Traffic‐Related Air Pollution and Carotid Plaque Burden in a Canadian City With Low‐Level Ambient Pollution

Abstract

Background

The association between fine particulate matter and cardiovascular disease has been convincingly demonstrated. The role of traffic‐related air pollutants is less clear. To better understand the role of traffic‐related air pollutants in cardiovascular disease development, we examined associations between NO₂, carotid atherosclerotic plaque, and cardiometabolic disorders associated with cardiovascular disease.

Methods and Results

Cross‐sectional analyses were conducted among 2227 patients (62.9±13.8 years; 49.5% women) from the Stroke Prevention and Atherosclerosis Research Centre (SPARC) in London, Ontario, Canada. Total carotid plaque area measured by ultrasound, cardiometabolic disorders, and residential locations were provided by SPARC medical records. Long‐term outdoor residential NO₂ concentrations were generated by a land use regression model. Associations between NO₂, total carotid plaque area, and cardiometabolic disorders were examined using multiple regression models adjusted for age, sex, smoking, and socioeconomic status. Mean NO₂ was 5.4±1.6 ppb in London, Ontario. NO₂ was associated with a significant increase in plaque (3.4 mm² total carotid plaque area per 1 ppb NO₂), exhibiting a linear dose‐response. NO₂ was also positively associated with triglycerides, total cholesterol, and the ratio of low‐ to high‐density lipoprotein cholesterol (P<0.05). Diabetes mellitus mediated the relationship between NO₂ and total carotid plaque area (P<0.05).

Conclusions

Our results demonstrate that even low levels of traffic‐related air pollutants are linked to atherosclerotic plaque burden, an association that may be partially attributable to pollution‐induced diabetes mellitus. Our findings suggest that reducing ambient concentrations in cities with NO₂ below current standards would result in additional health benefits. Given the billions of people exposed to traffic emissions, our study supports the global public health significance of reducing air pollution.

Clinical Perspective

What Is New?

• We found that even low levels of traffic‐related air pollution were significantly associated with higher carotid plaque burden, triglycerides, and total cholesterol, as well as a higher ratio of low‐ to high‐density lipoprotein cholesterol.

• The relationship between traffic‐related air pollution and carotid plaque burden appeared to be mediated by air pollution–induced diabetes mellitus.

• Our results suggest that air pollution–induced atherosclerotic plaque formation may play a role in cardiovascular disease susceptibility, particularly among diabetics; and that these processes occur even in cities with relatively low exposures.

What Are the Clinical Implications?

• We report that a 10 ppb increase in NO₂—the range observed in the low‐concentration city of London, Ontario, Canada—contributes to a 33.6 mm² increase in total plaque area. Plaque burden of this magnitude has previously been associated with a 5% increase in 5‐year risk of stroke, myocardial infarction, and vascular mortality.

• The association between NO₂ and plaque in this study is clinically significant because exposure to traffic‐related air pollution is ubiquitous; and because previous research suggests that even modest elevation or increase in total carotid plaque area is clinically relevant.

• Although these findings have their most obvious application in public health and regulatory policy, there may be a place for advising patients to follow local air quality indices and take steps to minimize their exposure to traffic‐related air pollution, particularly if they are diabetic or have existing cardiovascular disease.

Nonstandard Abbreviations and Acronyms.

CAC coronary artery calcification

IMT intima media thickness

LUR land use regression

SES socioeconomic status

SPARC Stroke Prevention and Atherosclerosis Research Centre

TRAP traffic‐related air pollutants

TPA total carotid plaque area

Introduction

The role of air pollution in exacerbating and triggering myocardial events, including mortality, has been convincingly demonstrated.1 There is compelling evidence that atherosclerosis—which forms the basis for much of cardiovascular disease (CVD) pathology—is driven by inflammation, as well as endothelial and metabolic dysfunction,1, 2, 3 and that air pollution contributes to cardiometabolic disease by promoting systemic inflammation.4, 5 However, the role of air pollution, particularly traffic‐related air pollutants (TRAP) such as NO₂ in the development of atherosclerosis, is still poorly understood. This study examines associations between long‐term exposure to NO₂, atherosclerotic plaque burden, and cardiometabolic disorders that may contribute to CVD development.

A growing number of studies have examined associations between air pollution, subclinical atherosclerosis, and vascular disease with varying results. Fine particulate matter (PM_2.5) has been positively associated with carotid intima media thickness (IMT) and coronary artery calcification (CAC).3, 6 Other studies reported no association between IMT, CAC, ankle brachial index, and PM_2.5.7, 8, 9, 10 NO₂ was associated with ankle brachial index but not IMT in 1 study,11 and left IMT in another.12 Nitrogen oxides (NOx) were also associated with CAC—but not IMT—progression.13

Previous studies examining air pollution and atherosclerosis had several limitations that reduce their interpretability. Most relied on IMT, which is biologically and genetically distinct from atherosclerosis14, 15, 16; CAC was also used in some studies. CAC is highly correlated with plaque burden,17 and both CAC and plaque burden are strong predictors of future CVD risk.15 However, CAC has several notable limitations. It is an indirect marker of atherosclerosis mediated by different biological processes,18, 19, 20, 21, 22, 23 with a much slower response to therapy than plaque burden measured by ultrasound,24 and CAC scans increase patient exposure to radiation.24 While MESA‐Air was a landmark prospective evaluation that demonstrated important associations between air pollution, CAC, and IMT, the study did not include any direct metric of atherosclerotic plaque. Some studies were also limited by using air pollution data from sparse regulatory monitoring networks rather than spatially refined data.

We examined associations between outdoor residential NO₂, atherosclerotic plaque burden, and cardiometabolic disorders in London, Ontario, Canada—a city with relatively low TRAP—using accurate, high‐resolution methods to assess atherosclerosis and air pollution. Atherosclerotic plaque was quantified using carotid total plaque area (TPA) measured by 2‐dimensional carotid ultrasound, a stronger predictor of CVD risk compared with IMT.25, 26, 27, 28 Unlike CAC, TPA is a direct metric of atherosclerosis and is more mechanistically and biologically related to the pathobiological process driving CVD events.16, 19 Only 1 previous study used TPA based on carotid ultrasound to examine air pollution impacts, and they saw no significant associations between TRAP and carotid artery atherosclerosis.29

Outdoor residential NO₂ was estimated by land use regression (LUR) modeling, which provides more accurate fine‐scale intraurban concentrations compared with proximity models, regulatory monitoring, and satellite‐derived estimates30, 31; and has been shown to have comparable, and in some instances better, performance compared with dispersion models.30 Cardiometabolic disorders included diabetes mellitus, hypertension, blood pressure (BP), total cholesterol (TC), high‐density lipoprotein cholesterol (HDL‐C), low‐density lipoprotein cholesterol (LDL‐C), LDL:HDL ratio, triglycerides, and body mass index (BMI).

Methods

Study Population

The study population comprised patients at the Stroke Prevention and Atherosclerosis Research Centre (SPARC) in London, Ontario, Canada, a small university city (population 385 000). Patients are referred to SPARC following a stroke or transient ischemic attack, or for asymptomatic carotid stenosis, early onset or severe vascular disease, or family history of vascular disease.

Approval to conduct the research described in this article was obtained from the Research Ethics Boards of Health Canada and the University of Western Ontario (protocol number 107051). The requirement to obtain informed consent to analyze the database was waived. Air pollution and neighborhood socioeconomic status (SES) data used in this study are available through The Canadian Urban Environmental Health Research Consortium repository at canue.ca or by request to the corresponding author. To comply with ethics and privacy requirements, clinical data used in this study will not be made available without additional Research Ethics Boards approval. Requests for clinical data from qualified researchers may be submitted to Dr David Spence at University of Western Ontario (dspence@robarts.ca).

Carotid Plaque Measurements

TPA was defined as the sum of cross‐sectional areas of all plaques between the clavicle and the angle of the jaw. TPA measurement has been described in detail previously.32 Briefly, plaque area was measured using high‐resolution duplex ultrasound scanners (Phillips ATL Mark 9, ATL 5000 HDI, and Philips IU‐22, Advanced Technology Laboratories). Plaque was defined as a local thickening of the intima >1 mm.

Measurements were performed in magnified longitudinal views of each plaque in the right and left common, internal, and external carotid arteries. The plane of measurement for each plaque was chosen by finding the plane that showed the largest extent of the individual plaque, and then freezing and magnifying the view. Plaque was measured by tracing the perimeter of the magnified plaque image with an onscreen cursor and recording the cross‐sectional plaque area calculated by the microprocessor in the scanner. This process was repeated until all visible plaques were measured.

Intraclass correlation for repeated measurements by the same technician was 0.94.32 For linear regression models, a cube root transformation was used to normalize the distribution of TPA, as previously reported.17, 33 Baseline TPA measurements included in the analyses were collected from 1990 to 2013.

Other Clinical Measurements and Health Outcomes

All clinical data and health outcomes were obtained from patient medical records. Diabetes mellitus and hypertension were based on a history of physician diagnosis. Clinical measurements were collected by SPARC. Clinical measurements included blood pressure (systolic and diastolic), TC, triglycerides, HDL‐C, LDL‐C, TC:HDL ratio, LDL:HDL ratio, and BMI. We also examined elevated risk categories defined as follows: high cholesterol (TC >240 mg/dL), high LDL (>160 mg/dL), low HDL (<40 mg/dL), high triglycerides (>200 mg/dL), high TC:HDL ratio (men: >4.5, women: >4.0), high LDL:HDL (men >3.6, women: >3.2), high blood pressure (systolic BP ≥140 and diastolic BP ≥90), overweight or obese (BMI ≥25), and obese (BMI≥ 30). TPA and clinical measurements were collected during the same examination visit.

Traffic‐Related Air Pollution

The methods used to generate outdoor residential NO₂ concentrations for this study have been described previously.34 Briefly, Ogawa passive samplers were used to measure 2‐week integrated ambient NO₂ concentrations at 50 locations throughout the city of London, Ontario, Canada in spring 2010. A land‐use regression model was developed to estimate long‐term NO₂ concentrations by regressing spatially varying land use characteristics against NO₂ at the 50 sampling sites. Model predictors included traffic density within 150 meters (m), dwelling density within 1000 m, distance to the nearest highway, industrial land use within 1600 m, and length of railways within 550 m. The LUR model explained 78% of the spatial variability in measured NO₂.

Exposure surfaces for intraurban variations in NO₂ based on both air monitoring and LUR models have been shown to be stable over time, suggesting that they are representative of long‐term gradients of exposure in urban populations.35, 36, 37, 38 Estimates of NO₂ from the LUR model were assigned to each patient based on their residential location, indicated by the 6‐character postal code reported during the clinical visit in which TPA was measured. In urban areas, Canadian 6‐character postal codes are highly local, typically representing an area smaller than a city block.

Covariates

Covariates included age, sex, smoking status (ever smoker, current smoker, past smoker, never smoker, and pack‐years) and SES. Age, sex, and smoking were obtained from patient medical records. SES was assessed using neighborhood measures from the Canadian census, including average income and percent of population with a university degree or diploma in the census dissemination area where each patient resided. A Canadian dissemination area is the smallest geographic unit for which all census data are available, with a population of ≈400 to 700 people.39 Dissemination area–level SES variables have been used effectively to address residual confounding associated with SES in previous analyses linking air pollution and health outcomes (eg, mortality, CVD, and lung cancer) in large population‐based cohorts.40, 41 SES variables were linked to patients based on their residential postal code.

Inclusion Criteria

Analyses were conducted on patients living within the city of London for whom LUR NO₂ were available. Analyses were limited to patients with at least 2 clinical visits to ensure inclusion of patients with detailed medical history, as well as complete residential location and demographic information including age, sex, smoking history, and SES.

Statistical Analysis

Multiple linear regression models were used to estimate associations between NO₂, TPA, and other continuous cardiometabolic outcomes. Logistic regression models were specified for binary outcomes. Covariates were selected based on their relevance and strength of association with the outcome of interest, and optimization of multiple linear regression models based on the Akaike Information Criterion. All models were adjusted for age, sex, smoking, and SES.

Mediation analyses were conducted using methods developed by Dudley et al42 and Jasti et al.43 Based on the large sample size, and the similarity in sample size between models used to assess mediation, we used the Sobel test to assess the significance of potential mediation.44 Effect modification was also considered. Further details are provided in Data S1 and Figure S1.

Data included in these analyses were collected between 1990 and 2013, with 84% of the examinations carried out between 2000 and 2013 (n=1878). We conducted multiple sensitivity analyses to test whether our results were sensitive to the date of the clinical examinations. We considered models that were fully or partially stratified by examination date. LUR models have been consistently shown to provide stable long‐term estimates of NO₂ over a 10‐year period35, 36, 37, 38; therefore, we stratified by 10 years from the date of NO₂ data collection in 2010, resulting in 2 cohorts: patients with examination dates from 1990 to 1999 (>10 years from NO₂ data collection, n=349), and patients with examination dates from 2000 to 2013 (within 10 years of NO₂ data collection, n=1878). In fully stratified models, we calculated separate estimates for all regression parameters in the 1990–1999 versus 2000–2013 patients. In partially stratified models, we reported separate estimates for all parameters that differed between the 1990–1999 and 2000–2013 groups, with combined estimates for parameters that did not vary between the 2 patient groups.

To further assess the potential impacts of examination date, we specified mixed effect linear regression models for TPA with time period of clinical examination date included as a random effect. Mixed models included a 4‐level random effect variable (grouped by quartiles) based on the clinical examination date. We also considered the impact of patient age, plaque transformation, and influential observations on reported model results. Sensitivity analyses and effect modifiers are further described in Data S1.

Statistical analyses were performed using SAS 9.3. Dose–response analysis was conducted using a natural cubic spline function in R 3.1.

Results

Descriptive Statistics

Descriptive statistics are provided in Table 1. Patients represented an older population, with a mean age of 62.9 (Min: 18.0, Max: 96.0) years. Patients evenly comprised women (49.5%) and men (50.5%). There was a high prevalence of smoking in the patient population; more than half reported either current (17%) or past (42%) smoking. Among ever smokers—comprising both past and current smokers—pack years ranged from 0 to 165, with a mean of 13.5 and median of 5 (Min: 0, Max: 165) pack years. Most patients lived in neighborhoods with a lower proportion of university graduates (Mean: 27%, Min: <1%, Max: 100%) and lower mean neighborhood income compared with national averages.45 TPA ranged from 0 to 156 mm², with a mean of 109 mm², and median of 70.0 (Min: 0, Max: 873) mm².

Table 1.

Descriptive Statistics for Sociodemographic Characteristics, Smoking, Air Pollution, Clinical Measurements, and Health Outcomes

	Sample Size (n)	Mean	SD	Median	Min	Max
Age, y	2227	62.9	13.8	64.0	18.0	96.0
Income (average income by DA)a	2227	$34 215	$13 160	$30 062	$13 842	$155 960
Education (% University Degree by DA)a	2227	27.0%	17.5%	23.7%	<1%	100%
Pack y	2227	13.5	18.6	5.0	0	165
Nitrogen dioxide (NO2)	2227	5.4	1.6	5.1	3.0	13.0

	Sample Size (n)	Frequency (%)
Female	2227	1102 (49.5)
Ever smoker	2227	1306 (58.6)
Current smoker	2227	375 (16.8)
Past smoker	2227	931 (41.8)
Never smoker	2227	921 (41.4)

	Sample Size (n)	Mean	SD	Median	Min	Max
Plaque area, mm2 b	2227	109	122	70.0	0	873
Systolic BP, mm Hg	2222	144	21.4	142	92.0	240
Diastolic BP, mm Hg	2221	82.3	12.8	82.0	43.0	140
Total cholesterol, mg/dL	1953	195	47.0	192	81.2	583
Triglycerides, mg/dL	1943	160	104	133	6.2	1594
HDL cholesterol, mg/dL	1935	51.9	16.9	49.5	3.9	192
TC:HDL ratio	1935	4.1	2.0	3.8	1.3	52.7
Triglyceride:HDL ratio	1930	3.7	4.0	2.8	0.1	71.1
LDL cholesterol, mg/dL	1898	112	41.3	108	10.0	387
LDL:HDL ratio	1896	2.4	1.2	2.1	0.2	14.9
BMI, kg/m2	1875	27.4	5.0	26.7	14.7	51.3

	Sample Size (n)	Frequency (%)
Diabetes mellitus	2221	328 (14.8)
High blood pressurec	2221	485 (21.8)
Hypertension	2059	1420 (69.0)
High cholesterol, >240 mg/dL	1953	324 (16.6)
Medium/high cholesterol, >200 mg/dL	1953	847 (43.4)
High triglycerides, >200 mg/dL	1943	463 (23.8)
High TC:HDL ratiod	1935	713 (36.8)
Low HDL, <40 mg/dL	1935	469 (24.2)
High LDL, >160 mg/dL	1898	240 (12.6)
High LDL:HDL ratioe	1896	305 (16.1)
Overweight or obese, BMI ≥25	1875	1240 (66.1)
Obese, BMI ≥30	1875	482 (25.7)

BMI indicates body mass index; BP, blood pressure; DA, dissemination area; HDL, high‐density lipoprotein; LDL, low‐density lipoprotein; Max, maximum; Min, minimum; and TC, total cholesterol.

^aA dissemination area (DA) is a “small, relatively stable geographic unit composed of one or more adjacent dissemination blocks…with a population of 400 to 700 people, It is the smallest standard geographic area for which all census data are disseminated” (Statistics Canada).

^bUntransformed plaque area (mm²).

^cHigh blood pressure was defined as both systolic ≥140 mm Hg and diastolic ≥90 mm Hg.

^dHigh TC:HDL ratio was defined as ≥4.5 for men and ≥4.0 for women.

^eHigh LDL:HDL ratio was defined as ≥3.6 for men and ≥3.2 for women.

The distribution of NO₂ in the city is shown in Figure 1. Outdoor residential NO₂ concentrations at patient homes ranged from 3.0 to 13.0 parts per billion (ppb), with a mean of 5.4 ppb at patient postal codes. NO₂ exposure in this patient population was consistent with the distribution of outdoor residential NO₂ in a population representative cohort; ie, among the 381  522 London residents of the Canadian Census Health and Environment Cohort (CanCHEC), outdoor residential NO₂ estimated using the same LUR model ranged from 2.6 to 18.9 ppb, with a mean of 4.7.41 Ambient pollutant concentrations in Canadian cities are typically much lower than in other cities internationally, falling well below World Health Organization's guideline of 21.2 ppb for mean annual NO₂.46 For example, annual average NO₂ ranged from 15 to 19 ppb in Toronto, Edmonton, and Calgary in 2010, while Hong Kong, China had an annual average of almost 40 ppb.46 London, Ontario displayed the lowest mean NO₂ concentrations among 10 Canadian cities with LUR models.41

Figure 1.

Distribution of NO₂ in the city of London, Ontario.

Higher NO₂ concentrations are localized mainly in the urban core, with much lower concentrations in the suburbs. NO₂ concentrations were largely driven by traffic indicators, which explained ≈70% of the variability in NO₂ captured by the model. Dwelling density (17%), industrial land use (10%), and railroad lines (4%) explained the remainder of the variation in NO₂ concentrations. LUR indicates land use regression; and NO₂, nitrogen dioxide. The map of NO₂ in London is reprinted by permission of Taylor & Francis, Ltd, http://www.tandf​online.com on behalf of Air & Waste Management Association, from a figure in Oiamo et al.,34 copyright ©2012 Air & Waste Management Association, www.awma.org.

Descriptive statistics for smoking, obesity, diabetes mellitus, and hypertension stratified by age and sex among study patients compared with the general population are provided in Data S2 and Tables S1 through S4. The prevalence of current smoking by age and sex among patients in the study was similar to the general population. However, the prevalence of obesity, diabetes mellitus, and hypertension was higher among patients. For obesity and diabetes mellitus, this was driven by elevated prevalence among younger patients, while the prevalence of hypertension was higher among patients in all age and sex groups. See Data S2 for further discussion.

Associations Among NO₂, TPA, and Other Clinical Measurements

Linear regression results for NO₂ are reported in Table 2. The NO₂‐TPA dose–response curve is provided in Figure 2. Mean NO₂ by TPA quartile is provided in Data S3 and Figure S2. NO₂ was significantly associated with TPA adjusting for age, sex, smoking, and SES, with a 1.2 mm² increase in cube root transformed TPA per 1 ppb increase in NO₂. For untransformed TPA, a 1 ppb increase in NO₂ was associated with a 3.4 mm² increase in TPA, or 33.6 mm² increase in TPA per 10 ppb (P<0.05). NO₂ was also positively associated with triglycerides, TC, and LDL:HDL ratio. NO₂ was marginally associated with TC:HDL ratio and triglyceride:HDL ratio. NO₂ was not associated with systolic or diastolic BP, LDL‐C, HDL‐C, or BMI.

Table 2.

Multiple Linear Regression Analysis of NO₂, Plaque Area, and Other Clinical Measurements

	Sample Size (n)	Regression Coefficient and 95% CI for NO2 (per 1 ppb)
		Unadjusted Models	Adjusted Models
Plaque area, mm2	2227	2.55 [1.42, 3.69]a	1.22 [0.29, 2.15]b
Systolic BP, mm Hg	2222	0.46 [−0.11, 1.02]	0.24 [−0.33, 0.82]
Diastolic BP, mm Hg	2221	−0.09 [−0.43, 0.25]	0.12 [−0.23, 0.48]
Total cholesterol, mg/dL	1953	1.35 [−0.02, 2.71]c	1.73 [0.36, 3.11]b
LDL‐C, mg/dL	1898	0.43 [−0.79, 1.65]	0.68 [−0.57, 1.93]
HDL‐C, mg/dL	1935	−0.44 [−0.93, 0.05]c	−0.24 [−0.73, 0.24]
TC:HDL ratio	1935	0.06 [0.00, 0.12]b	0.05 [−0.01, 0.11]c
LDL:HDL ratio	1896	0.04 [0.01, 0.08]b	0.04 [0.01, 0.08]b
Triglycerides, mg/dL	1943	5.36 [2.35, 8.37]a	4.61 [1.38, 7.84]a
Triglyceride:HDL ratio	1930	0.13 [0.01, 0.24]b	0.11 [−0.01, 0.23]c
BMI, kg/m2	1875	0.17 [0.03, 0.32]b	0.09 [−0.06, 0.24]

Models linking NO₂ concentrations (per 1 ppb) with continuous clinical measurements and health outcomes were adjusted for age, sex, smoking, and SES (ie, Plaque Area=β₀+β_NO2+β_Age+β_Sex+β_Smoking+β_SES). Plaque area was modeled as cube root transformed TPA. BMI indicates body mass index; BP, blood pressure; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; NO₂, nitrogen dioxide; ppb, parts per billion; SES, socioeconomic status; TC, total cholesterol; and TPA, total plaque area.

^a P<0.01.

^b P<0.05.

^c P<0.10.

Figure 2.

Dose–response curve for NO₂ and TPA.

Local NO₂ concentrations were significantly associated with TPA after adjustment for age, sex, smoking, and socioeconomic status. There was a 1.2 mm² increase in cube root transformed TPA, per 1 ppb increase in NO₂. ppb, parts per billion; and TPA, total carotid plaque area.

Logistic regression model results for NO₂ are reported in Table 3. NO₂ was significantly associated with increased odds of exceeding clinically relevant thresholds for triglycerides, TC, TC:HDL ratio, and LDL:HDL ratio, as well as with obesity in women. NO₂ was marginally associated with diabetes mellitus and obesity (P<0.10), but was not associated with hypertension, high blood pressure, high LDL‐C, low HDL‐C, or obesity in men.

Table 3.

Multiple Logistic Regression Analysis of NO₂, Clinical Measurements, and Health Outcomes

	Sample Size (n)	Odds Ratio and 95% Confidence Interval for NO2 (per 1 ppb)
		Unadjusted Models	Adjusted Models
Diabetes mellitus	2221	1.13 [1.06, 1.21]a	1.07 [0.99, 1.15]b
High BP	2221	0.96 [0.89, 1.02]	0.97 [0.90, 1.04]
Hypertension	2059	1.06 [0.99, 1.12]b	1.02 [0.95, 1.09]
High TC	1953	1.08 [1.00, 1.16]c	1.09 [1.01, 1.18]c
High TC:HDL ratio	1935	1.14 [1.07, 1.22]a	1.08 [1.01, 1.15]c
High LDL:HDL ratio	1896	1.09 [1.02, 1.15]a	1.11 [1.02, 1.21]c
High LDL‐C	1898	1.04 [0.98, 1.11]	1.07 [0.98, 1.17]
Low HDL‐C	1935	1.08 [1.00, 1.18]b	1.04 [0.96, 1.12]
High triglycerides	1943	1.13 [1.05, 1.22]a	1.12 [1.04, 1.20]a
Overweight or obese	1875	1.10 [1.03, 1.17]a	1.02 [0.96, 1.09]
Obese, all	1875	1.05 [0.98, 1.11]	1.06 [1.00, 1.14]b
Obese, women	949	1.14 [1.05, 1.25]a	1.13 [1.03, 1.24]c
Obese, men	926	1.06 [0.97, 1.16]	1.00 [0.91, 1.10]

Models linking NO₂ concentrations (per 1 ppb) with binary clinical measurements and health outcomes were adjusted for age, sex, smoking, and SES (ie, Diabetes=β₀+β_NO2+β_Age+β_Sex+β_Smoking+β_SES). BMI indicates body mass index; BP, blood pressure; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; ppb, parts per billion; SES, socioeconomic status; and TC, total cholesterol.

^a P<0.01.

^b P<0.10.

^c P<0.05.

Mediation

Further analyses were conducted to examine potential mediation of the relationship between NO₂ and TPA by cardiometabolic disorders (Table 4). Diabetes mellitus significantly mediated the relationship between NO₂ and TPA, while elevated TC:HDL was a marginally significant mediator. There were no other statistically significant mediators, and no evidence to suggest that diabetes mellitus and other cardiometabolic disorders significantly modified the association between NO₂ and plaque.

Table 4.

Mediating Effects of Cardiometabolic Disorders on the Association Between NO₂ and Plaque

	N	Percent of the Total Effect That Is Mediated	Ratio of the Indirect to the Direct Effect	Sobel P Value
Continuous
Systolic BP, mm Hg	2222	7.12%	0.08	0.32
Diastolic BP, mm Hg	2221	0.36%	<0.01	0.75
Total cholesterol, mg/dL	1953	≈(0.83%)	−0.01	0.77
Triglycerides, mg/dL	1943	6.07%	0.06	0.12
HDL‐C, mg/dL	1935	3.05%	0.03	0.37
TC:HDL ratio	1935	2.27%	0.02	0.33
Trig:HDL ratio	1930	1.64%	0.02	0.45
LDL:HDL ratio	1896	3.89%	0.04	0.28
LDL‐C, mg/dL	1898	0.31%	<0.01	0.83
BMI, kg/m2	1875	≈(0.83%)	−0.01	0.56
Binary
Diabetes mellitus	2221	9.25%	0.41	0.03
High blood pressure	2221	≈(2.75%)	−0.09	0.29
Hypertension	2059	6.07%	0.19	0.49
High cholesterol	1953	1.42%	0.06	0.57
High Trig	1943	2.22%	0.06	0.55
Low HDL‐C	1935	3.63%	0.11	0.27
TC:HDL level	1935	6.30%	0.16	0.10
High LDL‐C	1898	4.46%	0.22	0.21
LDL:HDL level	1896	6.65%	0.27	0.16
Obese	1875	≈(1.38%)	−0.04	0.52
Overweight or obese	1875	≈(0.10%)	<0.01	0.83

Models were adjusted for age, sex, smoking, and SES. BMI indicates body mass index; BP, blood pressure; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; SES, socioeconomic status; TC, total cholesterol; and Trig, triglycerides.

Sensitivity Analyses

We conducted sensitivity analyses to examine associations between NO₂ and plaque among patients whose ultrasound examination was conducted within 10 years of NO₂ data collection (ie, patients with examination dates from 2000 to 2013) compared with patients with examination dates from 1990 to 1999. Patients with clinical examinations from 1990 to 1999 were younger, and had lower plaque burden compared with patients from 2000 to 2013 (P<0.05). In stratified and partially stratified models, the association between NO₂ and TPA was slightly stronger among patients assessed from 1990 to 1999 (2.2 mm²/1 ppb) compared with patients from 2000 to 2013 (1.0 mm²/1 ppb). The relationship between SES and TPA also varied between patient cohorts, with a strong inverse association among patients from 2000 to 2013, and nonsignificant positive association among patients from 1990 to 1999. Associations between age, sex, smoking, and TPA were similar in both patient cohorts.

Limiting the analyses to older patients did not change the results for the full cohort (see Data S1 for details). However, in analyses stratified by examination date (1990–1999 versus 2000–2013), we saw stronger associations between NO₂ and TPA among older patients (40 years or older) in both patient groups (1990–1999: 2.4 mm²/1 ppb, P<0.05; 2000–2013: 1.1 mm²/1 ppb P<0.05).

Mixed model results testing examination date were comparable to linear regression models for TPA (1.1 mm²/1 ppb P<0.05); therefore, linear regression models were reported in the article. Sensitivity analyses for other clinical health outcomes are described in Data S1.

Discussion

We found that exposure to low levels of NO₂ was significantly associated with total carotid plaque area–an important measure of systemic atherosclerosis–and that heightened atherogenesis induced by TRAP may be at least partially mediated by air pollution–induced diabetes mellitus. TRAP was also related to other cardiometabolic abnormalities in this study.

Plaque

Exposure to NO₂ in London, Ontario, a small city with relatively low ambient concentrations, was positively associated with cumulative plaque burden in a high‐risk patient population. Previous studies have reported inconsistent results in relating air pollution with atherosclerosis, with interpretation of these findings hampered by methodological limitations.

Most studies to date relied on measurement of IMT, which is biologically and genetically distinct from atherosclerosis.16, 25 MESA‐Air investigators cited several major limitations in using IMT to examine air pollution and atherosclerosis in observational studies, including the weakness of IMT as a marker for atherosclerosis,13 and the sensitivity of the method to ultrasound device quality.47 Some meta‐analyses reported statistically or marginally significant associations between PM_2.5 and IMT.10, 48 However, both MESA‐Air 13 and a meta‐analysis of 4 cohorts in the ESCAPE study using high‐resolution (LUR) estimates of PM_2.5 and NO₂ 9 found no statistically significant associations between air pollution and IMT. The current study, which used a direct measure of atherosclerotic plaque burden and spatially resolved pollutant concentrations, provides additional evidence for a link between TRAP and atherosclerosis.

Although total carotid plaque area is a stronger indicator of atherosclerosis 16 and CVD risk 26, 27, 28, 49, 50 compared with IMT, only 1 previous study, the Multicultural Community Health Assessment Trial (M‐CHAT) in Vancouver, BC, examined associations between air pollution and TPA.29 Like the current study, M‐CHAT used high‐resolution LUR models for TRAP coupled with ultrasound measures of TPA. However, they found no associations between TRAP and plaque burden. Ambient NO₂ levels among M‐CHAT participants were slightly higher than in the London study population, with similar spatial variation. Compared with the current study, the M‐CHAT population was much younger—with a median age of 46 years in M‐CHAT versus 64 in London patients, and the sample size was much smaller (N=509). These limitations likely explain their null findings.

CVD is a multifactorial disease impacted by environmental, behavioral, and genetic risk factors. The association between NO₂ and plaque in this study is clinically significant because exposure to traffic pollution is ubiquitous; and because previous research suggests that even modest elevation or increase in TPA is clinically relevant.32, 51 For example, Spence et al.32 reported that elevation of plaque burden was a strong predictor of the 5‐year risk of stroke, myocardial infarction, and vascular mortality; after adjustment for a broad panel of risk factors, the 5‐year risk of those events by TPA quartile was 5.6%, 10.7%, 13.9%, and 19.5%. Our results suggest that exposure to TRAP was associated with an elevation of ≈3.4 mm² per ppb of NO₂—a difference of ≈33.6 mm² between patients with the highest and lowest levels of exposure—which suggests a possible increase in 5‐year risk of 5% for those patients. Thus, our results may be regarded as clinically important as well as statistically significant.

Diabetes Mellitus

There was a marginally significant positive association between NO₂ and diabetes mellitus in this study, which is consistent with emerging evidence that PM_2.5, NO₂, and other ambient pollutants may be associated with diabetes mellitus risk. Long‐term exposure to NO₂, NOx, PM_2.5, and PM₁₀ has been positively associated with incident52, 53, 54, 55 and prevalent53 diabetes mellitus, as well as diabetes mellitus–related hospitalizations and mortality52, 53, 54, 55, 56 and insulin resistance.53, 57 Recent reviews53, 58 highlight the importance of TRAP such as NO₂ in the development of diabetes mellitus.

Our results provide evidence that diabetes mellitus significantly mediated the relationship between NO₂ and plaque, suggesting that some of the observed association between NO₂ and plaque may be attributable to pathophysiology of diabetes mellitus. This finding is consistent with the increased risk of air pollution–mediated CVD outcomes among diabetics, and the hypothesized role of diabetes mellitus in atherosclerosis. Diabetes mellitus has been associated with increased risk of air pollution–mediated CVD morbidity and mortality,54, 57 as well as atherosclerosis and plaque development.59, 60, 61 Metabolic syndrome severity was also associated with carotid plaque area in a small adult cohort,62 providing additional evidence for a linkage between metabolic disease and plaque burden.

Several potential mechanisms have been proposed to explain susceptibility to air pollution–mediated–CVD outcomes among diabetics including increased inflammation, vascular reactivity, and endothelial dysfunction.4, 57 Similar mechanisms have been proposed to explain the increased severity of CVD and atherosclerosis observed among diabetics.63, 64, 65, 66 Renin–angiotensin–aldosterone‐system activation,60 overexpression of microRNA,67 and altered mineral metabolism66 may also contribute to atherosclerosis among diabetics.

Other Cardiometabolic Disorders

NO₂ was significantly associated with total cholesterol, elevated cholesterol, elevated TC:HDL ratio, total triglycerides, and elevated triglycerides, suggesting that air pollution–mediated atherosclerosis may be related to metabolic changes including dyslipidemia and hypertriglyceridemia. These results are consistent with the limited evidence available from previous studies, which reported associations between air pollutants (eg, PM_2.5, NO₂, and ozone), triglycerides, fasting glucose, apolipoprotein B, hemoglobin A1c, and elevated TC and LDL‐C, as well as reduced HDL‐C.68, 69, 70, 71 Our results provide further evidence that air pollution contributes to cardiometabolic disorders.

NO₂ was not significantly associated with obesity, BMI, BP, or hypertension in this patient population. Previous literature reviews found clear evidence for a causal association between ambient PM_2.5 and increased arterial BP,72 as well as growing evidence for an association between PM_2.5 and hypertension.6, 73, 74 However, evidence for NO₂ is limited.73 A meta‐analysis of 15 population‐based cohorts in Europe (N=113 926) also reported no association between air pollutants (including PM_2.5 and NO₂) and hypertension or BP.75 Mixed results in the literature may be attributable to differences in study design, BP collection methods (eg, standardized day, time, and activity before measurement), pollutant of interest, and statistical analysis methods.

Mechanistic studies in animals suggest that the inflammatory effects of air pollution could promote obesity,53, 76 and several longitudinal studies linked TRAP with the development of obesity in children.53, 76 However, there is less convincing evidence linking air pollution and obesity in adults.76 Associations between air pollution, BMI, and obesity in adults, particularly from longitudinal studies, are limited. Our results suggest that further work is needed to characterize associations between air pollution and obesity in adults.

Finally, there was a high prevalence of hypertension and obesity in the SPARC patients. Associations among NO₂, hypertension, high blood pressure, BMI, and obesity may differ in healthy populations.

Limitations

The results of the current study may be limited by several factors. Lack of residential history is a common limitation in air pollution epidemiology. In this study, NO₂ concentrations were assigned to patients based on their residential address at the time of TPA measurement. While residential mobility is typically lower among noninstitutionalized older adults,77 who comprise the majority of the study population, the potential impact of past exposures, particularly among patients who moved, is unknown.

LUR models have been widely used to estimate long‐term exposure in epidemiological studies because they accurately characterize the long‐term intraurban gradients required to support epidemiological analyses.30, 31 LUR modeling does not account for individual exposures that may vary as a result of exposure to air pollution at nonresidential (eg, work and school) locations, during daily commutes, or to indoor‐generated pollutants such as those produced by heating, cooking, and other combustion sources. However, ambient NO₂ has been shown to be highly correlated with total personal exposure in previous meta‐analyses,78 likely attributable to the infiltration of outdoor pollutants into the home, and the large percentage of time North Americans spend indoors at home. For example, Canadians spend 70% of their time, on average (≈17 hours/d), indoors at home.79 Furthermore, ambient NO₂ was more strongly associated with air‐pollution–related health outcomes in the few studies that attempted to differentiate between ambient and nonambient exposures.78 Therefore, outdoor residential concentrations generated by LUR models provide an appropriate measure of exposure in air pollution health studies.

NO₂ has been widely used as a marker for TRAP80 because it better reflects the local scale heterogeneity in traffic emissions compared with PM_2.5 and O₃ 78, and because NO₂ can be more reliably measured and modeled compared with other traffic pollutants.31 However, there is growing evidence to suggest that NO₂ may be directly linked to some health outcomes81, 82 rather than simply acting as a marker for other TRAP species.

Previous studies reported strong associations between NO₂ and CVD and diabetes mellitus development.82 In addition, previous experimental studies provide evidence that NO₂ exposure may lead to an increase in circulating and tissue (heart) specific inflammatory mediators generated from reactions with inhaled NO₂ in the epithelial lining, suggesting a potential mechanism for NO₂‐induced systemic inflammation.82 However, while the US Environmental Protection Agency's Integrated Science Assessment concluded that long‐term and short‐term exposure to NO₂ were causally associated with respiratory outcomes, they determined that the evidence for a causal association between NO₂ and the development of CVD and diabetes mellitus was “suggestive, but not sufficient,” because past studies do not sufficiently address potential confounding by other TRAP pollutants in studies linking NO₂ with CVD and diabetes mellitus development.82

While we reported associations between NO₂ and atherosclerosis, NO₂ can be highly correlated with other TRAP species, and this study was not designed to differentiate between effects of NO₂ and other TRAP. NO₂ concentrations were estimated using a LUR model, which may not easily distinguish between highly correlated pollutants with similar spatial distribution and sources. There are also nontraffic sources of NO₂, which are reflected in the LUR model predictors. Therefore, further evidence is needed to isolate the impact of disparate pollutants and sources on cardiometabolic disease.

We also reported only cross‐sectional associations. These results are supported by recent findings in MESA‐Air that NO₂ and PM_2.5 were significantly associated with CAC progression.13 However, it is important to consider this in more detail. The most significant limitation of cross‐sectional analyses is the greater potential for exposure misclassification during the period of time that played a biological role in TPA development (ie, if patients relocated from differing exposure settings or if relative exposures in London changed over time). However, previous studies suggest that LUR pollution estimates and long‐term concentration gradients are stable over the time‐frame of this study,35, 36, 37, 38 and that seasonal LUR models adequately capture long‐term concentration gradients, making these models appropriate for estimating long‐term exposure in health studies.82 Cesaroni et al.36 reported that NO₂ models from samples collected 12 years apart showed good agreement, and similar associations with mortality. Eeftens et al.35 found that LUR models for NO₂ measured in 2007 predicted a high proportion (77%) of the variation in NO₂ measurements from 1999 to 2000. Similarly, Wang et al.37 reported that LUR models for NO₂ predicted a high proportion of the variability in measurements collected 7 years prior. Finally, de Hoogh et al.38 found that NO₂ models developed using measurements collected 10 years apart showed good agreement across multiple countries in Western Europe. Furthermore, the population included in the analysis has been shown to have low residential mobility.77

Otherwise, longitudinal analyses are not unequivocally superior. Longitudinal studies of atherosclerosis have been limited by reliance on IMT and CAC progression. IMT represents a phenotype distinct from atherosclerosis,14, 15, 16 and IMT progression does not accurately predict CVD risk compared with single‐time measures.83 Furthermore, the extremely small and variable changes in the progression of IMT decrease the power of longitudinal studies relying on IMT. CAC progression has been inconsistently associated with atherosclerosis and adverse cardiovascular events, with evidence that CAC progression can be associated with plaque rupture, as well as with plaque stability and reduced CVD risk.21, 22, 23 Thus, we believe that our cross‐sectional analysis—given the robust methods of exposure and atherosclerosis assessment—is not a major limitation.

Mediation tests assume that models are correctly specified and assume a causal relationship between potential mediators (ie, diabetes mellitus) and plaque. Causality cannot be verified in the current study because the timing of diabetes mellitus diagnosis relative to plaque measurement is unknown. However, the assumption that diabetes mellitus contributes to plaque burden rather than vice versa is consistent with the existing literature.60, 61, 66

A further limitation is that we did not have data on medication use by individual patients. All patients received similar treatment (ie, usual care or aggressive therapy). Thus, it is unlikely that medication use confounded reported associations between NO₂ and cardiometabolic disorders. However, we cannot account for potential effect modification by medication subtypes.

Another limitation is the long period during which study data were collected. The analyses include data from patients under treatment at SPARC from 1990 to 2013, with 84% of the data collected from 2000 onward. Sensitivity analyses indicate a positive association between NO₂ and TPA in patients across the time span of the study. The association between NO₂ and TPA was stronger in patients with plaque measured from 1990 to 1999. Interestingly, patients from 1990 to 1999 were younger and had lower TPA. Furthermore, SES indicators were not strongly associated with TPA in the 1990–1999 patients (in contrast with the strong association between SES and TPA in 2000–2013 patients), suggesting that some of the increase in estimates of association between NO₂ and TPA among 1990–1999 patients may be the result of uncontrolled confounding by SES in that patient group. In addition to age, TPA, and SES, differences in unmeasured factors such as referral criteria and medication trends (ie, increases in statin use from 2000 onward) may have contributed to stronger associations between NO₂ and TPA in the 1990–1999 patient group. Despite the differences in magnitude, our results suggest a consistently positive association across time periods. Furthermore, hierarchical model results provide additional confirmation that the positive association between NO₂ and plaque was significant, accounting for differences between patients from 1990 to 1999 versus 2000 to 2013.

Finally, the patients in this study are not representative of the general population. Rather, they represent a subset of the general population who may be at higher risk for air pollution–mediated health impacts. Our results suggest that even low levels of TRAP may contribute to atherosclerosis and CVD development in high‐risk populations. Further research is needed to assess impacts of TRAP on atherosclerosis and CVD among healthy adults.

Conclusions

The potential role of air pollution as a contributor to CVD development has enormous public health implications. This study provides unique new evidence for a link between long‐term exposure to TRAP and atherosclerotic plaque burden, as well as cardiometabolic disorders including dyslipidemia, hypertriglyceridemia, and impaired glucose metabolism. Furthermore, our results suggest that air pollution–induced atherosclerotic plaque formation may play a role in CVD susceptibility, particularly among diabetics; and that these processes occur even in cities with relatively low exposures.

Sources of Funding

None.

Disclosures

None.

Supporting information

Data S1–S3Tables S1–S4Figures S1 and S2References 33, 42–44.

(J Am Heart Assoc. 2020;9:e013400 DOI: 10.1161/JAHA.119.013400.)