Traffic-Related Air Pollution: A Systematic Review-Based Human Health Risk Assessment of Mortality

Download the alternative format
(PDF format, 6 MB, 150 pages)

Organization: Health Canada

Published: 2025-03-18

Health Canada – 2025

Acknowledgements

This risk assessment was reviewed by the following external scientific experts:

Dan Crouse, M.Sc., Ph.D. (Health Effects Institute)
Stéphane Buteau, Ph.D. (Université de Montréal)

List of tables
List of figures
List of abbreviations
Executive summary
Chapter 1. Introduction
- 1.1 Background
- 1.2 Objectives and approach
Chapter 2. Methodology
Chapter 3. Literature review and evaluation
Chapter 4. Risk characterization and evaluation of causality
Chapter 5. Conclusion
- Key uncertainties and gaps
References
Appendix. Study details by exposure metric

List of tables

Table 1.1. Summary of mortality outcomes associated with TRAP exposure and conclusions from HEI (2010, 2022)
Table 1.2. Summary of health outcomes and classification of causal associations from Health Canada (2020; 2022c)
Table 2.1. Conversion factors for units in meta-analysis
Table 2.2. Weight of evidence for determination of causality (derived from US EPA 2015)
Table 3.1. Number of articles included in the risk assessment per publication date and exposure metric for each cause of mortality
Table 3.2. General study characteristics by geographical location and year of publication for long-term exposure to TRAP
Table 3.3. RoB assessment

List of figures

Figure 2.1. Study selection process for the scoping review and the systematic review-based assessment on mortality
Figure 3.1. Association between all-cause mortality and long-term exposure to NO₂ in the general population
Figure 3.2. Association between all-cause mortality and long-term exposure to NO_x in the general population
Figure 3.3. Association between all-cause mortality and long-term exposure to PM_2.5 in the general population
Figure 3.4. Association between all-cause mortality and long-term exposure to EC in the general population
Figure 3.5. Association between all-cause mortality and long-term exposure to PM₁₀ in the general population
Figure 3.6. Forest plot of risk estimates for long-term exposure to traffic proximity and all-cause mortality in the general population
Figure 3.7. Forest plot of risk estimates for long-term exposure to traffic proximity and all-cause mortality in patient populations
Figure 3.8. Forest plot of risk estimates for long-term exposure to traffic density and all-cause mortality in the general population
Figure 3.9. Association between CSD mortality and long-term exposure to NO₂ in the general population
Figure 3.10. Association between CHD and long-term exposure to NO₂ in the general population
Figure 3.11. Association between CBVD mortality and long-term exposure to NO₂ in the general population
Figure 3.12. Association between CHD mortality and long-term exposure to NO_x in the general population
Figure 3.13. Association between CBVD mortality and long-term exposure to NO_x in the general population
Figure 3.14. Association between CSD mortality and long-term exposure to PM_2.5 in the general population
Figure 3.15. Association between CHD mortality and long-term exposure to PM_2.5 in the general population
Figure 3.16. Association between CBVD mortality (including stroke) and long-term exposure to PM_2.5 in the general population
Figure 3.17. Association between CHD mortality and long-term exposure to EC in the general population
Figure 3.18. Association between CBVD mortality (including stroke) and long-term exposure to EC in the general population
Figure 3.19. Forest plot of risk estimates for long-term exposure to traffic proximity and CSD mortality in the general population
Figure 3.20. Forest plot of risk estimates for long-term exposure to traffic proximity and CHD and CBVD mortality in the general population
Figure 3.21. Forest plot of risk estimates for long-term exposure to traffic density and CSD and CHD mortality in the general population
Figure 3.22. Association between respiratory mortality and long-term exposure to NO₂ in the general population
Figure 3.23. Association between respiratory mortality and long-term exposure to NO_x in the general population
Figure 3.24. Association between respiratory mortality and long-term exposure to PM_2.5 in the general population
Figure 3.25. Forest plot of risk estimates for long-term exposure to traffic proximity and respiratory mortality in the general population
Figure 3.26. Forest plot of risk estimates for long-term exposure to traffic density and respiratory mortality in the general population
Figure 4.1. Forest plot of pooled risk estimates for exposure to TRAP and risk of all-cause mortality in the general population; n represents the number of studies included in the meta-analysis, and I² the heterogeneity
Figure 4.2. Forest plot of pooled risk estimates for exposure to TRAP and risk of circulatory mortality in the general population; n represents the number of studies included in the meta-analysis, and I² the heterogeneity
Figure 4.3. Forest plot of pooled risk estimates for exposure to TRAP and risk of respiratory mortality in the general population; n represents the number of studies included in the meta-analysis, and I² the heterogeneity

List of abbreviations

ACE: acute coronary event
ACS-CPSII: American Cancer Society – Cancer Prevention Survey II
BC: black carbon
BS: black smoke
CAD: Canadian dollar
CanCHEC: Canadian Census Health and Environment Cohort
CBVD: cerebrovascular disease
CCME: Canadian Council of Ministers of the Environment
CHD: coronary heart disease
CI: confidence interval
CO: carbon monoxide
COPD: chronic obstructive pulmonary disease
CPD: cardiopulmonary disease
CSD: circulatory system disease
DE: diesel exhaust
EC: elemental carbon
ESCAPE: European Study of Cohorts for Air Pollution Effects
GIS: geographic information system
HEI: Health Effects Institute
HF: heart failure
HR: hazard ratio
HRV: heart rate variability
ICD: International Classification of Disease
IHD: ischemic heart disease
LUR: land-use regression
MI: myocardial infarction
NA: not available
NHS: Nurses' Health Study
NLCS: Netherlands Cohort Study on Diet and Cancer
NO: nitric oxide
NO₂: nitrogen dioxide
NO_x: nitrogen oxide
O₃: ozone
OR(s): odds ratio(s)
PAH: polycyclic aromatic hydrocarbon
PM: particulate matter
PM_2.5: particulate matter with a diameter less than 2.5 micrometres
PM_2.5 abs: PM_2.5 absorbance
PM₁₀: particulate matter with a diameter less than 10 micrometres
RE: risk estimate
REVEAL-HBV: Risk Evaluation of Viral Load Elevation and Associated Liver Disease / Cancer Hepatitis B Virus
RoB: risk of bias
ROS: reactive oxygen species
RR(s): relative risk(s)
SALIA: Study on the influence of air pollution on lung function, inflammation and ageing
SES: socio-economic status
SIDIAP: Sistema d'Informació pel Desenvolupament de la Investigació en Atenció Primària
SR-MA: systematic review – meta-analysis
TRAP: traffic-related air pollution
UFPs: ultrafine particles
UK: United Kingdom
USA: United States of America
US EPA: United States Environmental Protection Agency
VOC: volatile organic compound
WHO: World Health Organization

Executive summary

Traffic-related air pollution (TRAP) is the contribution of on-road vehicles to air pollution. It is a mixture of vehicle exhausts, secondary air pollutants formed in the atmosphere from vehicle emissions, evaporative emissions from vehicles, and other non-combustion emissions (e.g., road dust, brake wear, and tire wear). TRAP is of particular concern in urban areas and near highways, where concentrations are the most elevated. Health Canada has estimated that approximately 4 out of 10 Canadians live within 250 m of a high-traffic roadway and that TRAP is responsible for 1,200 premature deaths per year in Canada, which has an estimated annual monetized value of $9 billion (CAD 2015) for this health outcome alone (Health Canada, 2022a, b).

Health Canada has previously evaluated the association between exposure to TRAP and asthma, allergy, lung function, and selected cancer types (Health Canada, 2020, 2022c). The objective of this risk assessment is to evaluate the association of mortality with TRAP exposure. This will inform and support programs and policies designed to mitigate exposure to, and the health impacts of, TRAP in Canada.

For this risk assessment, the epidemiological literature (from January 1, 2000 to July 11, 2022) regarding the associations between TRAP exposure and mortality was evaluated using systematic review techniques, including meta-analysis. Sixty-four primary articles were included in the risk assessment following a librarian-assisted search and screening process. From this evaluation, a weight of evidence approach was used to determine the causal role of TRAP exposure in mortality from all causes, circulatory mortality and some subtypes, as well as respiratory mortality and some subtypes. This weight of evidence included 1) the quantitative analysis of TRAP pollutant concentrations, 2) the qualitative analysis of TRAP pollutants for which meta-analysis could not be conducted, and 3) the qualitative evidence based on various metrics of proximity to traffic and road networks. In addition, mechanistic evidence gathered from risk assessments of TRAP and relevant primary literature was considered to assess the biological plausibility of these associations and to support the determinations of causality. The biological evidence supports a role for TRAP exposure in cellular and tissue dysfunction, including oxidative stress and inflammation, and enhancement of disease progression, which can ultimately lead to mortality.

Based on the overall weight of evidence, it is concluded that:

there is sufficient evidence of a causal relationship between long-term exposure to TRAP and all-cause mortality;
there is sufficient evidence that the relationship between long-term TRAP exposure and mortality from circulatory system disease (CSD) is likely to be causal;
there is sufficient evidence that the relationship between long-term TRAP exposure and mortality from coronary heart disease (CHD) is likely to be causal;
the evidence is suggestive of, but not sufficient to infer, a causal relationship between long-term TRAP exposure and mortality from respiratory disease;
the evidence is suggestive of, but not sufficient to infer, a causal relationship between long-term TRAP exposure and mortality from chronic obstructive pulmonary disease;
there is inadequate evidence to infer a causal relationship between long-term TRAP exposure and mortality from cerebrovascular disease (CBVD).

The size of the evidence base varied considerably depending on the specific cause of mortality evaluated and this is reflected in the causality conclusions. The causality conclusions for all-cause mortality, CSD mortality, and CHD mortality were each founded on a large evidence base and consistency in the meta-analyses. In contrast, a causal relationship could not be determined for CBVD mortality as a result of the limited evidence base and the lack of consistency in the meta-analyses. Additionally, only a limited number of studies were conducted on other causes of mortality such as heart failure and myocardial infarction, and on populations that may be disproportionately impacted, including patient populations.

Additional research and analyses to address the identified data gaps would be useful to further characterize and understand the role of TRAP exposure in mortality. Identification of populations that may be disproportionately impacted by TRAP exposure could also be used to develop targeted policies or programs that would reduce or mitigate their risks. As such, these conclusions may be updated and expanded in the future to include other specific causes of mortality and/or sub-populations.

Chapter 1. Introduction

1.1 Background

As a source of air pollution, traffic-related air pollution (TRAP) is ubiquitous and predominates in urban areas. TRAP refers to the mixture of vehicle exhausts, secondary air pollutants formed in the atmosphere from vehicle emissions, evaporative emissions from vehicles, and other non-combustion emissions (e.g., road dust, brake wear, and tire wear). Approximately 4 out of 10 Canadians live within 250 m of a high-traffic roadway (e.g., expressways, highways, arterial roads, and major roads), an area that is associated with a higher risk of exposure to TRAP (Health Canada, 2022a). Health Canada has also estimated that 1,200 premature deaths per year are attributable to TRAP based on contributions from on-road vehicle emissions to particulate matter with a median mass aerodynamic diameter less than 2.5 micrometres (PM_2.5), nitrogen dioxide (NO₂), and ozone (O₃) ambient concentrations in Canada (Health Canada, 2022b). This analysis also estimated that TRAP contributed to 2.7 million acute respiratory symptom days, 1.1 million restricted activity days, and 210,000 asthma symptom days per year in Canada. The total annual monetized^{Footnote 1} value of this health burden was estimated at $9.5 billion (CAD 2015), with $9 billion associated with premature deaths. As such, the study of the health effects and health burden of TRAP is an active field of research in Canada and around the world.

Exposure to TRAP is particularly challenging to study because of its high spatial and temporal variability, the lack of a unique marker for this source, and differences in vehicle fleet composition over space and time (Khreis and Nieuwenhuijsen, 2017). Two broad categories of surrogates have been widely used in the TRAP literature to assess the contribution of traffic emissions to ambient air pollution: (1) concentrations of individual traffic-related pollutants such as NO₂, nitrogen oxide (NO_x), PM_2.5, PM with a median mass aerodynamic diameter less than 10 micrometres (PM₁₀), and elemental carbon (EC); and (2) measures based on traffic and road network infrastructure such as distance to the nearest road and traffic density (Health Effects Institute [HEI] Panel on the Health Effects of Traffic-Related Air Pollution, 2010, 2022). A brief summary of how these surrogates specifically relate to TRAP follows:

NO_x is primarily emitted from combustion sources and is made up of nitric oxide (NO) and NO₂. Vehicle emissions are predominantly NO which is rapidly converted to NO₂ (HEI, 2010; Health Canada, 2016a). In urban areas, traffic is often the primary source of NO_x in the atmosphere and the main contributor to the variability in NO_x levels (Hamra et al., 2015). NO₂ is considered to be the most direct measure of TRAP exposure as local traffic sources contribute up to 80% of the ambient NO₂ in urban settings (Khreis and Nieuwenhuijsen, 2017). Given the availability of ambient NO₂ measurements, it is the most commonly used surrogate to estimate TRAP exposure.
Particulate matter (PM) is a complex mixture of small liquid and solid particles and is associated with many sources, including vehicle emissions and traffic. PM is categorized based on size: PM₁₀; PM_coarse (PM > 2.5 µm and < 10 µm in aerodynamic diameter); PM_2.5; and ultrafine particles (UFPs) (PM ≤ 0.1 µm in aerodynamic diameter). PM₁₀, PM_coarse, and PM_2.5 are typically measured based on mass concentration (i.e., µg/m³) while UFPs are commonly reported as a number concentration. With respect to TRAP, PM is an important pollutant in the mixture as local traffic is responsible for 9% to 53% of urban PM₁₀, and 9% to 66% of urban PM_2.5 (Khreis and Nieuwenhuijsen, 2017). Additionally, some components of PM are commonly used as surrogates of TRAP, including EC, black carbon (BC), black smoke (BS), and PM_2.5 absorbance (PM_2.5 abs). These carbonaceous pollutants are indicators of diesel exhaust (DE) emissions, especially in cities, and are defined by the techniques used for measurement rather than fundamental differences in chemical properties (HEI, 2022).
Anthropogenic carbon monoxide (CO) is primarily formed by the incomplete combustion of carbon-containing fuels. On-road vehicles are a notable contributor to total emissions of CO in Canada with contributions of 28% according to the 2015 Canadian emissions inventory (Health Canada, 2022b).
Benzene is a naturally occurring constituent of crude oil and is formed through the incomplete combustion of organic materials (Canada, 1993). Concentrations of benzene tend to be higher in vehicles and at urban roadsides, with levels decreasing as distance from the source increases (e.g., levels decreasing from roadside to urban areas to rural areas) (CCME, 2012; HEI, 2010).
Compared with the use of specific air pollutants, measures based on traffic and road network parameters such as distance to roadway and traffic density, are specific to traffic sources and are simple and cost-effective to obtain (Khreis and Nieuwenhuijsen, 2017). However, these metrics may not account for the volume of traffic or the types of vehicles (e.g., proportion of cars and trucks) making up the traffic, which are key determinants for the relative concentrations of pollutants in the TRAP mixture. These metrics may also represent more than air pollution (e.g., noise) and can vary substantially between studies (e.g., differing distances to roadways; different road classifications), limiting the ability to readily compare and contrast the results.

In 2010, the HEI published a critical review of the literature on emissions, exposure, and health effects of TRAP (HEI, 2010). With respect to health effects, the epidemiological literature was evaluated to infer the presence of causal associations between TRAP exposure and health outcomes. In support of that evaluation, the toxicological literature was reviewed to identify any biological mechanism(s) for the purposes of understanding the role of traffic emissions in the effects observed in the epidemiological studies. The HEI review classified the causal associations between exposure to TRAP and a number of health outcomes. As the result of the growing database on the health effects related to TRAP exposure and the significant advances in regulations and vehicular technology, HEI published a Special Report in 2022 focusing on the systematic review of the epidemiological evidence regarding the associations between long-term exposure to TRAP and selected adverse health outcomes (HEI, 2022). The Panel evaluated the confidence in the quality of the body of evidence and assessed the level of confidence in the presence of an association between long-term TRAP exposure and selected outcomes, rather than assessing causality. A summary of the mortality outcomes evaluated as well as the conclusions from the two HEI reports (2010, 2022) is provided in Table 1.1.

Table 1.1. Summary of mortality outcomes associated with TRAP exposure and conclusions from HEI (2010, 2022)^{Footnote a}
Mortality outcome	Classification of causal association (HEI, 2010)	Overall confidence in the evidence (HEI, 2022)
All-cause mortality	Suggestive but not sufficient	High
Cardiovascular mortality Circulatory mortality	Suggestive but not sufficient —	— High
IHD mortality	NA	High
Stroke mortality	NA	Low to moderate
Respiratory mortality	NA	Moderate
COPD mortality	NA	Low
COPD: chronic obstructive pulmonary disease; IHD: ischemic heart disease; NA: not available Footnote a Adapted from Table 7.3 of HEI (2010) and from the executive summary table of HEI (2022) Return to Footnote a referrer

Using an umbrella review-based approach, Health Canada has evaluated the association between TRAP and asthma, allergy, and lung function (Health Canada, 2020), and the association between TRAP and selected cancer types (Health Canada, 2022c). The conclusions of these assessments are provided in Table 1.2. This current report focuses on TRAP exposure and the risk of mortality.

Table 1.2. Summary of health outcomes and classification of causal associations from Health Canada (2020; 2022c)
Health outcome	Classification of causal association
Asthma incidence (children)	Causal relationship
Asthma prevalence (children)	Causal relationship
Asthma incidence (adults)	Inadequate to infer a causal relationship
Asthma prevalence (adults)	Suggestive of, but not sufficient to infer, a causal relationship
Lung function	Likely to be a causal relationship
Allergic sensitization and allergic responses	Suggestive of, but not sufficient to infer, a causal relationship
Lung cancer (adults)	Causal relationship
Breast cancer (adults)	Suggestive of, but not sufficient to infer, a causal relationship
Childhood leukemia	Likely to be a causal relationship

1.2 Objectives and approach

The objective of this risk assessment is to use systematic review techniques, including meta-analysis, in the evaluation of the epidemiological literature regarding the associations between exposure to TRAP and mortality. A weight of evidence approach was used to determine the causal role of TRAP in mortality, including mortality from all causes, circulatory mortality and some subtypes, and respiratory mortality and some subtypes. Additionally, mechanistic evidence gathered from recent risk assessments of TRAP and recent primary literature was considered to assess the biological plausibility of the associations identified in the review of the epidemiological evidence and to support the determinations of causality.

This risk assessment document is organized as follows:

Chapter 1 provides background information and describes the objectives and approach;
Chapter 2 describes the methodology undertaken for this risk assessment in detail;
Chapter 3 presents the evidence from the included epidemiological primary articles and data analysis;
Chapter 4 critically evaluates the evidence from Chapter 3 and presents biological evidence to support the findings from the epidemiological studies for determination of causality;
Chapter 5 presents the conclusions and identifies key uncertainties and gaps.

A supporting document is also available upon request. This document includes the refined search strategy for the literature update, the Risk of Bias (RoB) guidelines used for the assessment of cohort studies, additional quantitative analyses of the epidemiological studies, and tabular summaries of the biological evidence.

Chapter 2. Methodology

In this chapter, the methodology for this risk assessment is described in detail. Section 2.1 summarizes the process used to identify the relevant epidemiological literature relating to TRAP exposure and adverse health outcomes, including the scoping review, the TRAP and traffic exposure framework, and the subsequent literature search updates specific to mortality endpoints considered in this risk assessment. Section 2.2 details the screening, data extraction, RoB assessment, and data analysis. Section 2.3 describes how the biological evidence related to TRAP and the health endpoints of interest were used to assess biological plausibility. Lastly, section 2.4 presents the criteria used to determine the level of causality in the weight of evidence approach.

2.1 Identification of relevant literature

2.1.1 Scoping review

As a first step, a scoping review^{Footnote 2} of the epidemiological literature on the human health effects of TRAP was conducted (Matz et al., 2019). The primary research question for this scoping review was as follows: "What is the current body of scientific literature regarding the association between TRAP exposure and adverse human health endpoints, including effects on various systems (respiratory, circulatory, immunological, reproductive/developmental, and nervous), as well as other health endpoints such as cancer and mortality?". The scoping review included primary epidemiological research articles and some review types (as described below) that were published in peer-reviewed journals and address the scoping review objectives. The observational study designs that were included were case-control, cohort, cross-sectional, panel, ecological, time-series, and case-crossover designs. Biological studies were included only if human subjects were involved in the study (i.e., controlled human exposure studies). Review types included in the scoping review were systematic reviews, meta-analyses, scoping reviews, and critical reviews that included an evaluation of causal association. With respect to TRAP and traffic exposure metrics, the inclusion criteria were adapted from the critical review of TRAP by the HEI (2010). These criteria allowed the reviewers to identify the studies that were TRAP- or traffic-centric from a larger body of general air pollution studies. Exposure metrics meeting the inclusion criteria were distance to roadways; measures of traffic density; modelling (e.g., land-use regression [LUR] and dispersion) that estimated traffic-specific exposure; traffic-based source apportionment; occupations characterized by traffic exposure (e.g., taxi drivers and truckers); subjects in locations characterized by level of traffic exposure (e.g., high- vs. low-exposure sites); and monitoring of TRAP-related pollutants (e.g., NO₂ and BC) when the measurements could be reasonably related to traffic sources (e.g., roadway-specific monitoring). To target TRAP-related exposures, studies that characterized exposure based on proximity to gas stations or service stations were excluded from the scoping review.

The literature searches were conducted by a Health Canada librarian in two databases, Ovid Embase and Ovid MEDLINE, and covered the period from January 1, 2000, to April 4, 2018. The detailed search strategy and inclusion criteria are described in Matz et al. (2019). The references identified from the literature search were screened independently by two reviewers for eligibility, first by title and abstract and then by full text; disagreements were resolved by consensus. To generate the evidence map, data extraction included study design parameters and human health outcomes. Descriptive summary tables were developed to provide a high-level summary of the number and types of articles evaluating the different types of health effects and cross-tabulations by study design parameters. The entire review process was managed using DistillerSR (DistillerSR Inc., Ottawa, ON).

From the scoping review, the association between TRAP exposure and mortality was identified as a candidate for a human health risk assessment, as a relatively large body of primary literature was identified in the evidence map. In addition, mortality is the most severe health outcome, so it is important to understand the causes and associated risks for the population.

2.1.2 Framework for traffic and TRAP exposure assessment

HEI published a protocol for a systematic review and meta-analysis of the selected health effects of long-term exposure to TRAP (HEI, 2019) to support its updated assessment of TRAP (2022). This protocol included an updated exposure framework for the evaluation of the health effects of TRAP in epidemiological studies. This framework retained the fundamental concepts of the 2010 HEI report while refining the criteria for identifying studies in which the exposure contrast(s) of TRAP pollutant(s) were primarily due to traffic sources. This refinement aided in the identification and selection of TRAP-specific studies from the larger body of air pollution research (i.e., studies of ambient air pollution considering all sources). Additionally, the HEI protocol indicated that studies of occupational exposure to TRAP (e.g., in taxi drivers and truckers) had not been considered useful in HEI's 2010 report and that these studies would be difficult to combine with ambient exposure studies; and that, as a result, occupational exposures were not considered in scope for HEI's updated report on TRAP (HEI, 2019, 2022).

For the present risk assessment of TRAP exposure and mortality, the primary studies identified from the scoping review and literature search updates were also assessed for inclusion or exclusion based on the updated HEI exposure framework. The details of the exposure assessment framework developed by HEI are described in detail in their TRAP review documents (HEI, 2010, 2019, 2022) and the key concepts are summarized in this section.

For any urban area, the overall air quality is a composition of:

the regional background air pollutants entering the city;
the urban background air pollutants dispersed from primary emissions sources, which includes traffic and other sources, and associated secondary air pollutants of these sources; and
the direct contribution from local traffic sources (e.g., hot spots).

In urban areas, TRAP emissions are greater in areas with higher levels of traffic, and the increased levels of TRAP pollutants in these areas are readily attributed to the roadway network and traffic volumes. For this near-roadway environment, factors such as traffic volume, fleet composition, fuel source, and driving behaviour are the main determinants of traffic emissions. Diurnal patterns of pollutant concentrations are observable corresponding with traffic volumes (e.g., morning and evening peaks associated with rush-hour traffic). Additionally, as the distance from roadways increases, concentrations of the primary TRAP pollutants (e.g., CO, NO, EC, UFPs) decrease. As the primary traffic emissions are dispersed from the near-road environment, they mix with other pollutants and are subject to the same chemical and physical processes as pollutants from other sources, and eventually become a part of the overall urban background.

Given this complexity of urban air quality, three strategies were considered in combination to identify studies in which the exposure contrast(s) of TRAP pollutant(s) were primarily due to traffic sources:

the pollutant(s) or traffic metric(s);
the spatial scale; and
the exposure assessment method and its spatial resolution.

First, as there is no pollutant unique to TRAP, certain pollutants were considered to represent traffic sources better than others (e.g., NO₂ vs. polycyclic aromatic hydrocarbons [PAHs]). For pollutants with regional contributions from other sources (e.g., PM_2.5, PM_coarse, and PM₁₀), the exposure contrasts within an urban setting (i.e., within-city contrasts) were considered to be largely attributable to traffic sources, as the regional component is largely constant across large areas. Additionally, if the evidence for an association with a health endpoint was for PM but not for other pollutants with greater TRAP specificity, such as NO₂ or EC, the confidence in the association with TRAP was lowered. Measures of traffic based on road network infrastructure (e.g., distance to roadway or traffic density) were considered to be highly specific markers of variations in traffic exposure, though indirect measures of TRAP exposure.

Second, the spatial scale of a study is an important element of evaluating the specificity of the exposure contrast(s) for TRAP. Spatial scales were defined as regional (>50 km), urban (5 km to 50 km), neighbourhood (1 km to 5 km), and local (<1 km). At the regional and urban scales, it is difficult to isolate the contrasts in TRAP compared with the contrasts due to all sources. In comparison, exposure contrasts at the neighbourhood and local scales were considered relevant to TRAP. As such, studies that used within-city comparisons were included. However, studies that relied on between-city comparisons and studies that considered very large geographical areas (e.g., nationwide or statewide studies) were excluded due to lack of specificity for TRAP compared with other sources – the exposure contrast being considered mostly attributable to regional differences in all sources, and not attributable to differences in traffic.

Third, the spatial resolution of the exposure assessment method was evaluated to select the studies in which the exposure contrasts were due to within-city variations in traffic at the local to neighbourhood scale. Thus, the required resolution of the air pollutant exposure surface was ≤5 km (i.e., 5 x 5 km grid or smaller) corresponding to the upper range of the neighbourhood scale. For indirect measures of TRAP, based on road network infrastructure, the required resolution was ≤1 km (i.e., ≤1 km from a major roadway or highway), corresponding to the upper range of the local scale. Additionally, the required spatial resolution of the address data (e.g., exact address, detailed postal code, census blocks) of the study participants was determined to be 5 km for air pollutants and 100 m for indirect exposure metrics.

2.1.3 Literature search update and handsearching

For this risk assessment, literature search updates using a refined search strategy with the scope limited to studies of mortality, were conducted by a Health Canada librarian using two databases, Ovid Embase and Ovid Medline, on October 16, 2020 and July 11, 2022. This refined search strategy is provided in section 1 of the supporting documentation. The primary studies identified through the literature search updates were assessed for inclusion using the same criteria and method as employed in the scoping review (Matz et al., 2019) and with the updated HEI exposure framework (HEI, 2019). The screening and selection of relevant primary studies from the literature search updates were managed using DistillerSR (DistillerSR Inc., Ottawa, ON).

To supplement the literature searches, the references from two recent systematic review–meta-analyses (SR-MAs) of long-term exposure to NO₂ and mortality (Huangfu and Atkinson, 2020; Stieb et al., 2021) and one recent SR-MA of long-term exposure to PM_2.5 (Chen and Hoek, 2020) were handsearched. Two of these three SR-MAs (Huanfu and Atkinson, 2020; Chen and Hoek, 2020) had been conducted to inform air quality guidelines developed by the World Health Organization (WHO, 2021).

2.2 Screening, data extraction, risk of bias assessment, and data analysis

2.2.1 Screening and data extraction

All studies identified from the scoping review (n = 110), literature search updates (n = 902), and handsearching (n = 14) were screened based on the exposure assessment framework identified in section 2.1.2 and for evaluation of mortality as a health endpoint. For mortality, both all-cause and cause-specific mortality were considered. For this assessment, all-cause mortality referred to the total number of deaths (i.e., death from all causes) as well as deaths that were not the result of an accident or suicide (i.e., non-accidental or natural). Lung cancer mortality was excluded as the association between TRAP exposure and lung cancer in adults was previously assessed by Health Canada (2022c) and determined to have a causal relationship. Title and abstract screening and full-text screening were performed in duplicate; any discrepancies were resolved by consensus and/or consultation with a third reviewer. As a result, 64 primary articles were included in the risk assessment. None of the studies evaluating short-term exposure to TRAP pollutants met the criteria for traffic specificity as per the exposure framework described in section 2.1.2 (e.g., exposures were assigned at the city level).

The study selection process is depicted in Figure 2.1.

Figure 2.1: Text description

Figure 2.1 depicts the flow of information through the different phases of a systematic review and maps out the number of records identified, included, and/or excluded at each phase. For the scoping review: 16,328 records were identified through database searching, of which 11,797 records remained after the duplicates were removed. 11,797 records were then screened for title and abstract screening, at which point 9,435 records were excluded and the full-text of 2,362 primary and review articles were assessed for eligibility. Of these full-text articles, 1,334 were excluded with reason leaving 956 primary articles and 72 review articles in the evidence map. For the TRAP – Mortality assessment: 110 of the 956 primary articles were eligible for mortality and 49 of the 902 primary articles identified during literature search updates were eligible for inclusion. These literature search updates were conducted in October 2020 and July 2022. Full-text evaluation of the 159 articles resulted in the exclusion of 109 articles and an additional 14 articles were identified from secondary searches, leaving 64 articles in the assessment.

For each study, several domains of data were extracted: bibliographic information; study period and location; study size and demographics; pollutant or exposure metric, exposure assessment method, spatial scale, and exposure distribution; cause of death and International Classification of Disease (ICD) codes (if available); model covariates (including potential confounders); and risk estimates (REs) including 95% confidence intervals (CIs) and pollutant increments. Full cohort analyses were selected over case-cohort analyses. Data extraction was completed by two reviewers with partial duplication of approximately 30% of the included studies; any discrepancies were resolved by consensus and/or consultation with a third reviewer.

2.2.2 Risk of bias assessment

As an important element of the systematic review process, a RoB assessment was conducted to assess the validity of included studies and to establish transparency in the evidence synthesis of results (Higgins et al., 2011). The RoB assessment tool employed for this risk assessment was based on the criteria proposed by Lam et al. (2016) in a systematic review and meta-analysis on air pollution and autism spectrum disorder, and was utilized in Stieb et al. (2021) and Health Canada (2022d) with minor modifications. The definitions and guidelines for the RoB tool for the assessment of cohort studies are presented in section 2 of the supporting documentation. The RoB at the study level was evaluated for the following domains: selection bias and generalizability, exposure assessment with regards to modelling and monitoring, confounding, outcome assessment, completeness of outcome data, selective outcome reporting, conflict of interest, and other sources of bias. With respect to the confounding domain, age, sex, smoking, and individual- or area-level socio-economic status (SES) were chosen a priori as critical potential confounders. The same RoB guidelines were used to evaluate case-control studies, with the exception of the selection bias domain, where professional judgment was used. Two reviewers assessed the RoB for each study and discrepancies between the assessments of the two reviewers were resolved by consensus and consultation with a third reviewer when necessary.

2.2.3 Data analysis

Data analysis was performed separately for each individual exposure-outcome pair (e.g., NO₂ – all-cause mortality). The exposure candidates included NO₂, NO_x, NO, PM_2.5, PM₁₀, EC, benzene, CO, traffic proximity, and traffic density. The outcome candidates consisted of all-cause mortality, circulatory mortality and its specific causes, and respiratory mortality and its specific causes. For each exposure-outcome pair, the most appropriate RE (i.e., derived using the most appropriate statistical model and/or identified as the primary findings by the study authors) was selected from each study, and if two or more studies provided REs for the same cohort, only one study from the same cohort was selected for the analysis. The term risk estimate (RE) is used in this report to encompass each risk ratio measure identified by individual studies when referring to them collectively in the meta-analysis.

Statistical significance of the REs was assessed based on the 95% CI. A RE was considered significant if the CI did not include 1.0 (i.e., the null association) and borderline significant if, for REs greater than 1, the lower 95% confidence limit was 0.9 to 1.0, inclusively, or, for REs less than 1, the upper confidence limit was 1.0 to 1.1, inclusively.

Data analysis was undertaken quantitatively where possible, or qualitatively when quantitative analysis could not be performed. Pollutant-outcome pairs with a minimum number of REs (n = 4) were quantitatively analyzed, and pollutant-outcome pairs that did not meet the minimum number of REs were described qualitatively. Traffic and roadway infrastructure metrics (i.e., traffic proximity and traffic density) were analyzed qualitatively, as there is insufficient consistency and standardization in these metrics to allow for quantitative synthesis.

Quantitative analysis: meta-analysis

Meta-analysis was performed for traffic pollutant–outcome pairs that had a minimum of four REs available, similar to other published systematic reviews with meta-analysis in the field (Khreis et al., 2017). Only REs based on single-pollutant models that could be recalculated on a standardized scale (i.e., continuous, but not dichotomous or log-transformed results) were eligible to be included in the meta-analysis. REs from multipollutant models were not retained as the focus was on the TRAP mixture and not its individual components. Of note, for most of the studies, the analyses were conducted using single-pollutant models. Like HEI (2022), Health Canada determined that due to a high degree of heterogeneity between the general population and patient populations, studies from these two population groups would not be pooled together and are presented separately.

Meta-analysis was conducted in R version 4.1.2 (R Core Team, 2021) using the metafor package (Viechtbauer, 2010). For each exposure-outcome pair, the study-specific REs were pooled using the random-effects modelling approach. The random-effects approach was chosen over the fixed-effects approach as it allows generalization of the conclusions beyond the particular set of studies included in the analysis. The models were fitted using Restricted Maximum Likelihood. Hazard ratios (HRs), relative risks (RRs), and odds ratios (ORs) were all eligible for pooling in the same meta-analysis, given that the outcome of interest is common but the effect size is relatively small (Davies et al., 1998), in line with what has been done in previous meta-analyses (Khreis et al., 2017; HEI, 2022). Conversion factors for pollutants to be converted to µg/m³ were based on those presented in HEI (2022) and are outlined in Table 2.1. For all pollutants except EC, the pooled REs were converted to an increment of 10 µg/m³ of pollutant. The pooled REs for EC were converted to an increment of 1 µg/m³ of pollutant. Heterogeneity was evaluated using the I² statistic, which represents the percent of total variance attributable to heterogeneity. Values of 0%–40%, 30%–60%, 50%–90%, and 75%–100% correspond to the following categories: might not be important, may represent moderate heterogeneity, may represent substantial heterogeneity, and may represent considerable heterogeneity, respectively (Higgins et al., 2022).

Table 2.1. Conversion factors for units in meta-analysis
Pollutant	Conversion	Factor^{Footnote a}
NO₂	ppb to µg/m³	1.88
NO	ppb to µg/m³	1.23
NO_x	ppb to µg/m³	1.55
CO	ppm to mg/m³	1.15
BC	µg/m³ to µg/m³ EC	1.25
BS	µg/m³ to µg/m³ EC	0.11
PM_2.5 abs	10^-5/m to µg/m³ EC	1.1
Footnote a Based on HEI 2022 Report Appendix 5C. Scaling and conversion factors for use in meta-analysis Return to Footnote a referrer

Meta-analysis can only be performed on unique cohorts (Higgins et al., 2022). Therefore, the rationales for selecting the most appropriate RE for inclusion into the meta-analysis were as follows:

If two or more studies provided REs on the same cohort, the study with the cohort that had the longest length of follow-up, largest sample size, and greater overall completeness of data necessary to perform a meta-analysis (e.g., RE, 95% CI) was selected for inclusion in the meta-analysis.
If one study presented REs based on multiple exposure assignments (e.g., at baseline, average, last year in study), the RE calculated using a moving average, with longer ranging averages preferred, was selected. Moving averages are often chosen to stabilize REs (van Donkelaar et al., 2015).
If one study presented REs based on multiple methods of exposure measurement (e.g., LUR model, dispersion model), REs based on LUR models were chosen over dispersion models for better representation of TRAP (de Hoogh et al., 2014).

Additional analyses were conducted to ascertain the robustness of results of the main meta-analysis, specifically subgroup analyses, sensitivity analyses, and tests for influence and publication bias. First, subgroup analysis was conducted based on exposure assessment methodology (e.g., LUR or dispersion modelling). The LUR model subgroup analysis also included an exposure assessment using hybrid models (i.e., assessment methods that combined more than one modelling technique such as a dispersion model with a chemical transport model) as they were deemed sufficiently similar for the purposes of meta-analysis. Subgroup analysis by geographical region was not conducted due to an insufficient number of Canadian or North American studies. Second, sensitivity analyses were performed to identify whether studies with high RoB were influencing the results; only studies with low or probably low ratings for RoB in the exposure assessment and confounding domains were included. Finally, the influence of individual studies on the overall pooled RE was examined using leave-one-out analysis. Evidence of the potential effect of publication bias was investigated by identifying asymmetry in funnel plots, performing Begg's rank correlation and Egger's regression tests, and using the trim-and-fill method.

2.3 Biological evidence

An evaluation of biological evidence was conducted to assess the biological plausibility of the associations between TRAP exposure and mortality. The studies included in the evaluation of the biological evidence typically consider short-term exposures and are designed to detect subtle changes in biomarkers which are relatable to long-term health effects observed in the epidemiological literature. For efficiency, this evaluation was non-exhaustive and built on existing reviews of TRAP by Health Canada (2020, 2022c) and HEI (2010, 2022). In addition, a search of the primary literature from the past 10 years (i.e., 2013–2022) was conducted to identify the most relevant recent studies. This literature search focused on human and experimental animal studies that evaluated real-world TRAP exposures, as these provide the most direct evidence of the associated biological effects. Controlled exposure studies of vehicle exhaust (i.e., DE) were also included as they were considered to represent an important component of the total TRAP mixture and the study design addressed some of the confounding associated with panel study design (e.g., differences in noise between exposure locations). The literature review centred on health effects associated with the circulatory and respiratory systems, as these were determined to be the most relevant for this risk assessment, and they comprised a majority of the biological evidence literature.

2.4 Determination of causality

The quantitative estimates of TRAP pollutant concentrations were considered to provide the highest level of evidence, while the qualitative analysis of the metrics of traffic and the road network infrastructure and of TRAP pollutants for which quantitative analysis was not undertaken provided support in the evidence base for the determination of causality. The biological evidence was used to support the associations observed in the epidemiological literature as well as to support a determination of causality.

In the weight of evidence approach used in this assessment to determine the causal role of TRAP in the development of specific health effects, consideration is given to a number of criteria, including those of causal inference developed by Bradford Hill (1965). The criteria, widely used in reviews of epidemiological literature and considered collectively in the weight of evidence evaluation, are as follows:

Biological plausibility: there is a plausible mechanism between the exposure and the effect;
Temporal sequence: the exposure precedes the health outcome;
Consistency of the association: the association is reported by different researchers, for different study designs, in different populations, etc.;
Coherence: evidence from toxicological studies, controlled human exposure studies, and epidemiological studies of various types provides support for the effects observed and potential modes of action;
Biological gradient: there is evidence of an exposure-response relationship;
Strength of the association: the greater the magnitude of the RE, the less likely that the relationship is due to uncontrolled residual confounding; and
Robustness of the association: the associations are robust to model specifications and adjustment for potential confounders such as weather, temporal trends, and co-occurring pollutants.

These criteria are used to inform a conclusion as to whether the relationship between TRAP exposure and a health effect is causal, likely to be causal, suggestive of a causal relationship, inadequate to infer a causal relationship, or not likely to be causal. The definitions of each of these determinations of causality are derived from the United States Environmental Protection Agency (US EPA, 2015) and are provided in Table 2.2. Health Canada has previously used this causality framework in the risk assessments of DE (2016b); NO₂ (2016b); gasoline exhaust (2017); TRAP and asthma, allergy, and lung function (Health Canada, 2020); and TRAP and selected cancer types (Health Canada, 2022c).

Table 2.2. Weight of evidence for determination of causality (derived from US EPA 2015)
Relationship	Description
Causal relationship	Evidence is sufficient to conclude that there is a causal relationship with relevant pollutant exposures (e.g., doses or exposures generally within one to two orders of magnitude of recent concentrations). That is, the pollutant has been shown to result in health effects in studies in which chance, confounding, and other biases could be ruled out with reasonable confidence. For example: (1) controlled human exposure studies that demonstrate consistent effects, or (2) observational studies that cannot be explained by plausible alternatives or that are supported by other lines of evidence (e.g., animal studies or mode of action information). Generally, the determination is based on multiple high-quality studies conducted by multiple research groups.
Likely to be a causal relationship	Evidence is sufficient to conclude that a causal relationship is likely to exist with relevant pollutant exposures. That is, the pollutant has been shown to result in health effects in studies where results are not explained by chance, confounding, and other biases, but uncertainties remain in the evidence overall. For example: (1) observational studies show association, but co-pollutant exposures are difficult to address and/or other lines of evidence (controlled human exposure, animal, or mode of action information) are limited or inconsistent; or (2) animal toxicological evidence from multiple studies from different laboratories demonstrates effects, but limited or no human data are available. Generally, the determination is based on multiple high-quality studies.
Suggestive of, but not sufficient to infer, a causal relationship	Evidence is suggestive of a causal relationship with relevant pollutant exposures but is limited because chance, confounding, and other biases cannot be ruled out. For example: (1) when the body of evidence is relatively small, at least one high-quality epidemiologic study shows an association with a given health outcome and/or at least one high-quality toxicological study shows effects relevant to humans in animal species; or (2) when the body of evidence is relatively large, evidence from studies of varying quality is generally supportive but not entirely consistent, and there may be coherence across lines of evidence (e.g., animal studies or mode of action information) to support the determination.
Inadequate to infer a causal relationship	Evidence is inadequate to determine that a causal relationship exists with relevant pollutant exposures. The available studies are of insufficient quantity, quality, consistency, or statistical power to permit a conclusion regarding the presence or absence of an effect.
Not likely to be a causal relationship	Evidence indicates there is no causal relationship with relevant pollutant exposures. Several adequate studies, covering the full range of levels of exposure that human beings are known to encounter and considering at-risk populations and life stages, are mutually consistent in not showing an effect at any level of exposure.

Chapter 3. Literature review and evaluation

3.1 Characteristics of the included articles

Overall, 64 primary articles were included in the risk assessment (details are provided in section 2.2.1); all were long-term exposure studies. Most of these articles used a cohort study design (n = 62); the remaining two articles were case-control studies (Rosenlund et al., 2006, 2009).

The most frequently examined cause of death was all-cause (including natural and non-accidental) followed by circulatory and respiratory (Table 3.1). Other causes of mortality evaluated in more than one article included various types of cancer other than lung cancer (n = 6), a proportion of overall mortality (i.e., overall mortality not including certain specific causes of mortality; n = 6), diabetes (n = 3), digestive system diseases (n = 3), and endocrine disorders (n = 2). Table 3.1 provides a breakdown of the 64 articles included in the analysis with the type of exposure metric measured for each of the causes of mortality reported.

Table 3.1. Number of articles^{Footnote a} included in the risk assessment per publication date and exposure metric for each cause of mortality
Cause of mortality	Number of primary articles	Publication date from 2013–2022	TRAP pollutant	Traffic and road network infrastructure
All-cause	47	29	31	20
Circulatory	41	22	27	19
Respiratory	20	13	15	7
Other	14	6	7	8
Footnote a Numbers reported do not add up to the total number of articles included in this risk assessment as some articles reported more than one cause of mortality and/or used more than one exposure metric. Return to Footnote a referrer

For all-cause, circulatory, and respiratory mortality, the majority of the articles identified were relatively recent with a publication date after 2012. TRAP pollutants were more commonly used than traffic and the road network infrastructure to assess TRAP exposure in articles investigating all-cause, circulatory, and respiratory mortality.

General study characteristics of the 64 primary articles are provided in Table 3.2. An overall synthesis of these articles is presented in this section followed by a more in-depth evaluation by cause of mortality in the subsequent sections of this chapter. Overall, the cohort studies were conducted in Canada (n = 10), the United States of America (USA) (n = 12), Europe (n = 30), Asia (n = 8), and Australia (n = 2), while the case-control studies were conducted in Europe (n = 2). The studies were mostly based on the general population, with the exception of 16 articles that pertained to studies with various patient populations, a combination of general and patient populations, or patient populations for only certain mortality outcomes. Diverse exposure assessment methods were used to assess long-term exposure to TRAP and mortality, including modelling such as LUR (n = 22), dispersion (n = 13), and hybrid or other models (n = 7), as well as traffic and road network infrastructure metrics (n = 32). The most frequently assessed TRAP pollutants were NO₂ (n = 28), PM_2.5 (n = 20), PM₁₀ (n = 13), and NO_x (n = 13). A number of other pollutants attributed to TRAP were also used to evaluate exposure in a limited number of articles; they include BC (n = 6), NO (n = 5), PM_2.5 abs (n = 6), PM_coarse (n = 4), CO (n = 2), EC (n = 1), and benzene (n = 1). Although UFPs are a key marker for TRAP, no studies were identified for this pollutant. With regards to traffic and the road network infrastructure, traffic proximity (n = 26) was the most commonly used metric, followed by traffic density (n = 13).

Table 3.2. General study characteristics by geographical location and year of publication for long-term exposure to TRAP
First author and year	Cohort^{Footnote a} (Study period)	Study location	Study population	TRAP pollutant/ metric	Exposure assessment method	Cause of death^{Footnote b}
Canada
Finkelstein 2004	(1992–2001)	Hamilton, ON	Patient	Traffic proximity	Traffic and road network infrastructure	A
Finkelstein 2005	(1985–1999)	Hamilton, ON	Patient	Traffic proximity	Traffic and road network infrastructure	C, R
Jerrett 2009	(1992–2002)	Toronto, ON	Patient	NO₂ (all), traffic proximity (A, C, O)	LUR model, traffic and road network infrastructure	A, C, R, O
Gan 2010	(1994–2002)	Vancouver, BC	General	Traffic proximity	Traffic and road network infrastructure	C
Gan 2011	(1994–2002)	Vancouver, BC	General	NO, NO₂, PM_2.5, BC	LUR model	C
Chen 2013	Ontario Tax Cohort (1982–2004)	Toronto, Hamilton, Windsor, ON	General	NO₂, traffic proximity	LUR model, traffic and road network infrastructure	C
Gan 2013	(1999–2002)	Metropolitan Vancouver, BC	General	NO, NO₂, PM_2.5, BC	LUR model	R
Villeneuve 2013	Ontario Tax Cohort (1982–2004)	Toronto, ON	General	NO₂, benzene	LUR model	A, C, R, O
Crouse 2015	Canadian Census Health and Environment Cohort (CanCHEC) 1991 (1991–2006)	10 cities (Edmonton, Hamilton, London, Montreal, Sarnia, Toronto, Vancouver, Victoria, Windsor, Winnipeg)	General	NO₂	LUR model	A, C, R
Cakmak 2019	CanCHEC 1991 (1991–2011)	Canada	General	Traffic proximity	Traffic and road network infrastructure	A, C, R, O
USA
Jerrett 2005	American Cancer Society – Cancer Prevention Survey II (ACS-CPSII) (1982–2002)	Los Angeles, CA	General	PM_2.5	Hybrid or other models	A, C, O
Medina-Ramón 2008	Worcester Heart Failure Study (2000–2005)	Worcester, MA	Patient	Traffic proximity, traffic density	Traffic and road network infrastructure	A
Krewski 2009	ACS-CPSII (1982–2000)	New York City, NY Los Angeles, CA	General	PM_2.5	LUR model	A, C, O
von Klot 2009	Worcester Heart Attack Study (1995–2005)	Worcester metropolitan area, MA	Patient	EC	LUR model	A
Hart 2013	Nurses' Health Study (NHS) (1990–2008)	All 50 states	General	Traffic proximity	Traffic and road network infrastructure	A
Wilker 2013	(1999–2012)	Boston, MA	Patient	Traffic proximity	Traffic and road network infrastructure	A
Hart 2014	NHS (1986–2012)	All 50 states	General	Traffic proximity	Traffic and road network infrastructure	C
Blount 2017	(2000–2012)	California	Patient	Traffic density	Traffic and road network infrastructure	A
Alexeeff 2018	Kaiser Permanente Northern California (2010–2015)	Oakland, CA	General	NO, NO₂, BC	Hybrid or other models	C
Kulick 2018	The Northern Manhattan Study (1993–2016)	Northern Manhattan, NY	General	Traffic proximity	Traffic and road network infrastructure	A, C
DuPré 2019	NHS and NHS II (1988–2008)	NHS: 11 states NHSII: 14 states	General	Traffic proximity	Traffic and road network infrastructure	O
Villanueva 2021	(1996–2016)	California	General	Traffic proximity	Traffic and road network infrastructure	O
Europe
Hoek 2002	Netherlands Cohort Study on Diet and Cancer (NLCS) (1986–1994)	Netherlands	General	Traffic proximity	Traffic and road network infrastructure	A, C, O
Nafstad 2004	(1972–1998)	Oslo, Norway	General	NO_x	Dispersion model	A, C, R
Gehring 2006	Study on the influence of air pollution on lung function, inflammation and ageing (SALIA) (1985–2003)	North Rhine–Westphalia, Germany	General	Traffic proximity	Traffic and road network infrastructure	A, C, O
Rosenlund 2006	Case-control (1992–1994)	Stockholm, Sweden	General	NO₂, PM₁₀, CO	Dispersion model	C
Naess 2007	(1992–1998)	Oslo, Norway	General	NO₂ (C, R), PM₁₀ (C, R), PM_2.5 (all)	Dispersion model	A, C, R
Rosenlund 2008	(1998– 2005 [A]; 1998–2000 [C])	Rome, Italy	Patient (A), General (C)	NO₂	LUR model	A, C
Beelen 2008	NLCS-AIR Study (1987–1996)	Netherlands	General	Traffic proximity, traffic density	Traffic and road network infrastructure	A, C, R, O
Beelen 2009	NLCS (1987–1996)	Netherlands	General	Traffic proximity, traffic density	Traffic and road network infrastructure	C
Brunekreef 2009	NLCS-AIR Study (1986–1996; followed from 1987–1996)	Netherlands	General	Traffic proximity, traffic density	Traffic and road network infrastructure	A, C, R, O
Rosenlund 2009	Case-control (1985–1996)	Stockholm, Sweden	General	NO₂, PM₁₀, CO	Dispersion model	C
Huss 2010	Swiss National Cohort (2000–2005)	Switzerland	General	Traffic proximity	Traffic and road network infrastructure	C
Maheswaran 2010	(1995–2006)	London, United Kingdom (UK)	Patient	NO₂, PM₁₀	Hybrid or other models	A
Nawrot 2011	(1997–2009)	Leuven, Belgium	Patient	Traffic proximity	Traffic and road network infrastructure	A
Cesaroni 2012	Rome Longitudinal Study (2001–2006)	Rome, Italy	General	NO₂	LUR model	A
Raaschou-Nielsen 2012	Diet, Cancer and Health cohort study (1993–2009)	Copenhagen and Aarhus, Denmark	General	NO₂, traffic proximity, traffic density	Hybrid or other models, traffic and road network infrastructure	A, C
Carey 2013	Clinical Practice Research Datalink (2003–2007)	England	General	NO₂, PM_2.5, PM₁₀	Dispersion Model	A, C, R
Cesaroni 2013	Rome Longitudinal Study (2001–2010)	Rome, Italy	General	NO₂, PM_2.5, traffic proximity, traffic density	LUR model, dispersion model, traffic and road network infrastructure	A, C, R
Heinrich 2013	SALIA (1985–2008)	North Rhine-Westphalia, Germany	General	Traffic proximity	Traffic and road network infrastructure	A, C, R
Beelen 2014a	European Study of Cohorts for Air Pollution Effects (ESCAPE) (earliest 1985; follow-up range from 6.3 to 18.6 y)	13 countries	General	NO₂, NO_x, PM_2.5, PM_2.5 abs, PM₁₀, PM_coarse, traffic density	LUR model, traffic and road network infrastructure	A
Beelen 2014b	ESCAPE (earliest 1985; follow-up range from 6.3 to 18.6 y)	13 countries	General	NO₂, NO_x, PM_2.5, PM_2.5 abs, PM₁₀, PM_coarse, traffic density	LUR model, traffic and road network infrastructure	C
Dimakopoulou 2014	ESCAPE (earliest 1985; follow-up range from 6.3 to 18.6 y)	11 countries	General	NO₂, NO_x, PM_2.5, PM_2.5 abs, PM₁₀, PM_coarse, traffic density	LUR model, traffic and road network infrastructure	R
Goeminne 2014	(2006–2013)	Leuven, Belgium	Patient	Traffic proximity, traffic density	Traffic and road network infrastructure	A
Stockfelt 2015	Multifactor Primary Prevention Study (1973–2007)	Gothenburg, Sweden	General	NO_x	Dispersion model	A, C, R
Desikan 2016	(2005–2012)	London, UK	Patient	NO, NO₂, NO_x, PM_2.5, PM₁₀, PM_coarse	Dispersion model	A
Tonne 2016	(2003–2010)	Greater London, UK	Patient	NO₂, NO_x, PM_2.5 (exhaust and non exhaust), PM₁₀ (exhaust and non exhaust), traffic density	Dispersion model, traffic and road network infrastructure	A
Badaloni 2017	Rome Longitudinal Study (2001–2010)	Rome, Italy	General	PM_2.5, PM_2.5 abs, PM₁₀	LUR model	A, C
Ruttens 2017	(1987–2013)	10 countries	Patient	Traffic proximity	Traffic and road network infrastructure	A
Nieuwenhuijsen 2018	Sistema d'Informació pel Desenvolupament de la Investigació en Atenció Primària (SIDIAP) (2010–2014)	Barcelona, Spain	General	NO₂, PM_2.5, PM_2.5 abs, PM₁₀	LUR model	A
Bauleo 2019	(1996–2013)	Civitavecchia, Italy	General	NO_x	Dispersion model	A, C, R, O
Andersson 2020	Primary Prevention Study cohort (1970–2011)	Gothenburg, Sweden	General	NO_x	Dispersion model	A, C
So 2020	Danish Nurse Cohort (1993–2013)	Denmark (capital region only)	General	PM_2.5	Hybrid or other models	A
Carlsen 2022	Malmo Diet and Cancer cohort (1991–1996)	Malmo, Sweden	General	NO_x, PM_2.5, PM₁₀, BC	Dispersion model	C
Asia
Yorifuji 2010	Shizuoka elderly cohort (1999–2006)	Shizuoka, Japan	General	NO₂	LUR model	A, C, R, O
Yorifuji 2013	Shizuoka elderly cohort (1999–2009)	Shizuoka, Japan	General	NO₂	LUR model	A, C, R
Barratt 2018	Hong Kong Elderly Health Services (1998–2011)	Hong Kong, China	General	NO, NO₂, PM_2.5, BC	LUR model	A, C, R
Yang 2018	Hong Kong Elderly Health Services (2001–2011)	Hong Kong, China	General	NO₂, PM_2.5, BC	LUR model	A, C, R
Cohen 2019	(2004–2017)	Israel	Patient	NO_x	LUR model, hybrid or other models, dispersion model	A, O
Cohen 2021	(1992–2018)	Israel	General and patient	NO_x	LUR model	A
Pan 2021	Risk Evaluation of Viral Load Elevation and Associated Liver Disease/ Cancer Hepatitis B Virus (REVEAL-HBV) (2000–2014)	Taipei and Pingtung, Taiwan	General	Traffic density	Traffic and road network infrastructure	C
Hadley 2022	Golestan Cohort Study (2004–2008)	Golestan province, Iran	General	Traffic proximity	Traffic and road network infrastructure	A, C
Australia
Dirgawati 2019	Health in Men Study (1996–2012)	Perth, Western Australia	General	NO₂, NO_x, PM_2.5, PM_2.5 abs	LUR model	A, C
Hanigan 2019	45 and Up Study (2006–2015)	Sydney	General	NO₂, PM_2.5	Hybrid or other models	A
Abbreviations: LUR: land-use regression Footnote a Cohort or study name when specified in the article. Return to Footnote a referrer Causes of death were divided into the following categories: all-cause (A), circulatory (C), respiratory (R), and other (O). Return to Footnote b referrer

As indicated in section 2.2.2 of this risk assessment, an RoB assessment was conducted to evaluate the validity of the included studies and establish transparency in the evidence synthesis of the results. The RoB ratings (where a higher rating signifies a greater risk of a bias in that domain) for the individual studies are summarized in Table 3.3. Outcome assessment and selective outcome reporting were the most uniformly low RoB domains. Risk of conflict of interest was rated low or probably low for all studies. Completeness of outcome data was also rated low or probably low with the exception of three studies with a probably high rating, which was largely attributed to shorter lengths of follow-up (< 4 years or median follow-up of 5.5 years). Similarly, risk of selection bias was mostly rated low or probably low with the exception of four studies that had a probably high rating. This probably high rating was attributed to a low response rate, recruitment, and retention of selective participants (i.e., self-selection for enrolment of a preventive service), loss to follow-up specific to certain demographics (e.g., older, current smokers, diabetes, financially non-capable, and more polluted area), or inconsistencies in the selection process across the cohorts. RoB from exposure assessment was rated probably low for most studies, but four studies rated probably high or high in this domain. The latter was largely attributable to a lack of data on residential mobility or time-activity patterns and issues with spatial and/or temporal accuracy of exposure estimates. Among the domains evaluated, the confounding domain showed the most variability between low, probably low, and probably high RoB. For the confounding domain, a probably high RoB rating was mostly from lack of account for smoking, and one study did not account for SES. One study was rated high for RoB in the confounding domain as it provided only unadjusted REs.

Table 3.3. RoB assessment
Reference	Selection bias	Exposure assessment	Confounding	Outcome assessment	Completeness of outcome data	Selective outcome reporting	Conflict of interest	Other
Hoek et al., 2002	2	3	1	1	2	1	1	1
Finkelstein et al., 2004	2	2	3	1	2	1	2	1
Nafstad et al., 2004	2	1	1	1	1	1	1	1
Finkelstein et al., 2005	2	4	3	1	2	1	1	1
Jerrett et al., 2005	2	2	1	2	1	1	2	1
Gehring et al., 2006	2	2	1	1	1	1	2	1
Rosenlund et al., 2006^{Footnote a}	2	2	1	1	1	1	2	1
Naess et al., 2007	1	2	3	1	1	1	1	1
Beelen et al., 2008	1	2	2 for full cohort 1 for case-cohort	1	1	1	1	1
Medina-Ramón et al., 2008	1	2	1	1	1	1	1	1
Rosenlund et al., 2008	1	2	3	1	1	1	2	1
Beelen et al., 2009	2	3	1	1	1	1	1	1
Brunekreef et al., 2009	2	1	1	1	1	1	1	1
Jerrett et al., 2009	1	1	1	1	2	1	1	1
Krewski et al., 2009 (LA and NYC)	2	2	1	1	2	1	1	1
Rosenlund et al., 2009^{Footnote a}	2	2	3	1	1	1	2	1
von Klot et al., 2009	1	2	2	1	1	1	2	1
Gan et al., 2010	1	1	3	1	1	1	2	1
Huss et al., 2010	1	2	3	1	1	1	1	1
Maheswaran et al., 2010	2	1	1	1	1	1	1	1
Yorifuji et al., 2010	3	2	1	1	1	1	1	1
Gan et al., 2011	1	1	3	1	3	1	1	1
Nawrot et al., 2011	2	2	1	2	2	1	1	1
Cesaroni et al., 2012	1	2	3	1	1	1	1	1
Raaschou-Nielsen et al., 2012	2	1	1	1	1	1	1	1
Carey et al., 2013	1	2	1	1	1	1	1	1
Cesaroni et al., 2013	1	1	3	1	1	1	1	1
Chen et al., 2013	1	1	1	1	1	1	1	1
Gan et al., 2013	1	1	3	1	2	1	2	1
Hart et al., 2013	2	2	1	1	1	1	1	1
Heinrich et al., 2013	2	2	1	1	1	1	1	1
Villeneuve et al., 2013	1	2	3 for NO₂ 1 for benzene	1	1	1	1	1
Wilker et al., 2013	1	2	1	1	1	1	1	1
Yorifuji et al., 2013	3	1	1	1	1	1	1	1
Beelen et al., 2014a	2	2	1	1	2	1	1	1
Beelen et al., 2014b	2	2	1	1	2	1	1	1
Dimakopoulou et al., 2014	2	2	1	1	1	1	1	1
Goeminne et al., 2014	1	2	1	1	3	1	1	1
Hart et al., 2014	2	1	1	1	1	1	1	1
Crouse et al., 2015	1	1	1	1	1	1	1	1
Stockfelt et al., 2015	1	1	1	1	1	1	1	1
Desikan et al., 2016	1	2	3	1	1	1	1	1
Tonne et al., 2016	2	2	1	1	1	1	2	1
Badaloni et al., 2017	1	2	1	1	1	1	2	1
Blount et al., 2017	1	2	3	1	1	1	1	1
Ruttens et al., 2017	1	2	2	2	3	1	1	2
Alexeeff et al., 2018	1	2	1	1	2	1	2	1
Barratt et al., 2018	3	2	1	1	1	1	1	1
Kulick et al., 2018	1	2	1	2	1	1	1	1
Nieuwenhuisjen et al., 2018	1	1	2	1	1	1	1	1
Yang et al., 2018	2	1	1	1	1	1	1	1
Bauleo et al., 2019	1	3	3	1	1	1	1	1
Cakmak et al., 2019	1	2	1	1	1	1	2	1
Cohen et al., 2019	2	2	1	1	2	1	1	1
Dirgawati et al., 2019	1	1	1	1	1	1	1	1
DuPré et al., 2019	2	1	1	2	1	1	1	1
Hanigan et al., 2019	2	2	1	1	1	1	1	1
Andersson et al., 2020	2	2	1	1	1	1	1	1
Cohen et al., 2021	3	2	1	1	1	1	1	1
Pan et al., 2021	2	2	3	1	1	1	1	1
So et al., 2020 (capital region only)	2	1	1	1	1	1	1	1
Villanueva et al., 2021	1	2	4	1	1	1	1	1
Carlsen et al., 2022	1	1	1	1	1	1	1	1
Hadley et al., 2022	2	2	1	1	1	1	1	1
Legend: 1: low RoB; 2: probably low RoB; 3: probably high RoB; 4: high RoB Footnote a Case-control study design; all other studies are cohorts. Return to Footnote a referrer

A literature review and evaluation of the studies are presented in the sections below for all-cause mortality, circulatory mortality, and respiratory mortality. The study details, organized by exposure metric, are provided in Appendix A. Additional analyses, including subgroup, sensitivity, leave-one-out, and publication bias analyses, are provided in section 3 of the supporting documentation. Other causes of mortality were not further considered in this risk assessment due to the limited number of studies identified.

3.2 All-cause mortality

All-cause mortality (also referred to as natural cause, non-accidental cause, and total mortality) was evaluated in 47 of the 64 articles evaluating the association between long-term exposure to TRAP and mortality.

3.2.1 Nitrogen oxides (NO, NO₂, and NO_x)

NO₂ is the most frequently measured TRAP pollutant for all-cause mortality with 19 studies while NO was among the least measured TRAP pollutants with only 2 studies. Ten studies used NO_x to assess TRAP exposure.

NO₂

For studies of NO₂, the study details regarding study population, exposure assessment, confounders, and REs are provided in Appendix A.1. Most of the studies were conducted in Europe (n = 10); for North America, 3 studies were conducted in Canada while no studies conducted in the USA were identified in the literature searches. The majority of the studies (n = 14) were drawn from the general population with a cohort size ranging from 11,627 (Dirgawati et al., 2019) to 1,265,058 (Cesaroni et al., 2013). For exposure assessment of the general population cohorts, LUR modelling (n = 11) was the most common method used, followed by hybrid modelling (n = 2) and dispersion modelling (n = 1). Exposures were assigned based on the residential address, postal code, or census tract and the mean (or median) exposures ranged from 5.2 to 104 µg/m³. The lowest exposures were assigned to one of the Swedish cohorts included in the ESCAPE study (Beelen et al., 2014a) while the highest exposures were noted in a study conducted in Hong Kong, China (Yang et al., 2018). All 14 studies accounted for age and sex, as well as individual SES and/or area-level SES; 2 studies did not adjust for smoking (Cesaroni et al., 2012, 2013). Many studies also considered additional confounders, which are listed in Appendix A.1.

The most appropriately adjusted RE identified from each study pertaining to all-cause mortality is also specified in Appendix A.1. For the general population, 12 of the 14 REs were positive, of which 4 were statistically significant and 3 were borderline significant. The forest plot and results of the random-effects meta-analysis for all-cause mortality and NO₂ in the general population are presented in Figure 3.1. Ten studies were included in the meta-analysis with REs ranging from 1.00 to 1.12. The pooled RE for the association between NO₂ and all-cause mortality was significantly positive with a HR of 1.03 (95% CI: 1.01–1.05) and heterogeneity was substantial (I² = 89.62%).

Figure 3.1 depicts a forest plot and results of the random-effects meta-analysis for all-cause mortality and NO2 exposure in the general population. — **Figure 3.1. Association between all-cause mortality and long-term exposure to NO₂ in the general population**

Figure 3.1: Text description

Figure 3.1 depicts a forest plot and results of the random-effects meta-analysis for all-cause mortality and NO₂ exposure in the general population. The meta-analysis of ten individual studies resulted in a pooled RE of 1.03 (95% CI: 1.01–1.05) per 10 µg/m³ increase in NO₂. The following results were also reported for the statistical model: Q = 52.47, df = 9, p <0.001, and I² = 89.62%.

The following information on the individual studies included in the pooled analysis is included in this figure:

Study reference	Weight of study	Hazard ratio	95% CI
Dirgawati et al., 2019	5.04%	1.06	1.00–1.13
Hanigan et al., 2019	3.01%	1.06	0.97–1.16
Barratt et al., 2018	12.96%	1.00	0.99–1.01
Nieuwenhuijsen et al., 2018	11.65%	1.02	1.00–1.04
Crouse et al., 2015	12.44%	1.05	1.04–1.07
Beelen et al., 2014a	11.65%	1.01	0.99–1.03
Carey et al., 2013	11.91%	1.02	1.00–1.04
Cesaroni et al., 2013	13.61%	1.03	1.02–1.04
Villeneuve et al., 2013	11.22%	1.01	0.99–1.03
Yorifuji et al., 2013	6.52%	1.12	1.07–1.18

Heterogeneity remained substantial (I² = 91.63%) in subgroup analysis based on the exposure assessment method when it was limited to studies that used LUR / hybrid models (RE: 1.03; 95% CI: 1.01–1.06), which included all but one study from the main analysis. Sensitivity analysis based on RoB was not done for the exposure domain as all the studies included in the meta-analysis scored low or probably low in this domain. When the analysis was limited to only studies that were low and probably low RoB for the confounder domain, the pooled estimate was robust with a RE of 1.04 (95% CI: 1.01–1.07; I² = 88.60%). Similarly, leave-one-out analysis indicated the results were robust with pooled RE ranging from 1.02 to 1.03. There was evidence of publication bias as demonstrated by the asymmetric distribution of the funnel plot but this was not supported by either Egger's (p = 0.0898) or Begg's test (p = 0.7275). Additionally, the trim-and-fill method suggested that two hypothetical studies are needed to make the funnel plot symmetrical.

As indicated in Appendix A.1, five studies examining the association between all-cause mortality and NO₂ focused exclusively on various patient populations, including patients from a respiratory disease clinic in Toronto, Ontario (Jerrett et al., 2009); patients surviving a coronary event in Rome, Italy (Rosenlund et al., 2008); and stroke (Maheswaran et al., 2010; Desikan et al., 2016) and myocardial infarction (MI) (Tonne et al., 2016) patients in London, United Kingdom. The cohort size of these studies ranged from 1,800 (Desikan et al., 2016) to 18,138 (Tonne et al., 2016). The studies used LUR models (n = 2), dispersion modelling (n = 2), and hybrid models (n = 1) to assess exposure and the majority (n = 3) accounted for the key confounders of age, sex, SES, and smoking. As summarized in Appendix A.1, three of the five REs were positive, of which one was statistically significant and one was borderline significant. The REs from the four studies in patients with a cardiovascular etiology (Rosenlund et al., 2008; Maheswaran et al., 2010; Desikan et al., 2016; Tonne et al., 2016) were pooled using a random-effects meta-analysis; the forest plot and results of this meta-analysis are presented in the supporting document. The pooled RE was 1.05 (95% CI: 0.84–1.30) per 10 µg/m³ NO₂ exposure and heterogeneity was considered substantial (I² = 85.17%).

NO_x

For NO_x, 10 studies conducted in Europe (n = 7), Israel (n = 2), and Australia (n = 1) examined the association between long-term exposure to NO_x and all-cause mortality; the study details are depicted in Appendix A.2. Six of these studies were drawn from the general population and these cohorts ranged in size from 6,304 (Andersson et al., 2020) to 367,251 (Beelen et al., 2014a). An additional study with both population-based and patient-based cohorts of CHD also reported a RE for the general population (n = 2,393) (Cohen et al., 2021). For the general population cohorts, the majority of the studies used dispersion modelling (n = 4) followed by LUR modelling (n = 3) to assess NO_x exposure and all studies assigned NO_x exposure at the residential address; the mean (or median) exposures ranged from 5.80 to 107.3 µg/m³. The lowest exposures were estimated in a study conducted in Civitavecchia, Italy (Bauleo et al., 2019) while the highest exposures were modelled in one of the Italian cohorts included in the ESCAPE study (Beelen et al., 2014a). All studies accounted for the key confounders of age, sex, and SES (individual and/or area-level) and all but one study (Bauleo et al., 2019) also adjusted for smoking. Additional confounders were also considered in these studies.

The most appropriately adjusted RE for NO_x and all-cause mortality from each study is summarized in Appendix A.2. Six of the 10 REs were positive, of which 2 were statistically significant and 2 were borderline significant. The forest plot and results of the random-effects meta-analysis for all-cause mortality and NO_x in the general population are presented in Figure 3.2. For the general population, six studies, with the exception of Andersson et al. (2020) which provided HRs by exposure quintile, were included in the meta-analysis with REs ranging from 0.98 to 1.08 per 10 µg/m³ NO_x. The pooled RE for the association between NO_x and all-cause mortality in the general population was positive and borderline significant with a HR of 1.02 (95% CI: 0.99–1.06; I² = 88.02%). Heterogeneity was substantial and was largely attributed to Nafstad et al. (2004) as no heterogeneity (I² = 0.00%) was observed when this study was removed from the meta-analysis in the leave-one-out analysis.

Figure 3.2: Text description

Figure 3.2 depicts a forest plot and results of the random-effects meta-analysis for all-cause mortality and NO_x exposure in the general population. The meta-analysis of six individual studies resulted in a pooled RE of 1.02 (95% CI: 0.99–1.06) per 10 µg/m³ increase in NO_x. The following results were also reported for the statistical model: Q = 29.7, df = 5, p <0.001, and I² = 88.02%.

The following information on the individual studies included in the pooled analysis is included in this figure:

Study reference	Weight of study	Hazard ratio	95% CI
Cohen et al., 2021 (CHD-free)	11.72%	1.00	0.95–1.05
Bauleo et al., 2019	10.11%	0.98	0.93–1.04
Dirgawati et al., 2019	19.00%	1.02	1.00–1.04
Stockfelt et al., 2015	20.09%	1.02	1.01–1.04
Beelen et al., 2014a	20.93%	1.01	1.00–1.02
Nafstad et al., 2004	18.15%	1.08	1.06–1.11

Subgroup analysis based on exposure assessment methodology indicated associations similar to the main analysis when the analysis was limited to studies using LUR modelling and dispersion modelling with pooled REs of 1.01 (95% CI: 1.00–1.03; I² = 0.00%) and 1.03 (95% CI: 0.92-1.16; I² = 91.34%), respectively. No heterogeneity was observed for pooled studies using LUR modelling to assess NO_x exposure; in contrast, heterogeneity was greater than the main analysis when limited to studies using dispersion modelling. Sensitivity analysis with respect to RoB showed robust associations between NO_x and all-cause mortality when only studies with low or probably low RoB for the exposure domain or confounder domain were included; the same study (Bauleo et al., 2019) was excluded for each of these analyses, resulting in a pooled RE of 1.03 (95% CI: 0.99–1.07; I² = 89.75%). Leave-one-out analysis indicated that the main analysis was robust as the pooled RE ranged from 1.01 to 1.03 and that Nafstad et al. (2004) was the main source of heterogeneity as omitting this study removed heterogeneity from the main analysis (I² = 0.00%). Publication bias was assessed but may be of limited use as fewer than 10 studies were included in the meta-analysis. The asymmetric distribution of the funnel plot indicated some evidence of a publication bias but this was not supported by either Egger's (p = 0.4660) or Begg's (p = 1.0000) tests. Additionally, the trim-and-fill method suggested that two hypothetical studies are needed to make the funnel plot symmetrical.

Four studies examined the association between long-term exposure to NO_x and all-cause mortality in patient populations (Appendix A.2). Two of these studies were conducted in stroke (Desikan et al., 2016) and MI (Tonne et al., 2016) patients in London, United Kingdom. The other two studies were conducted in Israel in patients undergoing percutaneous coronary interventions (Cohen et al., 2019) and in CHD patient cohorts that were matched to population-based cohorts (Cohen et al., 2021). The cohort size of these studies ranged from 1,800 (Desikan et al., 2016) to 18,138 (Tonne et al., 2016). For the exposure assessment, Desikan et al. (2016) and Tonne et al. (2016) used dispersion modelling and Cohen et al. (2021) used LUR modelling, while Cohen et al. (2019) used three different methods (i.e., dispersion modelling, LUR modelling, and hybrid modelling) to estimate exposure to NO_x. All studies accounted for the key confounders of age, sex, SES, and smoking, with the exception of Desikan et al. (2016) whose study did not adjust for smoking. As summarized in Appendix A.2, four of the six REs pertaining to patient populations were positive of which one was statistically significant. As only three of these REs could be pooled, meta-analysis was not conducted.

NO

The literature searches identified two studies evaluating the association between all-cause mortality and NO; both were based on the general population. The study details are depicted in Appendix A.3. One study was conducted on a cohort of 1,800 participants in London, United Kingdom (Desikan et al., 2016) and the other on a cohort of 60,548 participants residing in Hong Kong, China (Barratt et al., 2018). Both studies estimated exposure at the residential address; one used dispersion modelling (mean NO = 34.39 µg/m³; Desikan et al., 2016) while the other used an LUR model (mean NO = 489 µg/m³; Barratt et al., 2018). Both studies also included adjustments for the key confounders of age, sex, and SES, as well as additional confounders, but Desikan et al. (2016) did not adjust for smoking. As indicated in Appendix A.3, neither RE indicated an association.

3.2.2 PM

PM_2.5

The literature searches identified 15 studies evaluating the association between all-cause mortality and PM_2.5 that was indicative of TRAP exposure (section 2.1.2). Study details including study population, exposure assessment, confounders, and REs are provided in Appendix A.4. A majority of the studies were conducted in Europe (n = 9); no studies conducted in Canada were identified in the literature searches. Most of the studies were conducted in general population cohorts (n = 13) and these cohorts ranged in size from 11,627 (Dirgawati et al., 2019) to 1,265,058 (Cesaroni et al., 2013). For exposure assessment, the most common methods were LUR (n = 7) and dispersion modelling (n = 6), and exposures were assigned based on the residential address, postal code, or census tract. The lowest mean or median exposures were estimated for a study conducted in Australia (4.49 µg/m³; Hanigan et al., 2019) and the highest exposure was estimated for a study conducted in Hong Kong, China (42.4 µg/m³; Barratt et al., 2018). Each of the studies included adjustments for the key confounders of age, sex, and SES, and typically included additional confounders. With respect to smoking, three studies did not include adjustment, either direct or indirect, for individual-level smoking (Naess et al., 2007; Cesaroni et al., 2013; Desikan et al., 2016).

Summarized in Appendix A.4 is the most appropriately adjusted RE identified from each study. For the general population-based studies, 11 of 13 REs were positive, and 7 were statistically significant and 4 were borderline significant. The forest plot and random-effects meta-analysis for all-cause mortality and PM_2.5 are presented in Figure 3.3. Ten studies were included in the meta-analysis with standardized REs ranging from 0.90 to 1.63. The pooled RE for the association between PM_2.5 and all-cause mortality was significant at 1.06 (95% CI: 1.03–1.08) and the heterogeneity was minimal and likely not of importance (I² = 0.14%).

Figure 3.3: Text description

Figure 3.3 depicts a forest plot and results of the random-effects meta-analysis for all-cause mortality and PM_2.5 exposure in the general population. The meta-analysis of ten individual studies resulted in a pooled RE of 1.06 (95% CI: 1.03–1.08) per 10 µg/m³ increase in PM_2.5. The following results were also reported for the statistical model: Q = 10.52, df = 9, p = 0.31, and I² = 0.14%.

The following information on the individual studies included in the pooled analysis is included in this figure:

Study reference	Weight of study	Hazard ratio	95% CI
So et al., 2020	1.57%	1.02	0.86–1.21
Dirgawati et al., 2019	1.57%	1.14	0.97–1.36
Hanigan et al., 2019	0.10%	1.63	0.84–3.18
Barratt et al., 2018	22.98%	1.06	1.01–1.10
Nieuwenhuijsen et al., 2018	9.52%	1.06	0.99–1.14
Badaloni et al., 2017	50.95%	1.05	1.02–1.08
Beelen et al., 2014a	4.24%	1.14	1.03–1.27
Carey et al., 2013	4.02%	1.00	0.90–1.11
Krewski et al., 2009 (NYC)	0.75%	0.90	0.70–1.15
Krewski et al., 2009 (LA)	4.30%	1.14	1.03–1.27

As the primary studies were evaluated as low or probably low RoB for the exposure and confounding domains, a sensitivity analysis based on RoB was not performed. Subgroup analysis based on the exposure assessment method (i.e., LUR modelling or dispersion modelling) did not indicate a difference from the main analysis. Leave-one-out analysis indicated that the main analysis was robust (pooled RE varied between 1.05 and 1.07) and no one study distorted the overall results. Neither test for publication bias indicated the presence of a bias (Egger's test p = 0.359; Begg's test p = 1.000). Also, trim-and-fill analysis indicated that one hypothetical study would be required to make the funnel plot symmetrical.

Two studies based on patient populations were identified in the literature search. Both studies were conducted in London, United Kingdom; one was based on stroke patients (N = 1800) (Desikan et al., 2016) and one on MI patients (N = 18,138) (Tonne et al., 2016). Both studies used dispersion modelling to estimate PM_2.5 exposure. Tonne et al. (2016) considered PM_2.5 from traffic exhaust sources and PM_2.5 from traffic non-exhaust sources (i.e., brake dust, tire wear, and dust resuspension). Each of the three REs were positive and significant or borderline significant (Appendix A.4), and the REs were similar for both the traffic and non-traffic exhaust sources. Desikan et al. (2016) reported a stronger, though less precise, association for the stroke patients (HR = 1.28, 95% CI: 1.08–1.52, per 1.86 µg/m³) compared with the associations reported in the general population studies.

EC, PM_2.5 abs, and BC

Seven studies that evaluated exposure to EC, PM_2.5 abs, or BC and the risk of mortality were identified; study details are provided in Appendix A.5. Given the similar interpretation of the TRAP-surrogates, BC and PM_2.5 abs were converted to EC-equivalent estimates to support pooling of the studies in a meta-analysis (as described in section 2.2.3). The studies were conducted in Europe (n = 3), Asia (n = 2), the USA (n = 1), and Australia (n = 1); the literature searches did not identify any studies conducted in Canada. Six of the studies considered the general population while one considered a patient population. For the general population studies, study size ranged from 11,627 (Dirgawati et al., 2019) to 1,249,108 (Badaloni et al., 2017). Each of these studies utilized LUR modelling to estimate exposure at residential address, and, based on EC-equivalents, the mean or median exposures ranged from 0.55 (Beelen et al., 2014a) to 16.4 µg/m³ (Barratt et al., 2018). Adjustments for the key confounders of age, sex, and SES were accounted for in each of the studies; a direct or indirect adjustment for smoking was accounted for in each study, except for Badaloni et al. (2017).

Summarized in Appendix A.5 is the most appropriately adjusted RE identified from each study. For the general population, all six REs were positive; two were significant and the other four were borderline significant. The forest plot and random-effects meta-analysis for all-cause mortality and EC are presented in Figure 3.4. Five studies were included in the meta-analysis with standardized REs ranging from 1.00 to 1.11. The pooled RE for the association between EC exposure and all-cause mortality was 1.02 (95% CI: 0.99–1.05) with substantial heterogeneity (I² = 86.03%).

Figure 3.4: Text description

Figure 3.4 depicts a forest plot and results of the random-effects meta-analysis for all-cause mortality and EC exposure in the general population. The meta-analysis of five individual studies resulted in a pooled RE of 1.02 (95% CI: 0.99–1.05) per 1 µg/m³ increase in EC. The following results were also reported for the statistical model: Q = 42.97, df = 4, p <0.001, and I² = 86.03%.

The following information on the individual studies included in the pooled analysis is included in this figure:

Study reference	Weight of study	Hazard ratio	95% CI
Dirgawati et al., 2019	3.74%	1.11	1.02–1.20
Barratt et al., 2018	33.89%	1.00	1.00–1.00
Nieuwenhuijsen et al., 2018	21.27%	1.02	1.00–1.04
Badaloni et al., 2017	31.17%	1.03	1.02–1.04
Beelen et al., 2014a	9.92%	1.02	0.97–1.06

As the primary studies were evaluated as low or probably low RoB for the exposure and confounding domains, a sensitivity analysis based on RoB was not performed. Similarly, as each of the studies used LUR models, no subgroup analysis based on exposure assessment was performed. Leave-one-out analysis indicated the results were robust as the pooled RE ranged from 1.01 to 1.03. There was no indication of publication bias, as trim-and-fill analysis did not identify any missing studies and the tests did not indicate a bias (Egger's test p = 0.1895; Begg's test p = 0.8167).

The one patient population study was conducted in the USA based on adults hospitalized with acute MI (von Klot et al., 2009; summarized in Appendix A.5). The study identified a significant positive association between all-cause mortality and EC exposure, and included adjustments for age, sex, SES, and smoking. As only the single study was identified, a meta-analysis was not conducted.

PM₁₀ and PM_coarse

The literature searches identified seven studies of PM₁₀ and two studies of PM_coarse that evaluated an association with all-cause mortality. As the literature available for PM_coarse is limited, this assessment will focus on the PM₁₀ literature. Study details including study population, exposure assessment, confounders, and REs are provided in Appendix A.6. Each of the studies was conducted in Europe, with four based on the general population and three based on patient populations. No studies conducted in Canada were identified in the literature searches. For the general population studies, the sample sizes were large, ranging from 322,159 (Beelen et al., 2014a) to 1,249,108 (Badaloni et al., 2017). For the exposure assessment, three of the studies employed LUR models and one utilized dispersion modelling, and exposures were assigned based on the residential address, postal code, or census tract. For PM₁₀, the mean exposures ranged from 13.5 to 48.1 µg/m³ (Beelen et al., 2014a). Each of the studies adjusted for the key confounders of age, sex, SES, and directly or indirectly adjusted for smoking, except for Badaloni et al. (2017), which did not account for smoking.

Summarized in Appendix A.6 is the most appropriately adjusted RE identified from each study. For the general population, two of four REs for PM₁₀ were positive and significant or borderline significant, and two indicated a null effect. The forest plot and random-effects meta-analysis for all-cause mortality and PM₁₀ are presented in Figure 3.5. All of the general population studies were included in the meta-analysis; the standardized REs ranged from 1.00 to 1.04 and the pooled RE was 1.02 (95% CI: 1.00–1.03). No heterogeneity (I² = 0.00%) was noted for this main analysis.

Figure 3.5 depicts a forest plot and results of the random-effects meta-analysis for all-cause mortality and PM10 exposure in the general population. — **Figure 3.5. Association between all-cause mortality and long-term exposure to PM₁₀ in the general population**

Figure 3.5: Text description

Figure 3.5 depicts a forest plot and results of the random-effects meta-analysis for all-cause mortality and PM₁₀ exposure in the general population. The meta-analysis of four individual studies resulted in a pooled RE of 1.02 (95% CI: 1.00–1.03) per 10 µg/m³ increase in PM₁₀. The following results were also reported for the statistical model: Q = 2.49, df = 3, p = 0.477, and I² = 0%.

The following information on the individual studies included in the pooled analysis is included in this figure:

Study reference	Weight of study	Hazard ratio	95% CI
Nieuwenhuijsen et al., 2018	7.08%	1.00	0.97–1.03
Badaloni et al., 2017	88.05%	1.02	1.01–1.03
Beelen et al., 2014a	3.43%	1.04	1.00–1.09
Carey et al., 2013	1.43%	1.00	0.94–1.07

As each of the primary studies in the main analysis was evaluated as low or probably low RoB for the exposure and confounding domains, a sensitivity analysis based on RoB was not performed. Subgroup analysis based on the exposure assessment method (i.e., LUR modelling or dispersion modelling) did not indicate a difference from the main analysis. Leave-one-out analysis did not indicate that any one study was influencing the main analysis (pooled RE = 1.01–1.02). There was no indication of publication bias, as trim-and-fill analysis did not identify any missing studies and the tests did not indicate a bias (Egger's test p = 0.8851; Begg's test p = 1.000).

Three studies based on patient populations were identified. Each of the studies was conducted in London, United Kingdom. Two considered stroke patients and one considered MI patients (summarized in Appendix A.6). The cohorts ranged in size from 1,800 (Desikan et al., 2016) to 18,128 (Tonne et al., 2016). Two of the studies used dispersion modelling and one used hybrid modelling. For PM₁₀, both REs were positive and significant or borderline significant. Tonne et al. (2016) evaluated the association of PM₁₀ from exhaust and non-exhaust sources separately, and both REs were positive and borderline significant. The mean exposures were an order of magnitude smaller in Tonne et al. (2016) compared with the other patient population studies. Due to the small number of studies, a meta-analysis was not conducted.

3.2.3 Other TRAP pollutants: benzene

Only one study that examined the association between long-term exposure to benzene, as a marker of TRAP, and all-cause mortality was identified in the literature searches. This Canadian study by Villeneuve et al. (2013) was conducted in the general population of Ontario, accounted for the key confounders including indirect adjustment for smoking, and reported a significant, positive association (summarized in Appendix A.7).

3.2.4 Traffic and the road network infrastructure

Traffic proximity

Traffic proximity or distance to roadway was the most commonly used metric of traffic and the road network infrastructure that examined the association between traffic and all-cause mortality with 18 studies, of which 11 were conducted on the general population and 7 on various patient populations. The study details regarding study population, exposure assessment, confounders, and REs are provided in Appendix A.9.

For the general population, the majority of the studies were conducted in Europe (n = 7) followed by the USA (n = 2), Canada (n = 1), and Asia (n = 1). They ranged in size from 3,287 (Kulick et al., 2018) to 2,644,370 (Cakmak et al., 2019). All the studies conducted in the USA, Canada, and Asia were based on unique cohorts while those conducted in Europe were based on four cohorts: NLCS-AIR (Hoek et al., 2002; Beelen et al., 2008; Brunekreef et al., 2009); SALIA (Gehring et al., 2006); the Diet, Cancer, and Health Cohort Study (Raaschou-Nielsen et al., 2012); and the Rome Longitudinal Study (Cesaroni et al., 2013). All studies accounted for the key confounders of age, sex, and SES (individual- and/or area-level) and all but two studies (Cesaroni et al., 2013; Cakmak et al., 2019) also adjusted for smoking. Many studies also considered additional confounders. Proximity to traffic was directly measured in a geographic information system (GIS) and was based on the exact residential address for all but two studies. The Canadian study used the residential postal code centroid (Cakmak et al., 2019) while the study conducted in Iran used the geographic centre of each village for rural residents and the location of the primary health centre for urban residents (Hadley et al., 2022). No consistent definitions or categories were used to assess traffic proximity across the studies.

The associations between traffic proximity and risk of all-cause mortality in the general population are summarized in Figure 3.6; only unique cohorts and non-transformed data are included in the forest plot. Two studies considered traffic proximity as a continuous variable. In a Canada-wide study, Cakmak et al. (2019) reported a positive and significant association. In comparison, Kulick et al. (2018) did not observe any association based on log-transformed proximity. When traffic proximity was evaluated as a categorical variable, 10 of the 14 REs were positive, of which one was significant and the remainder were borderline significant. Of the three studies that examined exposure-response relationships, Cesaroni et al. (2013) and Hart et al. (2013) identified an increased risk with increased exposure while Kulick et al. (2018) did not.

Figure 3.6: Text description

Figure 3.6 depicts a forest plot of risk estimates for long-term exposure to traffic proximity and all-cause mortality in the general population. The x-axis representing risk estimates and 95% CI ranges from 0.8 to 1.9. The following information is depicted in this figure:

Reference	Cohort, study location	Description	Data marker	Risk estimate	95% CI	Type of risk estimate
Cakmak et al., 2019	CanCHEC Cohort, Canada	total length of local roads ≤200 m radius from residence, per 1,108.6 m increase	A	1.05	1.04–1.05	Continuous (per IQR increase)
Hadley et al., 2022	Golestan Cohort Study, Asia	living ≤100 m from a minor highway or ≤500 m from a major highway	B	1.04	0.96–1.12	Categorical
Kulick et al., 2018	The Northern Manhattan Cohort, USA	distance to roadway categorized as <100 m (reference group ≥400 m)	C1	0.95	0.81–1.13	Categorical
Kulick et al., 2018	The Northern Manhattan Cohort, USA	distance to roadway categorized as 100 to <200 m (reference group ≥400 m)	C2	1.04	0.90–1.21	Categorical
Kulick et al., 2018	The Northern Manhattan Cohort, USA	distance to roadway categorized as 200 to <400 m (reference group ≥400 m)	C3	1.04	0.91–1.18	Categorical
Cesaroni et al., 2013	Rome Longitudinal Study, Europe	distance to high traffic road with >10,000 vehicles/day: <50 m (reference group ≥250 m)	D1	1.02	1.00–1.01	Categorical
Cesaroni et al., 2013	Rome Longitudinal Study, Europe	distance to high traffic road with >10,000 vehicles/day: 50–100 m (reference group ≥250 m)	D2	1.01	0.99–1.03	Categorical
Cesaroni et al., 2013	Rome Longitudinal Study, Europe	distance to high traffic road with >10,000 vehicles/day: 100–150 m (reference group ≥250 m)	D3	1.01	0.99–1.02	Categorical
Cesaroni et al., 2013	Rome Longitudinal Study, Europe	distance to high traffic road with >10,000 vehicles/day: 150–250 m (reference group ≥250 m)	D4	0.99	0.98–1.01	Categorical
Hart et al., 2013	NHS, USA	living ≤50 m from A3 road or ≤150 m from A1 or A2 road; consistently close (reference group: consistently far)	E1	1.05	1.00–1.10	Categorical
Hart et al., 2013	NHS, USA	living ≤50 m from A3 road or ≤150 m from A1 or A2 road; moved far to close (reference group: consistently far)	E2	1.17	1.00–1.37	Categorical
Hart et al., 2013	NHS, USA	living ≤50 m from A3 road or ≤150 m from A1 or A2 road; moved close to far (reference group: consistently far)	E3	0.98	0.94–1.03	Categorical
Heinrich et al., 2013	SALIA, Europe	living ≤50 m from a major roadway (≥10,000 cars/day)	F	1.42	1.12–1.79	Categorical
Raaschou–Nielsen et al., 2012	Diet, Cancer, and Health Cohort Study, Europe	living ≤50 m from a major roadway (≥10,000 vehicles/day)	G	0.94	0.85–1.05	Categorical
Brunekreef et al., 2009	NLCS-AIR Cohort, Europe	living near a major road (≤100 m from a freeway or ≤50 m from a major road with 10,000 motor vehicles/day)	H	1.05	0.97–1.12	Categorical

Traffic proximity and risk of all-cause mortality were also examined in a diverse group of patient populations (summarized in Appendix A.9 and Figure 3.7). These studies were conducted in Canada (n = 2), the USA (n = 2), and Europe (n = 3). With the exception of one study that did not account for any confounders (Finkelstein et al., 2004), all accounted for age and sex, five accounted for SES, and three accounted for smoking; many studies also accounted for additional confounders. For the majority of the studies, exposure was based on the exact residential address while two studies used postal or ZIP code to assign exposure (Finkelstein et al., 2004; Medina-Ramón et al., 2008). For the two studies that assessed exposure as a continuous variable (Medina-Ramón et al., 2008; Ruttens et al., 2017), no association was observed based on distance to a major road or freeway. Ruttens et al. (2017) identified positive associations that were significant or borderline significant when considering the total length of roads in buffer zones, with the greatest risk for a 100-m buffer zone. Goeminne et al. (2014) reported that each 10-fold increase in the distance to a major road resulted in a significant lower risk of dying (HR = 0.28; 95% CI: 0.10–0.77). When traffic proximity was evaluated as a categorical variable, positive associations that were significant or borderline significant were observed, and Wilker et al. (2013) observed an exposure-response relationship.

Figure 3.7: Text description

Figure 3.7 depicts a forest plot of risk estimates for long-term exposure to traffic proximity and all-cause mortality in patient populations. The x-axis representing risk estimates and 95% CI ranges from 0.8 to 2.1. The following information is depicted in this figure:

Reference	Cohort, study location	Description	Data marker	Risk estimate	95% CI	Type of risk estimate
Ruttens et al., 2017	Lung transplant patients across 10 European countries	distance to freeway, per 1,233 m	A1	0.987	0.964–1.012	Continuous (per IQR increase)
Ruttens et al., 2017	Lung transplant patients across 10 European countries	distance to major road, per 241 m	A2	1.000	0.976–1.024	Continuous (per IQR increase)
Ruttens et al., 2017	Lung transplant patients across 10 European countries	total road length within different buffers: 50 m, per 108 m	A3	1.055	0.955–1.112	Continuous (per IQR increase)
Ruttens et al., 2017	Lung transplant patients across 10 European countries	total road length within different buffers: 100 m, per 279 m	A4	1.111	1.025–1.202	Continuous (per IQR increase)
Ruttens et al., 2017	Lung transplant patients across 10 European countries	total road length within different buffers: 200 m, per 756 m	A5	1.094	1.030–1.779	Continuous (per IQR increase)
Ruttens et al., 2017	Lung transplant patients across 10 European countries	total road length within different buffers: 500 m, per 4,092 m	A6	1.085	1.000–1.130	Continuous (per IQR increase)
Ruttens et al., 2017	Lung transplant patients across 10 European countries	total road length within different buffers: 1000 m, per 15,403 m	A7	1.047	0.985–1.131	Continuous (per IQR increase)
Medina Ramón et al., 2008	Worcester Heart Failure Study, USA	distance to major roadway, per 2,008 m	B	1.00	0.93–1.08	Continuous (per IQR increase)
Wilker et al., 2013	Ischemic stroke patients, USA	distance to roadway with >10,000 vehicles/day; ≤100 m (reference >400 m)	C1	1.20	1.01–1.43	Categorical
Wilker et al., 2013	Ischemic stroke patients, USA	distance to roadway with >10,000 vehicles/day; >100 to 200 m (reference >400 m)	C2	1.08	0.88–1.31	Categorical
Wilker et al., 2013	Ischemic stroke patients, USA	distance to roadway with >10,000 vehicles/day; >200 to 400 m (reference >400 m)	C3	0.99	0.82–1.20	Categorical
Nawrot et al., 2011	Lung transplant patients with long-term follow-up at UZ Leuven, Belgium	living ≤171 m from a major road	D	1.99	1.09–3.61	Categorical
Jerrett et al., 2009	Patients from a respiratory disease clinic, Canada	living ≤50 m from a major road or ≤100 m from a highway	E	1.19	0.92–1.53	Categorical
Finkelstein et al., 2004	Pulmonary function testing at clinic, Canada	living ≤50 m from a major road or ≤100 m from a highway	F	1.18	1.02–1.38	Categorical

Traffic density

The literature search identified nine studies that evaluated the association between traffic density and all-cause mortality; five studies were conducted in the general population and four considered patient populations. Study details including study population, exposure assessment, confounders, and REs are provided in Appendix A.10. For the general population studies, each was conducted in Europe and considered risks in the general population. They ranged in size from 52,061 (Raaschou-Nielsen et al., 2012) to 1,265,058 (Cesaroni et al., 2013). Beelen et al. (2008) and Brunekreef et al. (2009) were conducted with the same cohort and had identical results. Each of the studies accounted for the key confounders of age, sex, and SES, and smoking was accounted for in each of the studies, except for Cesaroni et al. (2013).

The REs for traffic density and all-cause mortality in the general population are summarized in a forest plot (Figure 3.8). Both the ESCAPE study (Beelen et al., 2014a) and NLCS-AIR Cohort (Brunekreef et al., 2009) considered traffic as a continuous variable. For both these studies, positive, borderline-significant associations were observed when evaluated based on traffic on the nearest road to the residence and sum of traffic in a 100-m buffer around the residence. Cesaroni et al. (2013) evaluated quintiles of traffic density in a 150-m buffer around the residence and identified a positive association for each quintile and a significant trend increasing risk (p < 0.001).

Figure 3.8: Text description

Figure 3.8 depicts a forest plot of risk estimates for long-term exposure to traffic density and all-cause mortality in the general population. The x-axis representing risk estimates and 95% CI ranges from 0.95 to 1.1. The following information is depicted in this figure:

Traffic indicator	Reference	Cohort, study location	Description	Data marker	Risk estimate	95% CI	Type of risk estimate
Traffic on nearest road	Beelen et al., 2014a	ESCAPE, Europe	traffic on nearest road (per 5,000 vehicles/day)	A1	1.01	1.00–1.03	Continuous
Traffic on nearest road	Brunekreef et al., 2009	NLCS-AIR Cohort, Europe	traffic on nearest road (per 10,000 vehicles/day)	B1	1.03	1.00–1.08	Continuous
Sum of traffic in 100-m buffer	Beelen et al., 2014a	ESCAPE, Europe	sum of traffic intensity in 100-m buffer (per 4x 10⁶ vehicle-m/day)	A2	1.01	0.98–1.05	Continuous
Sum of traffic in 100-m buffer	Brunekreef et al., 2009	NLCS-AIR Cohort, Europe	sum of traffic intensity in 100-m buffer (per 335,000 vehicles/day)	B2	1.02	0.97–1.07	Continuous
Traffic density in 150-m buffer	Cesaroni et al., 2013	Rome Longitudinal Study, Europe	traffic density in 150-m buffer (vehicle-m/day); (0.25–1.63) x 10⁶ (reference: <0.25 x 10⁶)	C1	1.02	1.00–1.04	Categorical
Traffic density in 150-m buffer	Cesaroni et al., 2013	Rome Longitudinal Study, Europe	traffic density in 150-m buffer (vehicle-m/day); (1.63–3.23) x 10⁶ (reference: <0.25 x 10⁶)	C2	1.03	1.01–1.05	Categorical
Traffic density in 150-m buffer	Cesaroni et al., 2013	Rome Longitudinal Study, Europe	traffic density in 150-m buffer (vehicle-m/day); (3.23–6.66) x 10⁶ (reference: <0.25 x 10⁶)	C3	1.03	1.01–1.05	Categorical
Traffic density in 150-m buffer	Cesaroni et al., 2013	Rome Longitudinal Study, Europe	traffic density in 150-m buffer (vehicle-m/day); ≥6.66 x 10⁶ (reference: <0.25 x 10⁶)	C4	1.04	1.03–1.06	Categorical

Four studies conducted in patient populations were identified in the literature search (summarized in Appendix A.10). These studies, representing a diverse group of patient populations, were conducted in the USA (n = 2) and Europe (n = 2), and had a wide range of study sizes (183 [Goeminne et al., 2014] to 32,875 [Blount et al., 2017]). Each of the studies accounted for key confounders of age, sex, and SES, while smoking was only accounted for in Goeminne et al. (2014) and Tonne et al. (2016). A positive, borderline significant or significant association between traffic density and all-cause mortality was observed in each study.

3.2.5 Summary

The association between TRAP and all-cause mortality has been investigated in 47 cohort studies. The majority of these studies were based on a general population and were conducted in Europe. Over 60% of the studies assessed exposure to TRAP by modelling individual pollutants, while the remainder used metrics of traffic and the road network infrastructure as indirect measures of exposure. Nearly all the studies accounted for the key confounders of age, sex, SES, and smoking and many considered additional confounders. For the pooled analyses based on the general population, a positive association was determined for each of the pollutants evaluated (i.e., NO₂, NO_x, PM_2.5, EC, and PM₁₀), and these associations with pooled REs ranging from 1.02 to 1.06 were significant for NO₂ and PM_2.5 and borderline significant for NO_x, EC, and PM₁₀. While there was no or little heterogeneity for PM_2.5 and PM₁₀, heterogeneity was substantial for NO₂, NO_x, and EC. The results from the pooled analyses was supported by the qualitative analysis of the different metrics for traffic and the road network infrastructure, which also provided some indication of an exposure-response relationship. For patient populations, pooled analysis could only be conducted for NO₂, where a positive non-significant association was observed for patients with a cardiovascular etiology. For NO and benzene, the evidence was limited and did not support conducting a pooled analysis.

3.3 Circulatory mortality

Circulatory mortality (also referred to as cardiovascular mortality) was evaluated in 41 of the 64 articles evaluating the association between long-term exposure to TRAP and mortality. An evaluation of the studies examining the effects of long-term exposure to TRAP and circulatory mortality are presented below by TRAP pollutant or metric. These studies evaluated mortality from circulatory system diseases (CSD) as a group, as well as from more specific causes (e.g., CHD [also referred to as IHD], MI, heart failure [HF], and cerebrovascular disease [CBVD], including stroke). In the few instances that studies reported cardiopulmonary or cardiorespiratory mortality without presenting the results for circulatory and respiratory mortality separately, the studies were considered a measure of circulatory mortality, as these causes dominate the combined category (HEI, 2022). Similarly, as stroke is the most common cause of CBVD, and consistent with the approach taken by HEI (2022), CBVD mortality and stroke mortality were pooled for meta-analysis or qualitatively evaluated together in the absence of a quantitative analysis.

3.3.1 Nitrogen oxides (NO, NO₂, and NO_x)

NO₂

For NO₂ studies (n = 19), details including study population, exposure assessment, confounders, and REs are provided in Appendix A.1. Most of the studies were conducted in Europe (n = 8), Canada (n = 5), and Asia (n = 4). Two of the studies had a case-control study design (Rosenlund et al., 2006, 2009), and the rest were cohort studies. All but one Canadian study (Jerrett et al., 2009) were drawn from the general population with a study size ranging from 3,210 (Rosenlund et al., 2006) to 1,265,058 (Cesaroni et al., 2013). For exposure assessment of the studies conducted in the general population, LUR modelling (n = 12) was the most common method used, followed by dispersion modelling (n = 4) and hybrid modelling (n = 2). Exposures were assigned based on the residential address, postal code, census block, or at the neighborhood level, with the mean (or median) exposures ranging from 5.2 to 107 µg/m³. The lowest exposures were assigned to one of the Swedish cohorts included in the ESCAPE study (Beelen et al., 2014a) while the highest exposures were noted in a study conducted in Hong Kong, China (Barratt et al., 2018). Studies conducted in the general population each examined one or more endpoints pertaining to circulatory mortality, including mortality from CSD (n = 12), CHD (n = 12), CBVD (n = 10) and stroke (n = 1), MI (n = 3), and dysrhythmias, HF, and cardiac arrest (n = 2). All 18 studies accounted for age and sex, as well as individual SES and/or area-level SES; 5 studies did not adjust for smoking (Naess et al., 2007; Rosenlund et al., 2008, 2009; Gan et al., 2011; Cesaroni et al., 2013). Many studies also considered additional confounders.

The most appropriately adjusted RE identified from each study for each relevant outcome pertaining to circulatory mortality is also specified in Appendix A.1.

For the general population, 12 of the 15 REs pertaining to CSD mortality were positive, of which 7 were statistically significant and 5 were borderline significant. Four of these REs were derived from the same study (Naess et al., 2007) and correspond to different age groups and sex. The forest plot and results of the random-effects meta-analysis for CSD mortality and NO₂ are presented in Figure 3.9. Seven studies were included in the meta-analysis for the general population with REs ranging from 0.99 to 1.24. The pooled RE for the association between NO₂ and CSD mortality was borderline significant at 1.05 (95% CI: 0.98–1.12) with considerable heterogeneity (I² = 96.64%).

Figure 3.9 depicts a forest plot and results of the random-effects meta-analysis for CSD mortality and NO2 exposure in the general population. — **Figure 3.9. Association between CSD mortality and long-term exposure to NO₂ in the general population**

Figure 3.9: Text description

Figure 3.9 depicts a forest plot and results of the random-effects meta-analysis for CSD mortality and NO₂ exposure in the general population. The meta-analysis of seven individual studies resulted in a pooled RE of 1.05 (95% CI: 0.98–1.12) per 10 µg/m³ increase in NO₂. The following results were also reported for the statistical model: Q = 48.12, df = 6, p <0.001, and I² = 96.64%.

The following information on the individual studies included in the pooled analysis is included in this figure:

Study reference	Weight of study	Hazard ratio	95% CI
Barratt et al., 2018	15.20%	1.00	0.98–1.02
Crouse et al., 2015	14.95%	1.04	1.02–1.07
Beelen et al., 2014b	13.85%	1.01	0.97–1.06
Carey et al., 2013	15.04%	0.99	0.97–1.01
Cesaroni et al., 2013	15.47%	1.03	1.02–1.04
Chen et al., 2013	13.81%	1.07	1.03–1.12
Yorifuji et al., 2013	11.69%	1.24	1.15–1.33

Heterogeneity remained considerable (I² = 96.68%) when the analysis was limited to studies that used LUR models (RE: 1.06; 95% CI: 0.98–1.14), which included all but one study from the main analysis. Sensitivity analysis based on RoB was not done for the exposure domain as all the studies included in the meta-analysis scored low or probably low for this domain, but when the analysis was limited to studies with low and probably low RoB for the confounder domain, the pooled RE was robust (RE: 1.05; 95% CI: 0.97–1.14) and heterogeneity was unchanged (I² = 95.7%). Leave-one-out analysis indicated that the results were robust for all studies with the exception of Yorifuji et al. (2013); removing the latter resulted in a lower pooled RE of 1.02 and reduced heterogeneity (I² = 80.96%). Although the Egger's test indicated evidence of a publication bias (p = 0.04067), this was not supported by the Begg's test (p = 0.2389) or the trim-and-fill analysis, which did not identify any missing studies.

For CHD mortality in the general population, 11 of the 12 REs were positive, of which 7 were statistically significant and 2 were borderline significant. The forest plot and results of the random-effects meta-analysis for CHD mortality and NO₂ are presented in Figure 3.10. Nine studies were included in the meta-analysis for the general population with REs ranging from 1.00 to 1.29. The pooled RE for the association between NO₂ and CHD mortality was significant at 1.05 (1.03–1.07) and the heterogeneity was minimal and likely not of importance (I² = 0.4%).

Figure 3.10 depicts a forest plot and results of the random-effects meta-analysis for CHD mortality and NO2 exposure in the general population. — **Figure 3.10. Association between CHD and long-term exposure to NO₂ in the general population**

Figure 3.10: Text description

Figure 3.10 depicts a forest plot and results of the random-effects meta-analysis for CHD mortality and NO₂ exposure in the general population. The meta-analysis of nine individual studies resulted in a pooled RE of 1.05 (95% CI: 1.03–1.07) per 10 µg/m³ increase in NO₂. The following results were also reported for the statistical model: Q = 12.5, df = 8, p = 0.13, and I² = 0.4%.

The following information on the individual studies included in the pooled analysis is included in this figure:

Study reference	Weight of study	Hazard ratio	95% CI
Alexeeff et al., 2018	0.15%	1.13	0.85–1.50
Barratt et al., 2018	12.20%	1.03	1.00–1.07
Crouse et al., 2015	9.93%	1.05	1.02–1.09
Beelen et al., 2014b	1.53%	1.00	0.91–1.09
Cesaroni et al., 2013	58.54%	1.05	1.04–1.07
Chen et al., 2013	3.55%	1.09	1.02–1.15
Yorifuji et al., 2013	0.64%	1.29	1.12–1.48
Gan et al., 2011	7.79%	1.05	1.01–1.09
Rosenlund et al., 2008	5.67%	1.07	1.02–1.12

As all the primary studies included in the meta-analysis used LUR models to assess exposure and were evaluated as low or probably low RoB for the exposure domain, subgroup analysis by exposure assessment and sensitivity analysis based on RoB for the exposure domain were not performed. When analysis was limited to studies with low and probably low RoB for the confounder domain, which excluded three studies (i.e., Rosenlund et al., 2008, Gan et al., 2011, and Cesaroni et al., 2013), the pooled RE was robust (RE: 1.06) but there was increased uncertainty around the RE (95% CI: 0.99–1.14) and heterogeneity became substantial (I² = 61.9%). Leave-one-out analysis did not indicate that any one study was influencing the main analysis; the association remained significant with a pooled RE = 1.05. Neither test for publication bias suggested the presence of a bias (Egger's test p = 0.2371; Begg's test p = 0.1194) while trim-and-fill analysis indicated that two hypothetical studies would be required to make the funnel plot symmetrical.

For CBVD mortality (including stroke mortality) in the general population, 7 of the 11 REs were positive, of which 1 was statistically significant and 5 were borderline significant. The forest plot and results of the random-effects meta-analysis for CBVD mortality and NO₂ are presented in Figure 3.11. Eight studies were included in the meta-analysis for the general population with REs ranging from 0.93 to 1.57. The pooled estimate for the association between NO₂ and CBVD mortality was borderline significant at 1.01 (95% CI: 0.98–1.04) and the heterogeneity was minimal and likely not of importance (I² = 0.56%).

Figure 3.11 depicts a forest plot and results of the random-effects meta-analysis for CBVD mortality and NO2 exposure in the general population. — **Figure 3.11. Association between CBVD mortality and long-term exposure to NO₂ in the general population**

Figure 3.11: Text description

Figure 3.11 depicts a forest plot and results of the random-effects meta-analysis for CBVD mortality and NO₂ exposure in the general population. The meta-analysis of eight individual studies resulted in a pooled RE of 1.01 (95% CI: 0.98–1.04) per 10 µg/m³ increase in NO₂. The following results were also reported for the statistical model: Q = 12.65, df = 7, p = 0.081, and I² = 0.56%.

The following information on the individual studies included in the pooled analysis is included in this figure:

Study reference	Weight of study	Hazard ratio	95% CI
Dirgawati et al., 2019	0.39%	0.93	0.72–1.20
Alexeeff et al., 2018	0.08%	1.57	0.90–2.74
Barratt et al., 2018	20.33%	1.00	0.97–1.04
Crouse et al., 2015	9.06%	1.02	0.97–1.08
Beelen et al., 2014b	3.52%	1.01	0.93–1.10
Cesaroni et al., 2013	60.71%	1.01	0.99–1.03
Chen et al., 2013	4.10%	0.96	0.89–1.03
Yorifuji et al., 2013	1.81%	1.19	1.06–1.34

Subgroup analysis by exposure assessment and sensitivity analysis based on RoB for the exposure domain were not performed as all the primary studies included in the meta-analysis used LUR modelling or hybrid modelling to assess exposure and were evaluated as low or probably low RoB for the exposure domain. When analysis was limited to studies with low and probably low RoB for the confounder domain, which included all but Cesaroni et al. (2013), the pooled estimate was robust (RE: 1.02) but heterogeneity became moderate (I² = 50.81%). Limiting the analysis to studies exclusively evaluating CBVD mortality (i.e., excluding stroke mortality) did not change the pooled estimate (RE: 1.01) or the heterogeneity (I² = 0.54%). Similarly, leave-one-out analysis did not indicate that any one study was influencing the main analysis (pooled RE = 1.01–1.02). Neither test for publication bias indicated the presence of a bias (Egger's test p = 0.4139; Begg's test p = 0.7195) while trim-and-fill analysis indicated that one hypothetical study would be required to make the funnel plot symmetrical.

A limited number of studies (three or less) were identified in the literature searches pertaining to NO₂ exposure and MI mortality, dysrhythmias, HF, and cardiac arrest mortality in the general population; these studies reported mostly positive associations that were borderline significant or significant.

The only study to examine the association between long-term exposure to NO₂ and circulatory mortality in a patient population was conducted in adult patients from a respiratory disease clinic in Toronto, Ontario, (Jerrett et al., 2009). This Canadian study accounted for the key confounders including smoking, and reported a significant, positive association between NO₂ and CSD mortality (summarized in Appendix A.1).

NO_x

For NO_x, seven studies examined the association between long-term exposure to NO_x and circulatory causes of mortality; the study details are depicted in Appendix A.2. They were conducted in Europe (n = 6) and Australia (n = 1) and all were drawn from the general population with cohorts ranging in size from 6,304 (Andersson et al., 2020) to 367,383 (Beelen et al., 2014a). The majority of the studies used dispersion modelling (n = 5) followed by LUR modelling (n = 2) to assess NO_x exposure and all studies assigned NO_x exposure at the residential address; the mean (or median) exposures ranged from 5.80 to 107.3 µg/m³. The lowest exposures were estimated in a study conducted in Civitavecchia, Italy (Bauleo et al., 2019) while the highest exposures were modelled in one of the Italian cohorts included in the ESCAPE study (Beelen et al., 2014a). The studies each examined one or more endpoints pertaining to circulatory mortality, including mortality from CSD (n = 4), CHD (n = 4), CBVD (n = 4) and stroke (n = 1), MI (n = 3), cardiac diseases (n = 1), and acute coronary events (ACE; n = 1). All studies accounted for the key confounders of age, sex, and SES (individual- and/or area-level) and all but one study (Bauleo et al., 2019) also adjusted for smoking. Additional confounders were also considered in these studies.

The most appropriately adjusted RE from each study for each relevant outcome pertaining to circulatory mortality is summarized in Appendix A.2.

For CHD mortality, three of the four REs were positive of which two were borderline significant and one was statistically significant. The forest plot and results of the random-effects meta-analysis for CHD mortality and NO_x in the general population are presented in Figure 3.12. Four studies were included in the meta-analysis with REs ranging from 0.86 to 1.08. The pooled RE for the association between NO_x and CHD mortality was borderline significant at 1.02 (95% CI: 0.93–1.13) and the heterogeneity was substantial (I² = 77.84%).

Figure 3.12: Text description

Figure 3.12 depicts a forest plot and results of the random-effects meta-analysis for CHD mortality and NO_x exposure in the general population. The meta-analysis of four individual studies resulted in a pooled RE of 1.02 (95% CI: 0.93–1.13) per 10 µg/m³ increase in NO_x. The following results were also reported for the statistical model: Q = 11.24, df = 3, p = 0.01, and I² = 77.84%.

The following information on the individual studies included in the pooled analysis is included in this figure:

Study reference	Weight of study	Hazard ratio	95% CI
Bauleo et al., 2019	6.91%	0.86	0.74–1.01
Stockfelt et al., 2015	32.52%	1.02	0.99–1.05
Beelen et al., 2014b	31.85%	1.01	0.98–1.04
Nafstad et al., 2004	28.73%	1.08	1.03–1.12

When the analysis was limited to studies using dispersion modelling to assess exposure, which included all but one study from the main analysis, the pooled RE remained positive but was not significant (RE: 1.01; 95% CI: 0.78–1.30) and heterogeneity increased (I² = 91.16%). In contrast, the pooled RE and heterogeneity from the main analysis were robust to the sensitivity analyses based on RoB (i.e., excluding studies with high or probably high RoB in the exposure assessment or confounder domains). Leave-one-out analysis indicated the following: 1) Nafstad et al. (2004) was the main source of heterogeneity as omitting this study removed heterogeneity from the main analysis, and 2) omitting Beelen et al. (2014b) and Stockfelt et al. (2015) increased the uncertainty around the RE. There was no evidence of a publication bias as demonstrated by the Begg's (p = 0.7500) and Egger's (p = 0.2366) tests and the trim-and-fill method suggested that one hypothetical study was needed to make the funnel plot symmetrical.

For CBVD mortality (including stroke mortality), two of the five REs were positive and borderline significant. The forest plot and results of the random-effects meta-analysis for CBVD mortality and NO_x in the general population are presented in Figure 3.13. Five studies were included in the meta-analysis with REs ranging from 0.82 to 1.04. The pooled analysis indicated a null association (RE: 1.00; 95% CI: 0.96–1.05) between NO_x and CBVD mortality with minimal heterogeneity that was likely not of importance (I² = 0.10%).

Figure 3.13: Text description

Figure 3.13 depicts a forest plot and results of the random-effects meta-analysis for CBVD mortality and NO_x exposure in the general population. The meta-analysis of five individual studies resulted in a pooled RE of 1.00 (95% CI: 0.96–1.05) per 10 µg/m³ increase in NO_x. The following results were also reported for the statistical model: Q = 5.94, df = 4, p = 0.204, and I² = 0.1%.

The following information on the individual studies included in the pooled analysis is included in this figure:

Study reference	Weight of study	Hazard ratio	95% CI
Bauleo et al., 2019	2.25%	0.82	0.68–1.00
Dirgawati et al., 2019	8.46%	0.97	0.88–1.07
Stockfelt et al., 2015	23.73%	1.03	0.97–1.09
Beelen et al., 2014b	57.62%	1.00	0.96–1.04
Nafstad et al., 2004	7.95%	1.04	0.94–1.15

Except for the higher heterogeneity observed for dispersion modelling (I² = 70.44%), subgroup analysis based on the exposure assessment method (i.e., LUR modelling or dispersion modelling) did not indicate a difference from the main analysis. In contrast, sensitivity analyses based on RoB (i.e., excluding studies with high and probably high RoB for the confounder or exposure assessment domain) or limiting the analysis to studies exclusively evaluating CBVD mortality (excluding stroke mortality) resulted in a positive association (RE: 1.01) with borderline significance (95% CI: 0.97–1.04 and 95% CI: 0.94–1.07, respectively). In each of these instances, one study was excluded from the main analysis. Similarly, these studies (Bauleo et al., 2019 and Dirgawati et al., 2019) were the only ones noted to influence the main analysis in the leave-one-out analysis. Neither test for publication bias indicated the presence of a bias (Egger's test p = 0. 2724; Begg's test p = 0.4833) and trim-and-fill analysis suggested that one hypothetical study was needed to make the funnel plot symmetrical.

Four studies examined the association between CSD mortality and NO_x, one of which provided the REs by quintile of exposure. No exposure-response relationship could be determined from this study (Andersson et al., 2020). Four of the seven REs reported for CSD mortality were positive, of which three were borderline significant. Similarly, two of the three REs pertaining to MI mortality were positive and borderline significant. One RE was reported for each of cardiac disease mortality and ACE mortality; neither indicated a positive association.

NO

The literature searches identified three studies conducted in Canada, the USA, and Asia evaluating the association between circulatory causes of mortality and NO; the study details are depicted in Appendix A.3. They were drawn from the general population with cohorts ranging in size from 41,869 (Alexeeff et al., 2018) to 406,232 (Gan et al., 2011). NO exposure was assessed by LUR models (n = 2) or hybrid models (n = 1) and were based on the residential address or postal code. Mean NO ranged from 6.0 (Alexeeff et al., 2018) to 489 µg/m³ (Barratt et al., 2018). The studies each examined one or more endpoints pertaining to circulatory mortality, including mortality from CSD (n = 1), CHD (n = 3), and CBVD (n = 2). They included adjustments for the key confounders of age, sex, and SES, as well as additional confounders, but Gan et al. (2011) did not adjust for smoking. Based on a limited number of REs for each mortality endpoint, mixed results were observed for CHD and CBVD and there was no evidence of an association for CSD.

3.3.2 PM