.016–1.033)
Two-pollutant model
NonGreen150md + NonGreen1000me	1.010 (1.001–1.02)	1.024 (1.015–1.032)

CI, confidence interval; HR, hazard ratio; IQR, interquartile range.

^aAll estimates are per interquartile increase in exposure and adjusted for age, sex, calendar year, civil status, individual and family income, country of origin, occupational status, education, and neighbourhood-level percentage of population with low income, with only basic education, who are unemployed, with manual labour, with non-Western background, with criminal record, who are single-parents, and who live in social housing.

^bLdenMax: Traffic noise at most exposed façade.

^cLdenMin: Traffic noise at least exposed façade.

^dNonGreen1000m: Proportion of area < 1000 m from the residence that are not publicly accessible green areas (i.e. not recreational areas, forests, and open nature areas).

^eNonGreen150m: Proportion of areas < 150 m from the residence, not classified as agricultural areas, household gardens, recreational areas, forests, and open nature areas.

The pre-screened multi-exposure domain model with total air pollution concentrations included PM_2.5, UFP, LdenMax, and LdenMin and NonGreen150m and NonGreen1000m, and in the mutually adjusted model, a positive association only remained for NonGreen150m (HR per IQR: 1.018; 95% CI: 1.010–1.027), LdenMax (HR per IQR: 1.024; 95%CI: 1.012–1.035) and PM_2.5 (HR per IQR: 1.026; 95% CI: 1.002–1.050) (Figure 1; Supplementary material online, Table S5). In this model, the CRI per IQR was 1.089 (95% CI: 1.076–1.101). The CRI per IQR for the corresponding model with non-traffic air pollution concentrations was 1.104 (95% CI: 1.091–1.118) (Figure 1; Supplementary material online, Table S6). Non-traffic NO₂, LdenMax, and NonGreen150m were the ones exhibiting the least change in this multi-exposure model.

Figure 1

Hazard ratios of myocardial infarction per interquartile range increases in road traffic noise, lack of green space, and (A) total concentration of air pollution and (B) non-traffic contribution to air pollution in single- (circles) and multi-pollutant (squares) models. CI, confidence interval; CRI, cumulative risk index; UFP, ultrafine particle;

HRs differed little when we investigated 1- or 10-year time averaging periods (see Supplementary material online, Table S7).

Discussion

In a nationwide cohort with 71 285 cases, an increased risk of MI was associated with noise at the most exposed façade, lack of green space, and air pollution. Air pollution was best captured as PM_2.5 in models including total air pollution and NO₂ in models including only the non-traffic air pollution contributions. In the total air pollution model, air pollution, noise, and lack of green space contributed similarly to the CRI.

Air pollution

In Denmark, higher socio-economic status is positively associated with higher levels of air pollution at the residence,²⁹ and after adjusting for socio-economic indicators, air pollution was associated with MI in single-pollutant models. The associations were relatively weak. The HR per IQR for PM_2.5 and UFP correspond to rate differences of about 116 and 76 cases per 100 000 person-years, and the HRs for EC and NO₂ represent around 39 and 53 additional cases per 100 000 person-years.³⁰ In our multi-exposure models, adjusting for other exposures had the least effect on PM_2.5, which accords with the existing evidence for an association being more consistent for PM_2.5 than for NO₂ and other pollutants.³ However, studies performing two- and multi-pollutant analyses have produced conflicting results: two studies with PM_2.5 and NO₂ found PM_2.5 to be the dominant exposure.^7,31 A Canadian and a Dutch study found that in two-pollutant models combining UFP, NO₂, PM_2.5, and PM_2.5 absorbance, UFP was least affected by adjustment for other air pollutants.^4,9 Potential explanations for the conflicting results between studies include different exposure modelling qualities where associations shift towards the better-modelled pollutant. Also, the composition of air pollution may differ between studies. We have previously shown that the association with MI in our cohort was primarily due to non-traffic sources,⁸ the proportion of which may differ substantially between locations, mainly since some studies cover single cities, whereas others cover entire countries or states. On the same note, the primary non-traffic sources of PM may differ between locations; e.g. for Denmark, the primary national non-traffic emission source of PM_2.5 is non-industrial combustion, with residential heating as the primary contributor in this sector. We found that the association between PM_2.5 and MI remained after adjustment for green space. In a nationwide Dutch study, PM_2.5, but not NO_2, remained associated with ischaemic heart disease after adjusting for green space,³² whereas in an Italian study, an association between NO₂ and MI was unaffected by adjustment for noise and green space.³³ Further studies will be needed to assess the confounding potential of green space in air pollution studies.

Non-traffic air pollution

In a small study including UFP and PM_2.5, HRs for ischaemic heart disease per IQR for diesel and gasoline combustion were similar to HRs for total particle concentrations.¹⁰ Other studies have, however, associated ischaemic heart disease or MI primarily with non-traffic sources such as coal combustion or residential heating.^12,34 In the present cohort, we have previously found that per IQR, PM_2.5, UFP, and NO₂ from non-traffic sources were associated with higher HRs for MI than the same pollutants from traffic sources.⁸ In our multi-exposure model, including these air pollutants from non-traffic sources, NO₂ dominated and was the air pollutant least affected by adjustment for other exposures.

We have previously demonstrated that the MI risk from the same quantity of NO₂ differed between NO₂ from traffic and non-traffic sources indicating that NO₂ acts as a proxy for correlated exposures rather than being a causative agent.⁸ So that NO₂ from non-traffic sources dominates our analysis suggesting that our non-traffic NO₂ measure is a better proxy for the causal air pollutants than non-traffic UFP or PM_2.5.

EC total and from non-traffic sources were not or only weakly associated with MI in single pollutant models. When including other air pollutants in models with EC, the HR for these other pollutants generally increased, while EC exhibited a negative association with MI risk. This suggests that EC is not a significant risk factor for MI in our study and that single pollutant results for EC are affected by collinearity with other pollutants.

Road traffic noise

Noise at the most exposed façade was associated with a higher risk of MI both before and after adjustment for air pollution and green space. Other studies have also found an association with road traffic noise after adjusting for air pollution.^35,36 Only two previous studies associating road traffic noise and MI have adjusted for both air pollution and green space. A Dutch study found no association between road traffic noise and self-reported ‘heart attack’,⁶ and an Italian study found no association between noise and MI hospitalization.³³ In neither study did adjustment for other environmental factors alter the results. The Dutch study was cross-sectional and relied on self-reported outcome data, and both studies had less precise noise assessment than the present study, which may have contributed to the null results.

In single-pollutant models, noise both at the least and most exposed façade was associated with MI, with the strongest risk per IQR related to noise at the most exposed façade. In our models, including both noise measures, HRs for noise at the least exposed façade were reduced to the null.

Multiple studies have shown a link between night-time noise and sleep quality, which could be on a potential pathway to MI.¹⁴ However, under the assumption that most people choose a bedroom with low levels of ambient noise and that LdenMin is, therefore, closer associated with night-time noise, our result corroborates previous epidemiological studies not finding a stronger association for night-time than daytime or total noise at the most exposed façade.^35,36 Altogether, our data suggest that the effect of noise on MI is better captured at the most exposed façade.

Green space

Green space around the residence was associated with a reduced risk of MI, which accords with a recent review and meta-analysis where green space was inversely correlated with ischaemic heart disease mortality and cardiovascular mortality.¹⁵ We found a stronger association with the lack of green space close to the house (<150 m) than at a greater distance (<1000 m). Gardens (and fields) were only included in our 150 m metric. A UK Biobank study, which also found stronger associations at short distances, found that residential gardens were the type of green closest associated with reduced cardiovascular mortality.³⁷ This could indicate that the primary effect of greenness on the risk of MI relates to stress-reducing potential. Greenness has been associated with reduced levels of volatile organic compounds and reduced effect of air pollutants on arterial stiffness.³⁸ It could, therefore, also be that the observed stronger effect of vegetation close to the residence (<150 m) relates to reduced or mitigated effects of other environmental exposures. On the other hand, two recent European studies did not find substantial differences in risk estimates depending on inclusion of gardens or at different distances.^32,39 Our study and a range of previous studies have found that though attenuated, the association of green space with the risk of MI remained after adjustment for noise and/or air pollution.^32,37,39 So, while lack of green space appears to be independently associated with the risk of MI, the optimal green space metric remains elusive.

CRI

In the final multi-pollutant model, air pollution, noise, and lack of green space all remained positively associated with risk of MI indicating that they each represent independent risk factors for MI. The contribution of each to the CRI was of similar magnitude indicating that they per IQR contribute equally to the total risk and that intervention measures to reduce MI incidence are merited for all these exposures. In the developing world, all three exposures are generally worsening whereas air pollution levels have decreased in developed countries due to regulation and industry relocation. Access to green space can vary greatly even over short distances and there is increasing awareness of the benefits to health and local climate. Climate change initiatives may improve air quality and green space but are less likely to reduce road traffic noise, which has increased with growing traffic. With political will, noise reduction measures such as specialized tires, asphalt, and noise barriers can be implemented.

Strengths and limitations

The nationwide prospective cohort design was a major strength of this study. Another strength was the comprehensive registers of the Danish population, which provided detailed information on MI diagnoses and a comprehensive set of covariates at individual and area levels as well as the precise residential history of all participants.¹⁶ Furthermore, the simultaneous analysis of multiple air pollutants and measures of road traffic noise and green space, both as exposures and confounders, was a strength of the present study. We are only aware of two previous MI cohort studies to perform such analyses.^6,32 A further strength was the similar spatial resolution of all exposures. A 10 m × 10 m resolution map of Denmark was used to assess green space and validated address level; state-of-the art models were applied for noise at the least and most exposed façade and total and non-traffic air pollution.^17,26 Even so, modelling uncertainty and lack of information about non-residential exposures, e.g. at work or when commuting, make some exposure misclassification inevitable. Non-differential exposure misclassification and resulting classical or Berkson error may therefore have affected our risk estimates towards the null and decreased precision. Some exposure uncertainty may also relate to the exact timing of exposure. Still, we found very similar HRs for both shorter and longer exposure periods indicating that a 5-year period captures relevant exposures well. Additionally, competing risk of death may have affected our HRs as air pollution is associated with increased mortality.

Multi-pollutant analyses present several challenges. If pollutants are not estimated to equal standards, the most precisely modelled factor may dominate in the multi-pollutant model. Further, if included factors are highly correlated, results may become difficult to interpret due to statistical instability. To address these caveats, we preselected exposures from the three exposure domains: air pollution, noise, and green space using two, three, and four pollutant models. Only exposures consistently associated with MI in these models were included in the final multi-exposure model. Still, it cannot be ruled out that correlations between exposures influence the results of our multi-exposure analyses.

Another limitation of our study was the lack of information about body mass index (BMI), physical activity, smoking, and other lifestyle factors. In a random sample of the Danish population, using the same covariates as in the present study, we have, however, found associations of MI with air pollution virtually unchanged after additional adjustment for BMI, physical activity, smoking, and diet.⁴⁰ This indicates that adjustment for our battery of potential confounders leaves little room for residual confounding from lifestyle when considering air pollution. This has, however, not been tested for noise and green space and can, therefore, not rule out potential residual confounding concerning noise and green space. Medical conditions such as diabetes and hypertension may be on the pathway linking environmental exposures and MI. Therefore, adjusting for such factors could obscure the relationship. A detailed analysis of how these factors may mediate or modify the effect is outside the scope of this study.

Our study cohort was a predominantly Caucasian, older Western population in a temperate climate. These factors and potential differences in air pollution levels and composition should be heeded when generalizing results to other populations.

Conclusions

Lack of green space close to the home and road traffic noise and air pollution were independently associated with the risk of MI in this nationwide cohort study. PM_2.5 appeared to be the air pollutant most consistently associated with the harmful aspects of air pollution.

In multi-pollutant models, risk estimates for air pollution, noise, and lack of green space were similar, indicating that all may be equally relevant targets for regulatory measures.

Author contributions

A.H.P.: conceptualization, writing—original draft, and formal analysis. M.S. and U.A.H.: conceptualization and writing—review & editing. M.K., J.H.C., L.M.F., A.M., J.K., T.M., and J.B.: resources and writing—review & editing. O.R.: conceptualization, writing—review & editing, and funding acquisition. All authors read, gave final approval, and agreed to be accountable for all aspects of the work, ensuring integrity.

Supplementary material

Supplementary material is available at European Journal of Preventive Cardiology.

Funding

Research described in this article was conducted under contract to the Health Effects Institute (HEI), an organization jointly funded by the United States Environmental Protection Agency (EPA) (Assistance Award No. R-82811201) and certain motor vehicle and engine manufacturers. The contents of this article do not necessarily reflect the views and policies of the EPA or the vehicle and engine manufacturers. The study funder was not involved in the design of the study; the collection, analysis, and interpretation of the data; and writing the paper and did not impose any restrictions regarding the publication of the paper.

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 license for the current study and so are not publicly available. Access to data requires permission from Statistics Denmark and the Danish Cancer Society.

Author notes

Conflict of interest: none declared.