United States
Environmental Protection
»m Agency
EPA/600/R-21/198
September 2021
www. epa. gov/isa
Supplement to the 2019 Integrated
Science Assessment
for
Particulate Matter
(External Review Draft)
September 2021
Center for Public Health and Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Research Triangle Park, NC
-------
DISCLAIMER
This document is an external review draft, for review purposes only. This information is
distributed solely for predissemination peer review under applicable information quality guidelines. It has
not been formally disseminated by EPA. It does not represent and should not be construed to represent
any Agency determination or policy. Mention of trade names or commercial products does not constitute
endorsement or recommendation for use.
September 2021
ii
DRAFT: Do Not Cite or Quote
-------
CONTENTS
INTEGRATED SCIENCE ASSESSMENT TEAM FOR PARTICULATE MATTER XI
EXECUTIVE SUMMARY ES-1
1. INTRODUCTION AND SCOPE 1-1
1.1 Introduction 1-1
1.2 Rationale and Scope 1-1
1.2.1 Rational for Inclusion of Health and Welfare Effects 1-2
1.2.2 Scope 1-3
1.3 Organization of the Supplement 1-4
2. OVERVIEW OF MAIN CONCLUSIONS OF THE 2019 INTEGRATED SCIENCE
ASSESSMENT FOR PARTICULATE MATTER 2-1
2.1 Health Effects 2-1
2.1.1 Health Effects of PM2.5 2-3
2.1.1.1 Respiratory Effects 2-3
2.1.1.1.1 Respiratory Effects Associated with Short-Term PM2.5 Exposure 2-4
2.1.1.1.2 Respiratory Effects Associated with Long-Term PM2.5 Exposure 2-5
2.1.1.2 Cardiovascular Effects 2-6
2.1.1.2.1 Cardiovascular Effects Associated with Short-Term PM2.5 Exposure 2-6
2.1.1.2.2 Cardiovascular Effects Associated with Long-Term PM2.5 Exposure 2-8
2.1.1.3 Nervous System Effects 2-10
2.1.1.3.1 Nervous System Effects Associated with Long-Term PM2.5 Exposure 2-10
2.1.1.4 Cancer 2-11
2.1.1.4.1 Cancer Associated with Long-Term PM2.5 Exposure 2-11
2.1.1.5 Mortality 2-13
2.1.1.5.1 Mortality Associated with Short-Term PM2.5 Exposure 2-13
2.1.1.5.2 Mortality Associated with Long-Term PM2.5 Exposure 2-14
2.2 Policy-Relevant Considerations 2-24
2.2.1 Potential Copollutant Confounding 2-24
2.2.1.1 Short-Term PM2.5 Exposure 2-25
2.2.1.2 Long-Term PM2.5 Exposure 2-26
2.2.2 Timing of Effects 2-27
2.2.2.1 Averaging Time 2-27
2.2.2.2 Lag Structure of Associations 2-28
2.2.3 Concentration-Response Relationship 2-29
2.2.3.1 Short-Term Exposure 2-29
2.2.3.2 Long-Term Exposure 2-30
2.2.4 PM Components and Sources 2-31
2.2.4.1 Respiratory Effects 2-32
2.2.4.2 Cardiovascular Effects 2-33
2.2.4.3 Mortality 2-34
2.2.5 Populations and Lifestages at Potentially Increased Risk of a PM-Related Health Effect _ 2-34
2.3 Welfare Effects 2-38
2.3.1 Visibility Impairment 2-38
2.3.2 Climate Effects 2-38
September 2021
iii
DRAFT: Do Not Cite or Quote
-------
2.3.3 Materials Effects
2-39
3. EVALUATION OF RECENT HEALTH EFFECTS EVIDENCE 3-1
3.1 Cardiovascular Effects 3-2
3.1.1 Short-Term PIVte 5 Exposure 3-2
3.1.1.1 Summary and Causality Determination from 2019 Integrated Science
Assessment for Particulate Matter 3-2
3.1.1.2 Recent U.S. and Canadian Epidemiologic Studies 3-7
3.1.1.2.1 Ischemic Heart Disease and Myocardial Infarction 3-8
3.1.1.2.2 Cerebrovascular Disease and Stroke 3-11
3.1.1.2.3 Heart Failure 3-14
3.1.1.2.4 Arrhythmia 3-16
3.1.1.2.5 Combinations of Cardiovascular-Related Outcomes 3-18
3.1.1.2.6 Cardiovascular Mortality 3-19
3.1.1.2.7 Potential Copollutant Confounding 3-20
3.1.1.2.8 Lag Structure of Associations 3-20
3.1.1.3 Recent Studies Examining the PM2.s-Cardiovascular Effects Relationship
through Accountability Analyses and Causal Modeling Methods 3-21
3.1.1.4 Summary of Recent Evidence in the Context of the 2019 Integrated Science
Assessment for Particulate Matter Causality Determination for Short-Term
PM2.5 Exposure and Cardiovascular Effects 3-23
3.1.2 Long-Term PM2.5 Exposure 3-24
3.1.2.1 Summary and Causality Determination from 2019 Integrated Science
Assessment for Particulate Matter 3-25
3.1.2.2 Recent U.S. and Canadian Epidemiologic Studies 3-29
3.1.2.2.1 Ischemic Heart Disease and Myocardial Infarction 3-30
3.1.2.2.2 Cerebrovascular Disease and Stroke 3-33
3.1.2.2.3 Atherosclerosis 3-35
3.1.2.2.4 Heart Failure and Impaired Heart Function 3-36
3.1.2.2.5 Cardiac Electrophysiology and Arrhythmia 3-37
3.1.2.2.6 Blood Pressure and Hypertension 3-38
3.1.2.2.7 Cardiovascular Mortality 3-38
3.1.2.2.8 Copollutant Confounding 3-39
3.1.2.2.9 Examination of the Concentration-Response (C-R) Relationship between
Long-Term PM2.5 Exposure and Cardiovascular Effects 3-41
3.1.2.3 Recent Studies Examining the PM2.s-Cardiovascular Effects Relationship
through Accountability Analyses and Causal Modeling Methods 3-46
3.1.2.4 Summary of Recent Evidence in the Context of the 2019 Integrated Science
Assessment for Particulate Matter Causality Determination for Long-Term
PM2.5 Exposure and Mortality 3-48
3.2 Mortality 3-49
3.2.1 Short-Term PM2.5 Exposure 3-49
3.2.1.1 Summary and Causality Determination from 2019 Integrated Science
Assessment for Particulate Matter 3-49
3.2.1.2 Recent U.S. and Canadian Epidemiologic Studies 3-54
3.2.1.2.1 All Cause and Total (Nonaccidental) Mortality 3-55
3.2.1.2.2 Cause-specific Mortality 3-58
3.2.1.2.3 Potential Copollutant Confounding of the PM2.s-Mortality Relationship 3-59
3.2.1.2.4 Effect Modification of the PM2.s-Mortality Relationship 3-60
September 2021
iv
DRAFT: Do Not Cite or Quote
-------
3.2.1.2.5 Lag Structure of Associations 3-64
3.2.1.2.6 Examination of the Concentration-Response (C-R) Relationship between Short-
Term PM2.5 Exposure and Mortality 3-64
3.2.1.3 Recent Studies Examining the PIVh s-Mortality Relationship through
Accountability Analyses and Causal Modeling Methods 3-67
3.2.1.4 Summary of Recent Evidence in the Context of the 2019 Integrated Science
Assessment for Particulate Matter Causality Determination for Short-Term
PM2.5 Exposure and Mortality 3-69
3.2.2 Long-Term PM2 5 Exposure 3-70
3.2.2.1 Summary and Causality Determination from 2019 Integrated Science
Assessment for Particulate Matter 3-70
3.2.2.2 Recent U.S. and Canadian Cohort Studies 3-75
3.2.2.2.1 All Cause and Total (Nonaccidental) Mortality 3-76
3.2.2.2.2 Cause-Specific Mortality 3-84
3.2.2.2.3 Long-Term PM2.5 Exposure and Mortality in Populations with Preexisting
Conditions 3-91
3.2.2.2.4 Studies of Life Expectancy 3-92
3.2.2.2.5 Potential Copollutant Confounding of the PM2 5-Mortality Relationship 3-93
3.2.2.2.6 Novel Methods to Address Potential Confounding 3-95
3.2.2.2.7 Examination of the Concentration-Response (C-R) Relationship between
Long-Term PM2.5 Exposure and Mortality 3-99
3.2.2.3 Recent Studies Examining the PM2.s-Mortality Relationship through
Accountability Analyses and Causal Modeling Methods 3-105
3.2.2.4 Summary of Recent Evidence in the Context of the 2019 Integrated Science
Assessment for Particulate Matter Causality Determination for Long-Term
PM2.5 Exposure and Mortality 3-120
3.3 Key Scientific Topics that Further Inform the Health Effects of PM2.5 3-121
3.3.1 Recent Experimental Studies at Near-Ambient Concentrations 3-121
3.3.2 PM2.5 Exposure and COVID-19 Infection and Death 3-123
3.3.2.1 Short-Term PM2.5 Exposure 3-123
3.3.2.2 Long-Term PM2.5 Exposure 3-124
3.3.2.3 Summary of Recent Epidemiologic Studies Examining PM2.5 Exposure and
COVID-19 Infection and Death 3-126
3.3.3 Populations and Lifestages at Potentially Increased Risk of a PM-Related Health Effect 3-126
3.3.3.1 Socioeconomic Status 3-127
3.3.3.1.1 Exposure Disparity 3-128
3.3.3.1.2 Health Risk Disparity 3-131
3.3.3.2 Race/Ethnicity 3-138
3.3.3.2.1 Exposure Disparity 3-139
3.3.3.2.2 Health Risk Disparity 3-143
3.3.3.3 Summary of Recent Evidence on At-Risk Populations in the Context of
Conclusions of the 2019 Integrated Science Assessment for Particulate Matter 3-150
EVALUATION OF RECENT WELFARE EFFECTS EVIDENCE 4-1
4.1 Summary of Evidence for Visibility Effects from 2019 Integrated Science Assessment for
Particulate Matter 4-1
4.2 Recent Studies that Inform the Relationship between PM and Visibility Effects 4-2
4.2.1 Visibility Preference and Light Extinction 4-2
4.2.2 Recent Advancements in Visibility Monitoring and Assessment 4-7
4.3 Summary of Recent Evidence in the Context of the 2019 Integrated Science Assessment for
Particulate Matter Causality Determination for Visibility Effects 4-9
2021
v
DRAFT: Do Not Cite or Quote
-------
5. SUMMARY AND CONCLUSIONS 5-1
APPENDIX A A-1
REFERENCES R-1
September 2021 vi DRAFT: Do Not Cite or Quote
-------
LIST OF TABLES
Table 2-1
Table 2-2
Table 2-3
Table 3-1
Table 3-2
Table 3-3.
Table 3-4
Table 3-5
Table 3-6
Table 3-7
Causal and likely to be causal causality determinations for short- and long-term PM2.5
exposure. 2-3
Key evidence contributing to causal and likely to be causal causality determinations for
PM2.5 exposure and health effects evaluated in the 2019 Integrated Science Assessment
for Particulate Matter. 2-16
Key evidence contributing to causal causality determinations for PM exposure and welfare
effects evaluated in the 2019 Integrated Science Assessment for Particulate Matter. 2-40
Summary of evidence for a causal relationship between short-term PM2.5 exposure and
cardiovascular effects from the 2019 Integrated Science Assessment for Particulate
Matter. 3-3
Summary of evidence for a causal relationship between long-term PM2.5 exposure and
cardiovascular effects from the 2019 Integrated Science Assessment for Particulate
Matter. 3-26
Summary of studies examining the concentration-response (C-R) relationship or
conducted threshold analyses for long-term exposure to PM2.5 and cardiovascular
morbidity. 3-41
Summary of evidence for a causal relationship between short-term PM2.5 exposure and
total mortality from the 2019 Integrated Science Assessment for Particulate Matter. 3-50
Summary of evidence for a causal relationship between long-term PM2.5 exposure and
total mortality from the 2019 Integrated Science Assessment for Particulate Matter. 3-72
Summary of studies examining the concentration-response (C-R) relationship or
conducted threshold analyses for long term PM2.5 exposure and mortality. 3-100
Description of methods from epidemiologic studies using accountability analyses or causal
modeling methods to examine long-term exposure to PM2.5 and mortality. 3-106
September 2021
vii
DRAFT: Do Not Cite or Quote
-------
LIST OF FIGURES
Figure 3-9
Figure 3-15
¦14
¦16
Figure 3-1 Results of studies of short-term PM2.5 exposure and hospital admissions and emergency
department visits for ischemic heart disease. 3
Figure 3-2 Results of studies of short-term PM2.5 exposure and hospital admissions and emergency
department visits for stroke. 3
Figure 3-3 Results of studies of short-term PM2.5 exposure and hospital admissions and emergency
department visits for heart failure. 3
Figure 3-4 Results of studies of short-term PM2.5 exposure and hospital admissions and emergency
department visits for arrhythmia. 3-18
Figure 3-5 Associations between long-term PM2.5 exposure and ischemic heart disease or myocardial
infarction. 3-33
Figure 3-6 Associations between long-term PM2.5 exposure and the incidence of stroke. 3-35
Figure 3-7 Associations between long-term exposure to PM2.5 and cardiovascular morbidity in single
pollutant models and models adjusted for copollutants. 3-40
Figure 3-8 Concentration-response relationship for the association of PM2.5 concentration with acute
myocardial infarction. 3-44
Concentration-response relationship for the association of PM2.5 concentration with acute
myocardial infarction using SCHIF (A) and penalized splines (B) with 4 degrees of
freedom.
Figure 3-10 Concentration-response relationship for the association of PM2.5 concentration with first
admissions for myocardial infarction.
Figure 3-11 Predicted log hazard for incident non-fatal myocardial infarction versus previous 1-year
mean ambient PM2.5 concentration. 3-45
Figure 3-12 Analysis steps used by Danesh Yazdi et al. (2021) to examine long-term PM2.5 exposure
and cardiovascular-related hospital admissions. 3-47
Figure 3-13 Summary of associations between short-term PM2.5 exposure and total (nonaccidental)
mortality in multicity studies. 3-56
Figure 3-14 Summary of associations between short-term PM2.5 exposure and cardiovascular and
respiratory mortality in multicity studies. 3-59
Odds ratio and 95% confidence intervals for lag 0 and lag 0-2 days of nonaccidental
mortality across tertiles of lag 0, lag 0-2, and lag 0-4 oxidant gases across 24 Canadian
cities. 3-62
Figure 3-16 Associations between short-term PM2.5 exposure and nonaccidental mortality at lag 1 for
the 312 core-based statistical areas examined in Baxter et al. (2019). 3-63
Figure 3-17 Concentration-response curves for the U.S. (A) and Canada (B) using a B-spline function
with knots at the 25th and 75th percentiles of PM2.5 concentrations across all cities in each
location. 3-65
3-44
3-45
September 2021 viii DRAFT: Do Not Cite or Quote
-------
Figure 3-18 Concentration-response curves for nonaccidental, cardiovascular, and respiratory
mortality using natural cubic splines with 3 degrees of freedom for associations with
0-2-day PM2.5 across 24 Canadian cities. 3-66
Figure 3-19 Associations between long-term PM2.5 exposure and total (nonaccidental) mortality in
recent North American cohorts. 3-77
Figure 3-20 Associations between long-term PM2.5 exposure and all cause and total (nonaccidental)
mortality in cohort studies in the U.S. and Canada published since the 2019 Integrated
Science Assessment for Particulate Matter. 3-78
Figure 3-21 Hazard ratios for spatially-decomposed analyses for an interquartile range increase in
PM2.5 concentrations for all cause mortality. 3-80
Figure 3-22 Associations between long-term PM2.5 exposure and all cardiovascular disease and all
respiratory disease mortality in recent North American cohorts. 3-85
Figure 3-23 Associations between long-term PM2.5 exposure and cause-specific cardiovascular
mortality in recent North American cohorts. 3-88
Figure 3-24 Associations between long-term PM2.5 exposure and cause-specific respiratory mortality in
recent North American cohorts. 3-90
Figure 3-25 Life expectancy losses by county for females and males as estimated by Bennett et al.
(2019). 3-93
Figure 3-26 Mortality risk ratios for a 10 |jg/m3 increase in PM2.5 by length of study for the base model
and temporal adjusted models in Eum et al. (2018). 3-97
Figure 3-27 Shape Constrained Health Impact Function predictions by PM2.5 concentration for the
pooled CanCHEC cohort. 3-103
Figure 3-28 Estimated concentration-response associations between PM2.5 and all-cause mortality
with a flexible modeling approach within the NHIS cohort. 3-104
Figure 3-29 Percent change in county-level COVID-19 case fatality rate and mortality rate in single
and multipollutant models (January 22, 2020-July 17, 2020). 3-125
Figure 3-30 Differences in PM2.5 exposure by socioeconomic status. 3-130
Figure 3-31 Risk ratio of association between PM2.5 and mortality, stratified by socioeconomic status. 3-132
Figure 3-32 Hazard ratios for the association between PM2.5 and mortality, by greenspace and
community-level material deprivation. 3-133
Figure 3-33 County-level life expectancy losses due to PM2.5 exceeding 2.8 |jg/m3. 3-134
Figure 3-34 Estimated national and socioeconomic deprivation quintile-specific mortality rates. 3-136
Figure 3-35 Odds ratios for the association between PM2.5 and cardiometabolic outcomes by
neighborhood cluster. 3-138
Figure 3-36 Average PM2.5 experienced and caused, by racial-ethnic group. 3-141
Figure 3-37 Source contributions to racial-ethnic disparity in PM2.5 exposure. 3-142
Figure 3-38 Difference in PM2.5 exposure by race. 3-143
September 2021
ix
DRAFT: Do Not Cite or Quote
-------
Figure 3-39 Percent change in cardiovascular disease mortality by PM2.5 exposure, stratified by
census block group racial composition in Massachusetts (2001-2011). 3-145
Figure 3-40 Percent change in cardiovascular disease mortality by PM2.5 exposure, stratified by the
racial residential segregation and index of racial dissimilarity in Massachusetts
(2001-2011). 3-146
Figure 3-41 Relationship between life expectancy and PM2.5 exposure by income inequality and
percent Black. 3-148
Figure 3-42 Probability of cardiometabolic disease and PM2.5 exposure, stratified by race, gender, and
hypertension status. 3-149
Figure 4-1 Percent acceptability levels plotted against atmospheric extinction coefficient for each of
the images used in studies for Washington, DC (WASH), Phoenix, AZ (PHX), Chilliwack,
BC (CHIL), Abbotsford, BC (ABBT), Denver, CO (DEN), and the Grand Canyon, AZ
(GRCA). 4-4
Figure 4-2 Percent acceptability levels plotted against apparent contrast of distant landscape features
for each of the images used in studies for Washington, DC (WASH), Phoenix, AZ (PHX),
Chilliwack, BC (CHIL), Abbotsford, BC (ABBT), Denver, CO (DEN), and the Grand
Canyon, AZ (GRCA). 4-6
September 2021
x
DRAFT: Do Not Cite or Quote
-------
INTEGRATED SCIENCE ASSESSMENT TEAM FOR
PARTICULATE MATTER
Executive Direction
Dr. Steven J. Dutton (Director—Acting)—Center for Public Health and Environmental
Assessment, Office of Research and Development, U.S. Environmental Protection
Agency, Research Triangle Park, NC
Dr. Emily Snyder (Deputy Director—Acting)—Center for Computational Toxicology and
Exposure, Office of Research and Development, U.S. Environmental Protection Agency,
Research Triangle Park, NC
Dr. Jane Ellen Simmons (Branch Chief)—Center for Public Health and Environmental
Assessment, Office of Research and Development, U.S. Environmental Protection
Agency, Research Triangle Park, NC
Dr. Christopher Weaver (Branch Chief)—Center for Public Health and Environmental
Assessment, Office of Research and Development, U.S. Environmental Protection
Agency, Research Triangle Park, NC
Authors
Mr. Jason Sacks (Assessment Lead, Integrated Science Assessment for Particulate
Matter)—Center for Public Health and Environmental Assessment, Office of Research
and Development, U.S. Environmental Protection Agency, Research Triangle Park, NC
Dr. Barbara Buckley (Retired)—Center for Public Health and Environmental Assessment,
Office of Research and Development, U.S. Environmental Protection Agency, Research
Triangle Park, NC
Dr. Stephanie Deflorio-Barker—Center for Public Health and Environmental Assessment,
Office of Research and Development, U.S. Environmental Protection Agency, Research
Triangle Park, NC
Dr. Scott Jenkins—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
Dr. Ellen Kirrane—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
Dr. Alison Krajewski—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
Dr. Thomas Luben—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
Dr. Stephen McDow—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
September 2021
xi
DRAFT: Do Not Cite or Quote
-------
Contributors
Dr. Jean-Jacques Dubois—Center for Public Health and Environmental Assessment, Office
of Research and Development, U.S. Environmental Protection Agency, Research
Triangle Park, NC
Mr. Kevin Park—Oak Ridge Associated Universities, Inc., Center for Public Health and
Environmental Assessment, Office of Research and Development, U.S. Environmental
Protection Agency, Research Triangle Park, NC
Mr. R. Byron Rice—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
Technical Support Staff
Ms. Marieka Boyd—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
Mr. Ryan Jones—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
Mr. Kevin Park—Oak Ridge Associated Universities, Inc., Center for Public Health and
Environmental Assessment, Office of Research and Development, U.S. Environmental
Protection Agency, Research Triangle Park, NC
Reviewers
Dr. Elizabeth Chan—Office of Air Quality Planning and Standards, Office of Air and
Radiation, U.S. Environmental Protection Agency, Research Triangle Park, NC
Dr. Nicole Hagan—Office of Air Quality Planning and Standards, Office of Air and
Radiation, U.S. Environmental Protection Agency, Research Triangle Park, NC
Dr. Iman Hassan—Office of Air Quality Planning and Standards, Office of Air and
Radiation, U.S. Environmental Protection Agency, Research Triangle Park, NC
Dr. Archana Lamichhane—Office of Air Quality Planning and Standards, Office of Air and
Radiation, U.S. Environmental Protection Agency, Research Triangle Park, NC
Ms. Sheila Igoe—Office of General Counsel, U.S. Environmental Protection Agency,
Washington, DC
Dr. Anthony Jones—Office of Air Quality Planning and Standards, Office of Air and
Radiation, U.S. Environmental Protection Agency, Research Triangle Park, NC
Mr. David Orlin—Office of General Counsel, U.S. Environmental Protection Agency,
Washington, DC
Dr. Lars Perlmutt—Office of Air Quality Planning and Standards, Office of Air and
Radiation, U.S. Environmental Protection Agency, Research Triangle Park, NC
Dr. Erika Sasser—Office of Air Quality Planning and Standards, Office of Air and
Radiation, U.S. Environmental Protection Agency, Research Triangle Park, NC
September 2021
xii
DRAFT: Do Not Cite or Quote
-------
Mr. Timothy Watkins—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
Dr. Karen Wesson—Office of Air Quality Planning and Standards, Office of Air and
Radiation, U.S. Environmental Protection Agency, Research Triangle Park, NC
September 2021
xiii
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
EXECUTIVE SUMMARY
In June 2021, the U.S. EPA announced it will reconsider the December 2020 decision to retain
the particulate matter (PM) National Ambient Air Quality Standards (NAAQS). As part of the
reconsideration process, the U.S. EPA indicated that it would develop a supplement to the 2019 Integrated
Science Assessment for PM (PM ISA) to thoroughly evaluate the most up-to-date science that became
available after the literature cutoff date of the 2019 PM ISA that could either further inform the adequacy
of the current PM NAAQS or address key scientific topics that have evolved since the 2020 PM NAAQS
review was completed.
Within this Supplement, the U.S. EPA evaluated recent studies (i.e., published since the literature
cutoff date of the 2019 PM ISA) that are of potentially greatest relevance to the reconsideration of the PM
NAAQS in the context of the findings of the 2019 PM ISA. The studies that formed the basis of the
evaluation presented within this Supplement consist of: (a) U.S. and Canadian epidemiologic studies for
health effect categories for which the 2019 PM ISA concluded a causal relationship (i.e., short- and
long-term PM2 5 exposure and cardiovascular effects and mortality); (b) U.S. and Canadian epidemiologic
studies that employed causal modeling methods or conducted accountability analyses; (c) studies that
examined specific issues of importance including experimental studies conducted at near-ambient PM2 5
concentrations, U.S. and Canadian epidemiologic or exposure studies that examined disparities in PM25
exposure or health risks by race and ethnicity or socioeconomic status, and U.S. and Canadian
epidemiologic studies that examined the association between PM2 5 exposure and Coronavirus Disease
2019 (COVID-19) infection or death; and (d) U.S. and Canadian studies that examined public preferences
for visibility impairment and/or developed methodologies or conducted quantitative analyses of light
extinction. This Supplement to the 2019 PM ISA is not intended to represent the full multidisciplinary
evaluation of evidence that results in the formation of weight-of-evidence conclusions, but instead is an
assessment that puts the results of recent studies in the context of the scientific conclusions (i.e., causality
determinations) presented within the 2019 PM ISA.
This Supplement to the 2019 PM ISA finds that recent studies further support, and in some
instances extend, the evidence that formed the basis of the causality determinations presented within the
2019 PM ISA that characterizes relationships between PM exposure and health (i.e., cardiovascular
effects and mortality) and welfare effects (i.e., visibility impairment). In brief, this Supplement finds the
following:
• Recent U.S. and Canadian epidemiologic studies examining short- and long-term PM2 5 exposure
and cardiovascular effects and mortality provide evidence that is consistent with the conclusions
of the 2019 PM ISA. Relative to the studies evaluated in the 2019 PM ISA, many of the studies
report positive associations at lower PM2 5 concentrations (i.e., annual PM2 5 concentrations
ranging from 5.9 to 16.5 (ig/m3; mean 24-hour avg PM25 concentrations ranging from 7.1 to
15.4 (ig/m3).
September 2021
ES-1
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
o Recent U.S. and Canadian epidemiologic studies examining short- and long-term PM2 5
exposure and cardiovascular effects provide evidence that is consistent with the studies
evaluated in the 2019 PM ISA. Studies examining short-term PM2 5 exposure report
consistent positive associations for cardiovascular-related emergency department (ED)
visits and hospital admissions, specifically for ischemic heart disease (IHD), myocardial
infarction (MI), and heart failure (HF). For long-term exposure there remains strong
evidence for cardiovascular-related mortality with support from studies of cardiovascular
morbidity outcomes including coronary heart disease (CHD), stroke, and atherosclerosis
progression, among individuals with preexisting diseases or patients followed after a
cardiac event or procedure. Associations persisted across studies conducted in different
geographic locations, populations with diverse demographic characteristics, and study
designs (i.e., different exposure assessment methods, and confounder control).
o Relatively few recent U.S. and Canadian epidemiologic studies examined short-term
PM2 5 exposure and mortality; however, these studies continue to provide evidence of
positive associations with both all-cause and total (nonaccidental) mortality as well as
cause-specific mortality outcomes.
o A number of recent long-term PM2 5 exposure and mortality studies conducted in cohorts
consisting of diverse populations and encompassing large geographic locations report
consistent, positive associations, with most reporting mean annual PM2 5 concentrations
ranging from 5.9-11.65 (ig/m3.
¦ Across epidemiologic studies examining both cardiovascular effects and
mortality, sensitivity analyses as well as individual studies, further inform
uncertainties in the evidence base (i.e., copollutant confounding, control for
confounders such as temporal trends and temperature, and the
concentration-response [C-R] relationship). Such analyses increase confidence in
the relationship for both short- and long-term PM2 5 exposures and both health
effect categories, and further support the causality determinations presented in
the 2019 PM ISA.
¦ Since the completion of the 2019 PM ISA, numerous U.S. and Canadian
epidemiologic studies conducted accountability analyses or employed causal
modeling methods to examine both short- and long-term PM2 5 exposure and
cardiovascular effects and mortality. These studies, which used a variety of
statistical methods to control for confounding bias, consistently report positive
associations, which further supports the broader body of epidemiologic evidence
for both cardiovascular effects and mortality.
o Several recent U.S. and Canadian studies provide additional insight on the health effects
of PM2 5 including a recent controlled human exposure study conducted at near-ambient
concentrations, which provided initial evidence of both lung and cardiac function changes
in young, healthy participants as well as studies providing initial assessments of
short- and long-term PM2 5 exposure and COVID-19 infection and death. While some
studies examining COVID-19 report positive associations, these studies are subject to
methodological limitations and require additional exploration.
o The 2019 PM ISA provided evidence that specific lifestages and populations are at
increased risk of a PM2 5-related health effect. Recent U.S. and Canadian epidemiologic
studies support and expand the evidence base within the 2019 PM ISA and indicate that
there are both PM2 5 exposure and health risk disparities by race and ethnicity,
specifically among non-White populations. Additionally recent evidence supports the
September 2021
ES-2
DRAFT: Do Not Cite or Quote
-------
1 evidence presented in the 2019 PM ISA that there may be PM2 5 exposure and health risk
2 disparities by socioeconomic status (SES), specifically among people of low SES.
3 • Recent studies continue to support a relationship between PM and visibility impairment and
4 provide additional insights on the impact of choice of metric on preference study results, impacts
5 of changing PM composition on the relationship between PM and visibility impairment, and
6 alternative approaches to estimating light extinction.
September 2021
ES-3
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
1.INTRODUCTION AND SCOPE
1.1 Introduction
The U.S. Environmental Protection Agency (U.S. EPA) completed the Integrated Science
Assessment for Particulate Matter (2019 PM ISA) in December 2019 (U.S. EPA. 2019). The 2019 PM
ISA builds upon the evidence evaluated and scientific conclusions presented in prior assessments
including the 2009 PM ISA (U.S. EPA. 2009) and earlier assessments, e.g., 2004 PM Air Quality Criteria
Document [AQCD; (U.S. EPA. 2004)1 and 1996 PM AQCD (U.S. EPA. 1996). Within the 2019 PM ISA,
evidence spanning scientific disciplines (e.g., atmospheric chemistry, exposure science, animal
toxicological, human clinical, epidemiology) was evaluated to assess the causal nature of relationships
between PM and health and nonecological welfare effects using a weight-of-evidence approach
extensively detailed in the Preamble to the Integrated Science Assessments (U.S. EPA. 2015) and the
appendix of the 2019 PM ISA.1
The key science judgments (i.e., causality determinations) detailed within the 2019 PM ISA
directly informed the development of conclusions outlined within the Policy Assessment for the Review of
the PMNAAQS (2020 PM PA) (U.S. EPA. 2020b). These key science judgments formed the basis of the
discussion on potential alternative primary and secondary National Ambient Air Quality Standards
(NAAQS) for PM within the 2020 PM PA.
On June 10, 2021, the U.S. EPA announced it is reconsidering the December 2020 decision to
retain the PM NAAQS, and as part of this process, the Agency is developing a supplement to the 2019
PM (EPA Press Office. 2021). As a result, the evidence presented within the 2019 PM ISA, along with
the targeted identification and evaluation of new scientific information in this Supplement, provides the
scientific basis to support a robust and thorough reconsideration of the 2020 PM NAAQS.
1.2 Rationale and Scope
In completing the review of the PM NAAQS in December 2020, U.S. EPA provisionally
considered numerous studies published after the literature cutoff date (approximately January 2018) for
the 2019 PM ISA. In reviewing these studies, as explained in Responses to Significant Comments on the
2020 Proposed Decision on the National Ambient Air Quality Standards for Particulate Matter the
U.S. EPA "concluded that none of the studies materially change any of the broad scientific conclusions of
hereafter welfare effects refers to nonecological welfare effects, unless otherwise noted. The ecological effects
resulting from the deposition of PM and PM components are being considered in a separate assessment as part of the
review of the secondary (welfare-based) NAAQS for oxides of nitrogen, oxides of sulfur, and PM (U.S. EPA.
2020a).
September 2021
1-1
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
the ISA regarding the health and welfare effects of PM or warrant reopening the air quality criteria for
this review" (U.S. EPA. 2020c).
To inform the reconsideration of the PM NAAQS, the U.S. EPA determined that a thorough
evaluation is warranted of some studies that became available after the literature cutoff date of the 2019
PM ISA that could either further inform the adequacy of the current PM NAAQS or address key scientific
topics that have evolved since the literature cutoff date for the 2019 PM ISA. To facilitate the
identification and evaluation of recent studies that warrant review, the U.S. EPA developed a rationale
(Section 1.2.1) and scope (Section 1.2.2) for this Supplement to the 2019 PM ISA to focus on specific
PM-related health and welfare effects most pertinent to the U.S. EPA in support of the reconsideration of
the primary and secondary PM NAAQS.
1.2.1 Rational for Inclusion of Health and Welfare Effects
The causality determinations presented within the 2019 PM ISA and discussed in Section 2
below, in combination with the characterization of the science with respect to the health and welfare
effects of PM presented in the 2020 PM PA, form the basis of the rationale for the health and welfare
effects evaluated within this Supplement. The following section provides specific details on the rationale
for the types of evidence included, which ultimately forms the basis of the scope that governs the studies
considered for inclusion.
In selecting the health effects to evaluate within this Supplement, the primary rationale is based
on the causality determinations for health effect categories presented in the 2019 PM ISA, and the
subsequent use of the health effects evidence in the 2020 PM PA (U.S. EPA. 2020b). "In considering the
public health protection provided by the current primary PM2 5 standards, and the protection that could be
provided by alternatives, [the U.S. EPA, within the 2020 PM PA] emphasized health outcomes for which
the ISA determined that the evidence supports either a causal or a likely to be causal relationship with
PM2 5 exposures" (U.S. EPA. 2020b). Although the 2020 PM PA initially focused on this broader set of
evidence, the basis of the discussion on potential alternative standards primarily focused on health effect
categories where the 2019 PM ISA concluded a causal relationship (i.e., short- and long-term PM2 5
exposure and cardiovascular effects and mortality) as reflected in Figures 3-7 and 3-8 of the 2020 PM PA
(U.S. EPA. 2020b). Therefore, within this Supplement the focus is only on the health effects evidence
where the 2019 PM ISA concluded a causal relationship.
In addition, this Supplement also considers recent health effects evidence that addresses key
scientific topics where the literature has evolved since the 2020 PM NAAQS review was completed,
specifically since the literature cutoff date for the 2019 PM ISA. These key scientific topics include
experimental studies conducted at near-ambient concentrations, epidemiologic studies that employed
causal modeling methods or conducted accountability analyses, studies that assess the relationship
between PM2 5 exposure and Coronavirus Disease 2019 (COVID-19) infection and death; and in
September 2021
1-2
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
accordance with recent U.S. EPA guidance on addressing environmental justice [e.g., U.S. EPA (2021)1.
studies that examine disparities in PM2 5 exposure and the risk of health effects.
Consistent with the rationale for the health effects, the selection of welfare effects to evaluate
within this Supplement were based on the causality determinations reported in the 2019 PM ISA and the
subsequent use of scientific evidence in the 2020 PM PA. The 2019 PM ISA concluded a causal
relationship for each of the welfare effects categories evaluated (i.e., visibility, climate effects, and
materials effects). While the 2020 PM PA considered the broader set of evidence for these effects, for
climate effects and material effects, it concluded that there remained "substantial uncertainties with regard
to the quantitative relationships with PM concentrations and concentration patterns that limit[ed] [the]
ability to quantitatively assess the public welfare protection provided by the standards from these effects"
(U.S. EPA. 2020b). Given these uncertainties and limitations, the basis of the discussion on conclusions
regarding the secondary standards in the 2020 PM PA primarily focused on visibility effects. Therefore,
this Supplement focuses only on visibility effects in evaluating newly available scientific information.
1.2.2 Scope
Building on the rationale presented in Section 1.2.1. the scope of this Supplement provides
specific criteria for the types of studies considered for evaluation. The studies considered for inclusion
within the Supplement consist of peer-reviewed studies published from approximately January 2018
through March 2021 that satisfy the following criteria:
Health Effects
• Health effect categories where the 2019 PM ISA concluded a causal relationship (i.e., short- and
long-term PM2 5 exposure and cardiovascular effects and mortality). Additionally, for these health
effect categories the recent studies evaluated are limited to:
o U.S. and Canadian epidemiologic studies
o Epidemiologic studies that employed causal modeling methods or conducted
accountability analyses (i.e., examined the impact of a policy on reducing PM2 5
concentrations)2
Key Scientific Topics
• Experimental studies (i.e., controlled human exposure and animal toxicological) conducted at
near-ambient PM2 5 concentrations
• At-risk populations
2These studies do not include studies that instituted a specific action or intervention to reduce or mitigate exposure,
such as the installation of high efficiency particle filers (HEP A) or indoor air cleaners.
September 2021
1-3
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
o U.S.- and Canadian-based epidemiologic or exposure studies examining potential
disparities in either PM2 5 exposures or the risk of health effects by race/ethnicity or
socioeconomic status (SES)
• U.S.- and Canadian-based epidemiologic studies that examined the relationship between PM2 5
exposures and COVID-19 infection and/or death
Welfare Effects
• U.S. and Canadian studies that provide new information on public preferences for visibility
impairment and/or developed methodologies or conducted quantitative analyses of light
extinction
Given the narrow scope of this Supplement (i.e., not focusing on the broader body of
experimental studies) it is important to recognize the evaluation conducted does not encompass the full
multidisciplinary evaluation presented within the 2019 PM ISA as described in the Preamble to the
Integrated Science Assessments (U.S. EPA. 2015) that would result in weight-of-evidence conclusions on
causality (i.e., causality determinations). Therefore, this Supplement critically evaluates and provides key
study specific information for those recent studies deemed to be of greatest significance for impending
regulatory decisions regarding the PM NAAQS in the context of the body of evidence and scientific
conclusions presented in the 2019 PM ISA. As such, the Supplement indicates whether recent evidence
supports (is consistent with), supports and extends (is consistent with and reduces uncertainties), or does
not support (is not consistent with) the causality determinations in the 2019 PM ISA.
1.3 Organization of the Supplement
The Supplement to the 2019 PM ISA is not intended to be a standalone document, but instead to
build on the established scientific record regarding the health and welfare effects of PM presented in the
2019 PM ISA and prior assessments. As a result, this Supplement evaluates selected recent studies
(i.e., studies published since the literature cutoff date of the 2019 PM ISA and fall within the scope as
outlined above) in the context of the scientific conclusions presented in the 2019 PM ISA.
This Supplement includes chapters and sections incorporated verbatim from the 2019 PM ISA to
provide the background information and scientific conclusions necessary to put recent studies in the
appropriate context. Section 2 of this Supplement consists of the Integrated Synthesis chapter (Chapter 1)
of the 2019 PM ISA, which integrates and summarizes the overall scientific conclusions of the 2019 PM
ISA. Section 3 represents the evaluation of the health effects evidence (i.e., short- and long-term PM2 5
exposure and cardiovascular effects and mortality) that falls within the scope of this Supplement. Within
each section of Section 3. the summary and causality determination from the 2019 PM ISA is presented to
capture the scientific conclusions of the ISA, which recent literature builds upon. The sections that follow
in Section 3 evaluate and integrate the evidence from recent studies and ultimately assess the results of
recent studies in the context of the causality determinations presented in the 2019 PM ISA. Additionally,
September 2021
1-4
DRAFT: Do Not Cite or Quote
-------
1 Section 3 evaluates recent studies that assess key science topics that have evolved since the completion of
2 the 2019 PM ISA. Section 4 consists of an evaluation of recent studies that inform visibility effects and is
3 organized similar to the health effects chapter. Therefore, Section 4 first presents the summary and
4 causality determination from the 2019 PM ISA, then evaluates recent studies, and concludes by assessing
5 new evidence in the context of the conclusions for visibility impairment presented in the 2019 PM ISA.
6 Lastly, Section 5 provides a summary and presents overarching conclusions based on the evaluation of
7 recent studies within this Supplement.
September 2021
1-5
DRAFT: Do Not Cite or Quote
-------
2.OVERVIEW OF MAIN CONCLUSIONS OF THE
2019 INTEGRATED SCIENCE ASSESSMENT FOR
PARTICULATE MATTER
Overall ( onclnsion.\ oj the 2tH{> Particulate Matter (I'M) Integrated Science Assessment (ISA)
• I !\ idcncc spaiiuum scientific disciplines (i c.. atmospheric chemistry. exposure science, dosimetry.
epidemiology. controlled human exposure. ;ind animal to.xicolouy i hmli upon c\ idcncc detailed in ihc
2<>()1) I'M IS \ ;md reaffirmed ;i c .ms.il rel,iii"iislii/< between short- ;md loim-tcrni I'M exposure ;nid
cardio\ ;iscnl;ii' effects ;ind total (uouaccidciital) mortality. ;ind ;i lilelv i" he > tius.il reLiiimishii' l\.»r
respiratory cllccls
• I Apcriniciilal ;ind epidemiologic c\ idcncc supported ;i lilelv i" he > .msul rcLiiimishii' hclwccn
loim-tcrni I'M exposure ;md ucr\ oiis s\ sicni cl lccls
• I !\ idcncc. primarily from sindics of hum c;mccr incidence ;md mortality. in combination w illi llie
decides til' resc;ircli on llie niulaucuicity ;md carcinogenicity of I'M supported ;i lilelv1" he > uus.il
rehiiiiiiishii' hclwccn loim-icrni I'M e\posnrc ;ind cancer
• Rcniaiuiim iiuccri;iiiilies ;ind limitations in llie sciculific c\ idcncc coiiirihulcd lo ;i surestive "/. hut ii"i
sufficient to infer, a causal relationship and inadequate lo infer the presence or absence of a causal
rcLiiiniiship I'orall oilier exposure, si/c fraction, and health effects catcuoix combinations
• I !\ idcncc hiuli upon and reaffirmed that there is a c .ms.il reLiiiniiship between I'M and the
uouccolomcal welfare effects \ isibility inipairnieiit. climate effects, and materials effects.
• I lie assessment of I'M sources and components confirmed and continued lo support the conclusion
from the 2009 PM ISA: lany P\ 12.5 components and sources are associated with many health effects,
and the evidence does not indicate that any one source or component is more strongly related with
health effects than l'.\ I2.5 mass.
• Manx populations (c.u . healths. diseased, etc laud hfesiaues ic u . children, older adults, etc iha\c
heeu show u to he at risk of a health effect iu response lo short- or louu-ierni I'M exposure, particularly
I'M I lowc\ er. of the populations and lifestaucs examined, scientific c\ idcncc indicated that only
some populations may he at disi>nii>urii
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Preface of the 2019 PM ISA, the evaluation of the health effects evidence focuses on exposures
conducted at concentrations of PM that are relevant to the range of human exposures across ambient
microenvironments (up to 2 mg/m3, which is one to two orders of magnitude above ambient
concentrations), and studies that (1) include a composite measure of PM3 or (2) apply some approach to
assess the direct effect of a specific PM size-fraction when the exposure of interest is a source-based
mixture (e.g., diesel exhaust, gasoline exhaust, wood smoke).
Consistent with the Integrated Synthesis chapter (Chapter 1) of the 2019 PM ISA, the subsequent
sections and accompanying table (Table 2-2) summarize the key evidence that informed the causality
determinations for relationships between PM exposure and health effects detailed in the 2019 PM ISA,
specifically those relationships where it was determined that a causal or likely to be causal relationship
exists (Table 2-1). While the following sections of this chapter focus on health effects categories where
the evidence supported a causal or likely to be causal relationship, this Supplement as reflected in the
Scope (Section 1.2.2) focuses on a narrower evidence base in subsequent chapters. These causality
determinations draw from evidence related to the biological plausibility of PM-related health effects and
the broader health effects evidence described in detail within the 2019 PM ISA in Chapter 5-Chapter 11,
as well as information on dosimetry in Chapter 4 and exposure assessment in Chapter 3. Those
relationships between PM and health effects where the 2019 PM ISA was concluded that the evidence
supported a causality determination of suggestive of, but not sufficient to infer, a causal relationship or
inadequate to infer the presence or absence of a causal relationship are not discussed within this chapter,
but are more fully discussed in the 2019 PM ISA.
3Composite measures of PM may include mass, volume, surface area, or number concentration.
September 2021 2-2 DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Table 2-1 Causal and //Ae/y to be causal causality determinations for short- and
long-term PM2.5 exposure.
Size Fraction Health Effects Category Exposure Duration Causality Determination Section
PM2.5 Respiratory Short-term Likely to be causal 2.1.1.1.1
Long-term Likely to be causal 2.1.1.1.2
Cardiovascular
Short-term
Causal
2.1.1.2.1
Long-term
Causal
2.1.1.2.2
Nervous system
Long-term
Likely to be causal
2.1.1.3.1
Cancer
Long-term
Likely to be causal
2.1.1.4.1
Mortality
Short-term
Causal
2.1.1.5.1
Long-term
Causal
2.1.1.5.2
PM2.5 = particulate matter with a nominal mean aerodynamic diameter less than or equal to 2.5 |jm.
2.1.1 Health Effects of PM2.5
Substantial scientific evidence exists across disciplines (i.e., animal toxicology, controlled human
exposure, and epidemiology) showing that both short- and long-term PM2 5 exposure can result in a range
of health effects, from changes in circulating biomarkers to mortality. However, the strength of the PM25
exposure-health effects relationship varies depending on the exposure duration (i.e., short- or long-term)
and broad health effects category (e.g., cardiovascular effects, respiratory effects) examined. Across the
broad health effects categories examined in the 2019 PM ISA, the evidence supporting biological
plausibility varies, but generally includes modulation of the autonomic nervous system and inflammation
as part of the pathways leading to overt health effects. Discussions of subsequent events that could occur
due to deposition of inhaled PM2 5 in the respiratory tract are detailed in the biological plausibility
sections of each health chapter in the 2019 PM ISA and summarized in the following sections.
2.1.1.1 Respiratory Effects
Scientific evidence presented in the 2019 PM ISA continues to support the conclusion of the 2009
PM ISA that there is a likely to be causal relationship between both short- and long-term PM2 5 exposure
and respiratory effects. These causality determinations are based on the consistency of findings within
disciplines; the coherence of evidence across disciplines, including epidemiologic, and animal
toxicological studies, with more limited evidence from controlled human exposure studies; and the
September 2021
2-3
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
evidence supporting biologically plausible pathways for respiratory effects, such as asthma exacerbation,
development of asthma, chronic obstructive pulmonary disease (COPD) exacerbation, and respiratory
mortality.
2.1.1.1.1 Respiratory Effects Associated with Short-Term PM2.5 Exposure
Epidemiologic studies provide strong evidence for overt respiratory effects, including
respiratory-related emergency department (ED) visits and hospital admissions and respiratory mortality
associated with short-term PM2 5 exposure, with coherence provided by some evidence of respiratory
effects from experimental studies. Collectively this evidence supported the conclusion of the 2009 PM
ISA that there is a likely to be causal relationship between short-term PM2 5 exposure and respiratory
effects (Table 2-2). This conclusion is based on multiple epidemiologic studies demonstrating generally
consistent, positive associations with ED visits and hospital admissions for asthma, COPD, and combined
respiratory-related diseases, as well as with respiratory mortality. Evidence from animal toxicological
studies, although limited, was supportive of and provided biological plausibility for the associations
observed in the epidemiologic studies related to exacerbation of asthma and COPD as well as respiratory
infection.
Epidemiologic studies evaluated in the 2019 PM ISA continue to provide strong evidence for a
relationship between short-term PM2 5 exposure and several respiratory-related endpoints, including
asthma exacerbation (2019 PM ISA, Section 5.1.2.1), COPD exacerbation (2019 PM ISA,
Section 5.1.4.1), and combined respiratory-related diseases (2019 PM ISA, Section 5.1.6), particularly
from studies examining ED visits and hospital admissions. The consistent positive associations between
short-term PM2 5 exposure and asthma and COPD ED visits and hospital admissions across studies that
used different approaches to control for the potential confounding effects of weather (e.g., temperature)
are supported by epidemiologic studies demonstrating associations with other respiratory-related effects,
such as symptoms and medication use that are indicative of asthma and COPD exacerbations (2019 PM
ISA, Section 5.1.2.2 and Section 5.1.4.2). The collective body of epidemiologic evidence for asthma
exacerbation was more consistent in children than in adults. Epidemiologic studies examining the
relationship between short-term PM2 5 exposure and respiratory mortality provided evidence of consistent
positive associations, indicating a continuum of effects from morbidity to mortality (2019 PM ISA,
Section 5.1.9).
Building off the studies evaluated in the 2009 PM ISA, epidemiologic studies evaluated in the
2019 PM ISA expanded the assessment of potential copollutant confounding. There was some evidence
that PM2 5 associations with asthma exacerbation, combined respiratory-related diseases, and respiratory
mortality remain relatively unchanged in copollutant models with gaseous pollutants (i.e., O3, NO2, SO2,
with more limited evidence for CO) and other particle sizes (i.e., PM10-2.5; 2019 PM ISA,
Section 5.1.10.1). The uncertainty as to whether there is an independent effect of PM25 on respiratory
health, was partially addressed by findings from animal toxicological studies.
September 2021
2-4
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Animal toxicological studies of short-term PM2 5 exposure provided coherence and biological
plausibility for asthma and COPD exacerbations by demonstrating asthma-related responses in an animal
model of allergic airways disease and enhanced lung injury and inflammation in an animal model of
COPD (2019 PM ISA, Section 5.1.2.4.4 and Section 5.1.4.4.3). There was also abroad body of animal
toxicological studies examining respiratory effects due to short-term PM2 5 exposure, but most of this
evidence was from studies conducted in healthy animals, and therefore, does not provide coherence with
the results of epidemiologic studies examining effects in people with asthma or COPD. This evidence
base also provided consistent evidence for respiratory irritant effects; limited evidence for altered host
defense, greater susceptibility to bacterial infection, and allergic sensitization; and some evidence for
pulmonary injury, inflammation, and oxidant stress. Controlled human exposure studies conducted in
people with asthma or COPD provided minimal evidence of effects due to short-term PM2 5 exposure,
such as decrements in lung function and pulmonary inflammation. These studies are limited in terms of
endpoints evaluated and the size and health status of study subjects.
2.1.1.1.2 Respiratory Effects Associated with Long-Term PM2.5 Exposure
Epidemiologic studies provided strong evidence for effects on lung development, with additional
evidence for the development of asthma in children due to long-term PM2 5 exposure. Evidence from
animal toxicological studies, although limited, was supportive of and provided biological plausibility for
the associations reported in epidemiologic studies related to lung development and the development of
asthma. There was also epidemiologic evidence supporting a decline in lung function in adults in response
to long-term PM2 5 exposure. Collectively this evidence supported the conclusions of the 2009 PM ISA
that there is a likely to be causal relationship between long-term PM2 5 exposure and respiratory effects
(Table 2-2).
Epidemiologic studies evaluated in the 2019 PM ISA continued to support an association between
long-term PM2 5 exposure and several respiratory-related endpoints in children and adults. In children,
studies in multiple cohorts provided strong evidence for decrements in lung function growth (2019 PM
ISA, Section 5.2.2.1.1). Robust and persistent effects were observed across study locations, exposure
assessment methods, and time periods. An animal toxicological study demonstrating impaired lung
development resulting from pre- and postnatal PM2 5 exposure provided biological plausibility for these
findings (2019 PM ISA, Section 5.2.2.1.2). Results of prospective cohort studies in children also provided
some evidence for asthma development in children and are supported by other studies examining asthma
prevalence in children, childhood wheeze, and pulmonary inflammation (2019 PM ISA, Section 5.2.3).
Biological plausibility was provided by an animal toxicological study of long-term PM2 5 exposure
demonstrating the development of an allergic phenotype and increase in airway responsiveness (2019 PM
ISA, Section 5.2.3.3.2). There was limited evidence of increased bronchitic symptoms and hospitalization
in children with asthma in relation to long-term PM2 5 exposure (2019 PM ISA, Section 5.2.7). In adults,
long-term PM2 5 exposure was found to be associated with accelerating lung function decline (2019 PM
September 2021
2-5
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
ISA, Section 5.2.2.2.2). Consistent evidence was observed for respiratory mortality and cause-specific
respiratory mortality for COPD and respiratory infection (2019 PM ISA, Section 5.2.10), providing
evidence of a continuum of effects in response to long-term PM2 5 exposure.
Only a few epidemiologic studies evaluated in the 2019 PM ISA have further examined potential
copollutant confounding. There was some evidence that PM2 5 associations with respiratory mortality
remained robust in models with some gaseous pollutants (2019 PM ISA, Section 5.2.10); however, there
was limited assessment of potential copollutant confounding when examining respiratory morbidity
outcomes. The uncertainty related to the independence of PM2 5 effects was partially addressed by
findings of animal toxicological studies. Long-term exposure to PM2 5 resulted in oxidative stress,
inflammation, and morphologic changes in both upper and lower airways (2019 PM ISA, Section 5.2.8),
in addition to the asthma-related and lung development-related effects mentioned above. Epidemiologic
studies examining the effects of declining PM2 5 concentrations provided additional support for a
relationship between long-term PM2 5 exposure and respiratory health by demonstrating improvements in
lung function growth and bronchitic symptoms in children, and improvement in lung function in adults in
association with declining PM2 5 concentrations (2019 PM ISA, Section 5.2.11). However, the limited
examination of copollutant confounding in studies of declining PM2 5 concentrations was a notable
uncertainty given the corresponding decline in other pollutants over the time period of the evaluated
studies.
2.1.1.2 Cardiovascular Effects
Consistent with the conclusions of the 2009 PM ISA, more recently published scientific evidence
further strengthens the conclusion that there is a causal relationship between both short- and long-term
PM2 5 exposure and cardiovascular effects. These causality determinations are based on the consistency of
findings within disciplines; coherence among evidence from controlled human exposure, epidemiologic,
and animal toxicological studies; and evidence supporting biologically plausible pathways for
cardiovascular effects, such as reduced myocardial blood flow, altered vascular reactivity, myocardial
infarctions, and cardiovascular mortality.
2.1.1.2.1 Cardiovascular Effects Associated with Short-Term PM2.5
Exposure
Strong evidence from epidemiologic studies demonstrating associations between cardiovascular
ED visits and hospital admissions in combination with evidence for PM2 5-induced cardiovascular effects
from controlled human exposure and animal toxicological studies confirmed and extended the conclusion
of a causal relationship between short-term PM2 5 exposure and cardiovascular effects from the 2009 PM
ISA (Table 2-2). This conclusion was based on multiple epidemiologic studies demonstrating associations
with cardiovascular effects such as ischemic heart disease (IHD)- and heart failure (HF)-related ED visits
September 2021
2-6
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
and hospital admissions, as well as cardiovascular mortality. The epidemiologic evidence was supported
by experimental studies demonstrating endothelial dysfunction, changes in blood pressure (BP), and
alterations in heart function in response to short-term PM2 5 exposure. Additional evidence from
epidemiologic, controlled human exposure, and animal toxicological studies also provided ample
evidence of biologically plausible pathways by which short-term exposure to PM2 5 can result in overt
cardiovascular effects.
Consistent with the 2009 PM ISA, the strongest evidence comes from epidemiologic studies that
reported consistent positive associations between short-term PM2 5 exposure and cardiovascular-related
ED visits and hospital admissions particularly for IHD (2019 PM ISA, Section 6.1.2.1) and HF (2019 PM
ISA, Section 6.1.3.1), as well as cardiovascular-related mortality (2019 PM ISA, Section 6.1.9) across
studies that used different approaches to control for the potential confounding effects of weather
(e.g., temperature). While associations remained relatively unchanged across the copollutants evaluated,
the evidence was especially consistent for air pollutants that are not typically associated with traffic
(i.e., ozone, SO2, PM10-2.5). In some instances, associations in copollutant models were attenuated, but this
was only observed for the traffic-related pollutants (i.e., NO2, CO), which generally had higher
correlations with PM2 5 than other copollutants. This evidence from copollutant analyses from studies
evaluated in the 2019 PM ISA generally indicates that the associations observed between short-term
PM2 5 exposure and cardiovascular effects are not artifacts due to confounding by another air pollutant
(2019 PM ISA, Section 6.1.14.1). These epidemiologic studies reduce akey uncertainty identified in the
2009 PM ISA by providing evidence that gaseous pollutants are not likely to confound the
PM2 5-cardiovascular effects relationship.
The independence of PM2 5 effects is further addressed by findings of controlled human exposure
and animal toxicological studies evaluated in the 2019 PM ISA. The most consistent evidence from
controlled human exposure studies was for a PM2 5 effect on endothelial function (2019 PM ISA,
Section 6.1.13). Multiple recent controlled human exposure studies reported that PM2 5 impaired some
measure of vessel dilation following reactive hyperemia or pharmacological challenge relative to filtered
air. Given the relationship between endothelial function and BP, these results were coherent with multiple
controlled human exposure studies that reported changes in BP following short-term PM2 5 CAPs
exposure (2019 PM ISA, Section 6.1.6.3). However, these results were inconsistent with some controlled
human exposure studies from previous reviews that did not find changes in endothelial function or BP.
The results of controlled human exposure studies evaluated in the 2019 PM ISA are also coherent with
evidence from animal toxicological studies demonstrating endothelial dysfunction and changes in BP or
the renin angiotensin system following short-term PM25 exposure (2019 PM ISA, Section 6.1.13.3 and
Section 6.1.6.4). Moreover, changes in endothelial function and BP reported in recent experimental
studies were consistent with epidemiologic studies reporting associations between short-term PM2 5
exposure and IHD, as well as with limited epidemiologic panel study evidence of associations with BP. In
addition, animal toxicological studies demonstrating that short-term PM2 5 exposure results in decreased
September 2021
2-7
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
cardiac contractility and changes in left ventricular pressure were coherent with epidemiologic studies
reporting associations between short-term PM2 5 exposure and HF.
Collectively, the evidence from controlled human exposure, animal toxicological, and
epidemiologic panel studies provided a biologically plausible pathway by which short-term PM2 5
exposure could result in cardiovascular effects such as those leading to an ED visit, hospital admission, or
mortality. This proposed pathway (2019 PM ISA, Section 6.1.1) begins with pulmonary inflammation
and/or activation of sensory nerves in the respiratory track and progresses to autonomic nervous system
imbalance and/or systemic inflammation that can potentially affect cardiovascular endpoints such as
endothelial function, HRV, hemostasis, and/or BP. Changes in the aforementioned cardiovascular
endpoints may then lead to the development of arrhythmia, thrombosis, and/or acute myocardial ischemia,
potentially resulting in outcomes such as myocardial infarction, IHD, HF, and possibly death.
Overall, across the scientific disciplines, recent studies extended and supported the previous
evidence for a continuum of cardiovascular-related health effects following short-term exposure to PM2 5.
These effects range from relatively modest increases in biomarkers related to inflammation, to subclinical
cardiovascular endpoints such as endothelial dysfunction, the overt outcomes of ED visits and hospital
admissions, specifically for IHD and HF, and ultimately cardiovascular-related mortality.
2.1.1.2.2 Cardiovascular Effects Associated with Long-Term PM2.5
Exposure
Multiple epidemiologic studies evaluated in the 2019 PM ISA and previous assessments that
extensively control for potential confounders provided strong evidence of positive associations with
cardiovascular mortality, which in combination with supporting evidence from recent studies examining
cardiovascular morbidity reaffirmed the conclusion of a causal relationship between long-term PM2 5
exposure and cardiovascular effects in the 2009 PM ISA (Table 2-2). This conclusion was based on U.S.
and Canadian cohort studies evaluated in the 2019 PM ISA that demonstrated consistent, positive
associations between long-term PM2 5 exposure and cardiovascular mortality, with more limited evidence
from studies examining long-term PM2 5 exposure and cardiovascular morbidity.
Epidemiologic studies consisting of U.S.-based cohorts and subsequent analyses of these cohorts,
provided the basis of the conclusions in the 2009 PM ISA. These studies, in combination with cohort
studies evaluated in the 2019 PM ISA, continued to demonstrate consistent, positive associations and
support a strong relationship between long-term PM2 5 exposure and cardiovascular mortality. The results
of these cohort studies are consistent across various spatial extents, exposure assessment techniques, and
statistical techniques in locations where mean annual average concentrations are near or below 12 (ig/m3
(2019 PM ISA, Section 6.2.10).
The body of literature examining the relationship between long-term PM2 5 exposure and
cardiovascular morbidity has greatly expanded since the 2009 PM ISA. Epidemiologic studies evaluated
September 2021
2-8
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
in the 2019 PM ISA examining cardiovascular morbidity endpoints consisted of several large U.S. cohort
studies each focusing on populations with distinct demographic characteristics (e.g., postmenopausal
women, male doctors, etc.) and extensive consideration of potential confounders. These studies have
reported heterogeneous results, with several studies that adjusted for important confounders, including
socioeconomic status (SES), reporting positive associations for cardiovascular morbidity endpoints. The
strong associations reported between long-term PM2 5 exposure and coronary events (e.g., coronary heart
disease [CHD] and stroke) among postmenopausal women in the Women's Health Initiative (WHI)
cohort, highlighted in 2009 PM ISA, were strengthened in an extended analysis that considered individual
and neighborhood-level SES (2019 PM ISA, Section 6.2.3; Section 6.2.10). Recent analyses of other
cohorts of women (i.e., Nurses' Health Study, California Teachers Study) that were comparable to WHI
in that they considered menopausal status or hormone replacement therapy did not show consistent
positive associations with CHD, myocardial infarction, or stroke. Longitudinal studies demonstrated that
changes in the progression of atherosclerosis in relation to long-term exposure to PM2 5 were variable
across cohorts and found to depend, in part, on the vascular bed in which atherosclerosis was evaluated
(2019 PM ISA, Section 6.2.4.1). However, within a study focusing on the progression of atherosclerosis
in a healthy population, the Multi-Ethnic Study of Atherosclerosis and Air Pollution (MESA-Air), an
association was observed between long-term PM2 5 exposure and coronary artery calcification (CAC),
which is a strong predictor of CHD (2019 PM ISA, Section 6.2.4). A small number of studies reported
positive associations between long-term PM2 5 exposure and HF, BP changes, and hypertension.
Longitudinal epidemiologic analyses also supported the observation of positive associations with markers
of systemic inflammation, coagulation, and endothelial dysfunction. These HF studies were coherent with
animal toxicological studies demonstrating decreased contractility and cardiac output and increased
coronary artery wall thickness following long-term PM2 5 exposure (2019 PM ISA, Section 6.2.4.2).
Moreover, animal toxicological studies finding a relationship between long-term exposure to PM2 5 and
changes in BP in rats and mice were coherent with epidemiologic studies reporting positive associations
between long-term exposure to PM2 5 and hypertension. Similarly, evidence of atherosclerotic plaque
progression in a genetically susceptible mouse model was consistent with epidemiologic studies reporting
associations between atherosclerosis and long-term PM2 5 exposure.
The body of evidence evaluated in the 2019 PM ISA also reduced uncertainties identified in the
2009 PM ISA related to potential copollutant confounding and the shape of the concentration-response
(C-R) relationship for CVD effects following long-term PM2 5 exposure. Generally, most of the PM2 5
effect estimates relating long-term PM2 5 exposure to cardiovascular mortality remained relatively
unchanged or increased in copollutant models adjusted for O3, NO2, SO2, and PM10-2.5 (2019 PM ISA,
Section 6.2.15). In addition, most of the results from analyses examining the C-R function for
cardiovascular mortality supported a linear, no-threshold relationship for cardiovascular mortality,
especially at mean annual PM2 5 concentrations <12 (ig/m3 (2019 PM ISA, Section 6.2.10). Some studies
reported that the slope of the C-R curve tended to be steeper at lower concentrations, especially for IHD
mortality, suggesting a supralinear C-R relationship. A limited number of cardiovascular morbidity
studies examined the shape of the C-R relationship and generally reported a steeper C-R curve at lower
September 2021
2-9
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
concentrations (starting at -10 (.ig/nr1) with the slope of the C-R curve decreasing at higher PM2 5
concentrations (2019 PM ISA, Section 6.2.16).
Evidence from animal toxicological and epidemiologic studies also provided biologically
plausible pathways by which long-term PM2 5 exposure could lead to cardiovascular effects such as CHD,
stroke, and CVD-related mortality (2019 PM ISA, Section 6.2.1). These pathways initially involve
autonomic nervous system changes and/or systemic inflammation that can potentially effect endpoints
related to vascular function, altered hemostasis, hypertension, atherosclerotic plaque progression, and
arrhythmia. Changes in cardiovascular endpoints such as these may then lead to IHD, HF, and possibly
death.
Overall, there was consistent evidence from multiple epidemiologic studies that long-term
exposure to PM2 5 is associated with cardiovascular mortality. Associations with CHD, stroke, and
atherosclerosis progression were observed in several recent epidemiologic studies providing coherence
for PM2 5-related cardiovascular mortality. Results from copollutant models generally support the
independence of PM25 associations. Additional evidence of the direct effect of PM25 on the
cardiovascular system was provided by experimental studies in animals demonstrating effects including
atherosclerosis plaque progression and changes in cardiac contractility and BP.
2.1.1.3 Nervous System Effects
2.1.1.3.1 Nervous System Effects Associated with Long-Term PM2.5
Exposure
The 2009 PM ISA evaluated a small number of animal toxicological studies pertaining to the
effects of long-term exposures to PM2 5 on the nervous system. Since the 2009 PM ISA, the literature base
has greatly expanded with studies evaluated in the 2019 PM ISA providing new information that
strengthens the lines of evidence indicating that long-term PM2 5 exposure may lead to effects on the brain
that are associated with neurodegeneration (i.e., neuroinflammation and reductions in brain volume), as
well as cognitive effects in older adults (Table 2-2). Animal toxicological studies provided evidence for a
range of nervous system effects including neuroinflammation and oxidative stress, neurodegeneration,
cognitive effects, and effects on neurodevelopment. Although the epidemiologic evidence was more
limited in terms of the number of studies conducted, multiple studies generally supported associations
between long-term PM2 5 exposure and changes in brain morphology, cognitive decrements, and dementia
in adult populations. The consistency and coherence of the evidence across disciplines as it relates to
region-specific brain inflammation, morphologic changes in the brain, cognitive effects, and dementia in
adult populations supported that there is a likely to be causal relationship between long-term PM2 5
exposure and nervous system effects. Thus, the expanded evidence base allowed for the first-time, a
causality determination for long-term PM2 5 exposure and nervous system effects.
September 2021
2-10
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
There was strong evidence for biologically plausible pathways that may underlie nervous system
effects resulting from long-term exposure to PM2 5. Studies demonstrated modulation of the autonomic
nervous system leading to downstream consequences including cardiovascular effects (2019 PM ISA,
Section 6.2.1). In addition, the pathway involving neuroinflammation in specific regions of the brain
(i.e., the hippocampus, cerebral cortex, and hypothalamus) and morphologic changes in the brain
indicative of neurodegeneration, is well substantiated and coherent across animal toxicological and
epidemiologic studies (2019 PM ISA, Section 8.2.3 and Section 8.2.4). Specifically, morphologic changes
induced in the hippocampus of animals were accompanied by impaired learning and memory and there
was consistent evidence from multiple epidemiologic studies that long-term PM2 5 exposure is associated
with reduced cognitive function (2019 PM ISA, Section 8.2.5). Further, the presence of early markers of
Alzheimer's disease pathology was demonstrated in animals following long-term exposure to PM2 5
CAPs, which was consistent with a small number of epidemiologic studies that reported positive
associations with neurodegenerative changes in the brain (i.e., decreased brain volume) and Alzheimer's
disease or all-cause dementia (2019 PM ISA, Section 8.2.6). Although the loss of dopaminergic neurons
in the substantia nigra, which is a hallmark of Parkinson's disease, was demonstrated in animals
(2019 PM ISA, Section 8.2.4), epidemiologic studies did not report associations with Parkinson's disease
(2019 PM ISA, Section 8.2.6). Overall, the lack of consideration of copollutant confounding introduces
some uncertainty in the interpretation of the epidemiologic studies but this uncertainty was addressed, in
part, by the direct evidence of effects provided by animal toxicological studies.
In addition to the findings described above, which are mostly relevant to adults, several recent
studies of neurodevelopmental effects in children have also been conducted. Positive associations
between long-term exposure to PM2 5 during the prenatal period and autism spectrum disorder (ASD)
were consistently observed across multiple epidemiologic studies (2019 PM ISA, Section 8.2.7.2).
However, several studies of performance on tests of cognitive function provided little support for an
association. Overall, these epidemiologic studies of developmental effects were limited due to their lack
of control for potential confounding by copollutants, the small number of studies, and uncertainty
regarding critical exposure windows. A study in animals that found inflammatory and morphologic
changes in the corpus collosum and hippocampus, as well as ventriculomegaly in young animals
following prenatal exposure to PM2 5 CAPs provided initial evidence indicating a potential biologically
plausible pathway for a relationship between PM2 5 and ASD.
2.1.1.4 Cancer
2.1.1.4.1 Cancer Associated with Long-Term PM2.5 Exposure
Experimental and epidemiologic evidence indicating genotoxicity, epigenetic effects (e.g., DNA
methylation), and increased carcinogenic potential due to PM2 5 exposure, along with strong
September 2021
2-11
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
epidemiologic evidence for increases in lung cancer incidence and mortality, supported a likely to be
causal relationship between long-term PM2 5 exposure and cancer (Table 2-2). This causality
determination represented a change from the suggestive of a causal relationship4 determination reported
in the 2009 PM ISA. The evidence base underlying this conclusion encompasses the decades of research
on whole PM exposures and research evaluated in the 2019 PM ISA focusing specifically on PM2 5.
PM2 5 exhibits various characteristics of carcinogens, as shown in studies demonstrating
genotoxic effects (e.g., DNA damage), epigenetic alterations, oxidative stress, and electrophilicity. The
examination of the role of PM2 5 in cancer development has often focused on whether whole PM, not
specific size fractions, has mutagenic properties and whether exposure to whole PM results in
genotoxicity or carcinogenicity. Additionally, it has been well characterized that some components of
PM2 5, specifically hexavalent chromium, nickel, arsenic, and PAHs are known human carcinogens.
Extensive analyses of PM25 and PM25 extracts in the Ames Salmonella/mammalian-microsome
mutagenicity assay demonstrated that PM2 5 contains mutagenic agents (2019 PM ISA, Section 10.2.2.1).
Additional in vitro and in vivo toxicological studies indicated the potential for PM2 5 exposure to result in
DNA damage, which was supported by limited human evidence (2019 PM ISA, Section 10.2.2.2). Some
studies have also demonstrated that PM2 5 exposure can result in cytogenetic effects, specifically
micronuclei formation and chromosomal aberrations (2019 PM ISA, Section 10.2.2.3), as well as
differential expression of genes potentially relevant to genotoxicity or other aspects of cancer
pathogenesis (2019 PM ISA, Section 10.2.2.4). Although inconsistently examined across studies, changes
in cellular and molecular markers of genotoxicity and epigenetic alterations, which may lead to genomic
instability, are demonstrated in response to PM2 5 exposure. Further, the carcinogenic potential of PM2 5
was demonstrated in an animal toxicological study in which chronic inhalation enhanced tumor formation
that was initiated by exposure to urethane (2019 PM ISA, Section 10.2.4). Additionally, epidemiologic
studies evaluated in the 2019 PM ISA encompassing multiple cohorts that are diverse in terms of both
geographic coverage and population characteristics have provided evidence of primarily consistent
positive associations between long-term PM2 5 exposure and lung cancer incidence and mortality,
particularly in never smokers (2019 PM ISA, Section 10.2.5.1). Experimental and epidemiologic evidence
of genotoxicity, epigenetic effects, and carcinogenic potential provides biological plausibility for
epidemiologic results of lung cancer incidence and mortality. Although evaluated in a limited number of
studies, the assessment of potential copollutant confounding, particularly with O3, indicated that PM2 5
associations with lung cancer incidence and mortality are relatively unchanged in copollutant models
(2019 PM ISA, Section 10.2.5.1.3). There was limited evidence that long-term PM2 5 exposure is
associated with cancers in other organ systems; however, there was initial evidence that PM2 5 exposure
may reduce survival in individuals with cancer.
4Since the 2009 PM ISA, the causality determination language has been updated and this category is now stated as
suggestive of, but not sufficient to infer, a causal relationship.
September 2021
2-12
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
2.1.1.5 Mortality
Consistent with the conclusions of the 2009 PM ISA, evidence from studies evaluated in the 2019
PM ISA reaffirmed and further strengthened that there is a causal relationship between both short- and
long-term PM2 5 exposure and total mortality. These causality determinations were based on the
consistency of findings across a large body of epidemiologic studies. Evidence from controlled human
exposure, epidemiologic, and animal toxicological studies of respiratory and cardiovascular morbidity
also provided coherence, as well as biological plausibility. Together, the consistency and coherence in the
evidence collectively supported a continuum of effects by which short- and long-term PM2 5 exposure
could result in mortality.
2.1.1.5.1 Mortality Associated with Short-Term PM2.5 Exposure
Strong epidemiologic evidence from studies evaluated in the 2019 PM ISA as well as previous
assessments that examined total (nonaccidental) mortality in combination with evidence for
cause-specific respiratory and cardiovascular mortality continued to support the conclusion of the 2009
PM ISA that there is a causal relationship between short-term PM2 5 exposure and total (nonaccidental)
mortality (Table 2-2). This conclusion was based on multiple recent multicity studies conducted in the
U.S., Canada, Europe, and Asia that continued to provide evidence of consistent, positive associations
between short-term PM2 5 and total mortality, across studies that used different approaches to control for
the potential confounding effects of weather (e.g., temperature). In addition, there was evidence of
biological plausibility for cause-specific mortality and ultimately total mortality as demonstrated by the
consistent and coherent evidence across scientific disciplines for cardiovascular morbidity, particularly
ischemic events and HF (2019 PM ISA, Chapter 6), and respiratory morbidity, with the strongest
evidence coming from studies of exacerbations of COPD and asthma (2019 PM ISA, Chapter 5).
Multicity studies evaluated in the 2019 PM ISA added to the body of evidence evaluated in the
2009 PM ISA and continued to support a positive association between short-term PM2 5 exposure and
total mortality with percentage increases in mortality ranging from 0.19-2.80% at lags of 0 to 1 day in
studies in which mean 24-hour avg concentrations were primarily <20 (ig/m3 (2019 PM ISA, Figure 11-1;
Table 11-1).5 The positive associations observed across studies reflected traditional analyses using
ambient monitors as well as analyses conducted in both urban and rural locations that used new exposure
assignment techniques and relied on multiple sources of PM2 5 data (e.g., ambient monitors, statistical
models, and satellite data). Whereas the analysis of potential copollutant confounding was limited to
single-city studies and studies of PM10 in the 2009 PM ISA, recent multicity studies conducted in Europe
and Asia indicated that PM2 5-mortality associations were relatively unchanged in copollutant models with
throughout this supplement, as detailed in the Preface of the 2019 PM ISA (Section P.3.2.2), risk estimates from
epidemiologic studies examining short-term exposures are for a 10-|ig/m3 increase in 24-hour avg PM2 5
concentrations and long-term exposures are for a 5-|ig/m3 increase in annual concentrations, unless otherwise noted.
September 2021
2-13
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
gaseous pollutants and PM10-2.5 (2019 PM ISA, Section 11.1.4). These results from copollutant models
further supported an independent effect of PM2 5 on mortality. The associations reported for total
mortality were also supported by analyses demonstrating increases in cause-specific mortality,
specifically for cardiovascular and respiratory mortality which comprise -33 and -9%, respectively, of
total mortality [NHLBI (2017); Figure 11-2], The consistent and coherent evidence across scientific
disciplines for cardiovascular morbidity, particularly ischemic events and HF (2019 PM ISA, Chapter 6),
and to a lesser degree for respiratory morbidity, with the strongest evidence for exacerbations of COPD
and asthma (2019 PM ISA, Chapter 5), provided biological plausibility for cause-specific mortality and
ultimately total mortality. The relationship between short-term PM2 5 exposure and total mortality was
additionally supported by analyses of the concentration-response (C-R) relationship. Although
alternatives to linearity have not been systematically evaluated, mortality studies evaluated in the 2019
PM ISA continued to support a linear, no-threshold C-R relationship (2019 PM ISA, Section 11.1.10).
2.1.1.5.2 Mortality Associated with Long-Term PM2.5 Exposure
Strong epidemiologic evidence from cohorts in the U.S., Canada, and Europe evaluated in the
2019 PM ISA, as well as previous assessments, continued to support the conclusion of the 2009 PM ISA
that there is a causal relationship between long-term PM2 5 exposure and total mortality (Table 2-2). This
conclusion was based on the evaluation of multiple cohorts that continued to provide evidence of
consistent, positive associations, across studies that controlled for a range of individual- and ecological
covariates, such as smoking status and SES. Additional evidence indicated coherence of effects across
scientific disciplines for cardiovascular and respiratory morbidity and metabolic disease, which provided
biological plausibility for cause-specific mortality and supported a causal relationship with total
mortality.
Additional reanalyses and extensions of the American Cancer Society (ACS) and Harvard Six
Cities (HSC) cohorts as well as new cohorts consisting of Medicare participants, people that live in
Canada, or people employed in a specific job (e.g., teacher, nurse, etc.) provided further evidence of
positive associations between long-term PM2 5 exposure and total mortality, particularly in areas with
annual mean concentrations <20 (ig/m3, and in some cases below 12 (ig/m3 (2019 PM ISA, Figure 11-17
and Figure 11-18). Across studies, positive associations were consistently observed regardless of the
exposure assignment approach employed, with some studies relying on ambient monitors while others
using modeled or remote sensing data or hybrid methods that combine two or more data sources. Recent
studies have conducted analyses to examine potential copollutant confounding and indicated that
associations between long-term PM2 5 exposure and total mortality are relatively unchanged in copollutant
models, particularly for O3, with fewer studies examining NO2, and PM10-2.5 (2019 PM ISA,
Section 11.2.3; Figure 11-20, Figure 11-21). The evidence for total mortality was further supported by
analyses of cause-specific mortality, which reported positive associations with cardiovascular, respiratory,
and lung cancer mortality. Biological plausibility for mortality due to long-term PM2 5 exposure was
September 2021
2-14
DRAFT: Do Not Cite or Quote
-------
1 provided by the coherence of effects across scientific disciplines for cardiovascular morbidity,
2 particularly for CHD, stroke and atherosclerosis, and for respiratory morbidity, particularly for the
3 development of COPD. Recent studies extensively examined the C-R relationship between long-term
4 PM2 5 exposure and total mortality, specifically in several U.S. and Canadian cohorts, and collectively
5 continued to support a linear, no-threshold C-R relationship (2019 PM ISA, Section 11.2.4; Table 11-7).
6 A series of studies evaluated in the 2019 PM ISA, examined the relationship between long-term
7 exposure to PM2 5 and mortality by examining the temporal trends in PM2 5 concentrations to test the
8 hypothesis that decreases in PM2 5 concentrations are associated with increases in life expectancy (2019
9 PM ISA, Section 11.2.2.5). These studies reported that decreases in long-term PM2 5 concentrations were
10 associated with an increase in life expectancy across the U.S. for the multiple time periods examined.
September 2021
2-15
DRAFT: Do Not Cite or Quote
-------
Table 2-2 Key evidence contributing to causal and likely to be causal causality determinations for PM2.5
exposure and health effects evaluated in the 2019 Integrated Science Assessment for Particulate
Matter.
Key Evidence in
2019 PM ISA Health Effect Category3 and Causality Determination PM2.5 Concentrations Associated with Effects
Respiratory Effects and Short-Term PM2.5 Exposure (2019 PM ISA, Section 5.1): Likely to be Causal Relationship
No Change in Causality Determination from the 2009 PM ISA; New Evidence Further Supports the Previous Determination.
Section 5.1.12 Epidemiologic evidence, consisting mainly of ED visits and hospital
Table 5-18 admissions, strongly supported a relationship with asthma
exacerbation, COPD exacerbation, and combinations of
respiratory-related diseases. Evidence for associations with respiratory
symptoms and medication use are coherent with other findings for
asthma and COPD exacerbation. Some epidemiologic studies
examined copollutant confounding and reported that results are robust
in models with gaseous pollutants (i.e., O3, NO2, SO2, and with more
limited evidence for CO) and other particle sizes (i.e., PM10-2.5),
especially for asthma exacerbation, combinations of respiratory-related
ED visits and hospital admissions, and respiratory mortality. There was
a large body of experimental evidence demonstrating respiratory effects
due to short-term PM2.5 exposure. These experimental studies provided
evidence for biologically plausible pathways by which PM2.5 exposure
could cause a respiratory effect. Specifically, animal toxicological
studies provided biological plausibility for asthma exacerbation, COPD
exacerbation, and respiratory infection with some evidence of an
independent effect of PM2.5 on respiratory endpoints. Controlled human
exposure studies provided minimal evidence of respiratory effects such
as altered lung function and pulmonary inflammation. Consistent
positive associations with respiratory mortality provide evidence of a
continuum of effects.
Mean ambient concentrations from epidemiologic studies for:
Hospital admissions and ED visits for asthma, COPD, respiratory
infections, and combinations of respiratory-related diseases:
U.S. and Canada: 4.7-24. 6 |jg/m3
Europe: 8.8-27.7 |jg/m3
Asia: 11.8-69.9 |jg/m3
Respiratory mortality:
U.S. and Canada: 7.9-19.9 |jg/m3
Europe: 8.0-27.7 |jg/m3
Asia: 11.8-69.9 |jg/m3
Concentrations from animal toxicological studies for:
Allergic airway disease:
442-596 |jg/m3
COPD: 171-1,200 pg/m3
Altered host defense:
100-350 |jg/m3
September 2021
2-16
DRAFT: Do Not Cite or Quote
-------
Table 2-2 (Continued): Key evidence contributing to causal and likely to be causal causality determinations for
PM2.5 exposure and health effects evaluated in the 2019 Integrated Science Assessment
for Particulate Matter.
Key Evidence in
2019 PM ISA
Health Effect Category3 and Causality Determination
PM2.5 Concentrations Associated with Effects
Respiratory Effects and Long-Term PM2.5 Exposure (2019 PM ISA, Section 5.2): Likely to be Causal Relationship
No Change in Causality Determination from the 2009 PM ISA; New Evidence Further Supports the Previous Determination.
Section 5.2.13 Epidemiologic evidence strongly supported a relationship with
Table 5-27 decrements in lung function growth in children. Additional epidemiologic
evidence supported a relationship with asthma development in children,
increased bronchitic symptoms in children with asthma, acceleration of
lung function decline in adults, and respiratory mortality, including
cause-specific respiratory mortality for COPD and respiratory infection.
Some epidemiologic studies examined copollutant confounding and
reported that results are robust in models with O3, NO2, and CO,
especially for respiratory mortality. There was limited experimental
evidence for respiratory effects from long-term PM2.5 exposure.
However, animal toxicological studies provided biological plausibility for
decrements in lung function and asthma development in children, and
they reduced the uncertainty regarding the independent effect of PM2.5
for these endpoints. Animal toxicological studies also provided evidence
for a wide variety of other subclinical effects, such as oxidative stress,
inflammation, and morphologic changes. Epidemiologic studies
examining the effects of declining PM2.5 concentrations strengthened
the relationship between long-term PM2.5 exposure and respiratory
health by demonstrating improvements in lung function growth and
reduced bronchitic symptoms in children and improved lung function in
adults as a result of lower PM2.5 concentrations. However, these studies
have a limited examination of copollutant confounding, which was a
notable uncertainty because concentrations of other pollutants have
also declined.
Mean ambient concentrations from epidemiologic studies for:
Decrement in lung function growth:
6-28 |jg/m3
Asthma development in children:
5.2-16.5 |jg/m3
Bronchitic symptoms in children with asthma:
9.9-13.8 |jg/m3
Accelerated lung function decline in adults:
9.5-17.8 |jg/m3
Respiratory mortality:
6.3-23.6 |jg/m3
Concentrations from animal toxicological studies for:
Impaired lung development:
16.8 |jg/m3
Development of allergic airway disease:
100 |jg/m3
September 2021
2-17
DRAFT: Do Not Cite or Quote
-------
Table 2-2 (Continued): Key evidence contributing to causal and likely to be causal causality determinations for
PM2.5 exposure and health effects evaluated in the 2019 Integrated Science Assessment
for Particulate Matter.
Key Evidence in
2019 PM ISA
Health Effect Category3 and Causality Determination
PM2.5 Concentrations Associated with Effects
Cardiovascular Effects and Short-Term PM2.5 Exposure (2019 PM ISA, Section 6.1): Causal Relationship
No Change in Causality Determination from the 2009 PM ISA; New Evidence Further Supports the Previous Determination.
Section 6.1.16 There was strong evidence for coherence of effects across scientific
Table 6-34 disciplines and biological plausibility for a range of cardiovascular
effects in response to short-term PM2.5 exposure. Consistent
epidemiologic evidence from multiple studies at relevant PM2.5
concentrations provided evidence of increases in ED visits and hospital
admissions for IHD and HF, as well as cardiovascular mortality in
multicity studies conducted in the U.S., Canada, Europe, and Asia.
These associations remained positive, but in some cases were reduced
with larger uncertainty estimates, in copollutant models with gaseous
pollutants. Controlled human exposure studies provided coherence and
consistent evidence for changes in various measures of endothelial
dysfunction and generally consistent evidence of changes in BP. These
controlled human exposure studies were consistent with animal
toxicological studies also demonstrating endothelial dysfunction, as well
as changes in BP and the renin-angiotensin system. In addition, animal
toxicological studies demonstrating that short-term PM2.5 exposure
results in decreased cardiac contractility and left ventricular pressure
were coherent with epidemiologic studies reporting associations
between short-term PM2.5 exposure and HF.
Mean ambient concentrations from epidemiologic studies for:
IHD: 5.8-18.6 pg/m3
HF: 5.8-18.0 pg/m3
Concentrations from controlled human exposure studies:
24-325 pg/m3 for 2 h
Concentrations from animal toxicological studies:
178-190 pg/m3
September 2021
2-18
DRAFT: Do Not Cite or Quote
-------
Table 2-2 (Continued): Key evidence contributing to causal and likely to be causal causality determinations for
PM2.5 exposure and health effects evaluated in the 2019 Integrated Science Assessment
for Particulate Matter.
Key Evidence in
2019 PM ISA Health Effect Category3 and Causality Determination PM2.5 Concentrations Associated with Effects
Cardiovascular Effects and Long-Term PM2.5 Exposure (2019 PM ISA, Section 6.2): Causal Relationship
No Change in Causality Determination from the 2009 PM ISA; New Evidence Further Supports the Previous Determination.
Section 6.2.18 Multiple epidemiologic studies continued to provide evidence of
Table 6-54 consistent, positive associations between long-term PM2.5 exposure
and cardiovascular mortality at lower ambient concentrations. The
cardiovascular mortality associations were observed across different
exposure assignment and statistical methods and were relatively
unchanged in copollutant models with both gaseous (i.e., O3, NO2, SO2)
and particulate (i.e., PM10-2.5) pollutants. The evidence for
cardiovascular mortality was supported by a smaller body of
epidemiologic studies that further explored associations between
long-term PM2.5 exposure and cardiovascular morbidity. These studies
reported some evidence for increased risk of PM2.5-related Ml and
stroke, specifically in individuals with a preexisting cardiovascular
disease or diabetes. Recent epidemiologic studies also presented
evidence for an effect of long-term PM2.5 exposure on subclinical
features of cardiovascular morbidity, particularly progression of
atherosclerosis as reflected by associations with CAC, with more limited
evidence for other measures, such as cIMT. Key evidence from animal
toxicological studies included consistent evidence for changes in BP, as
well as some evidence for decreases in measures of heart function
(e.g., contractility and cardiac output) and cardiac remodeling.
Moreover, as in the previous review, there was also some additional
evidence for atherosclerotic plaque progression in a genetically
susceptible mouse model.
Mean ambient concentrations from epidemiologic studies for:
Cardiovascular mortality.
4.1-17.9 |jg/m3
Coronary events'.
13.4 |jg/m3
CAC:
14.2 |jg/m3
CHD and stroke (in those with preexisting disease):
13.4-23.9 |jg/m3
Concentrations from animal toxicological studies for:
BP:
85-375 |jg/m3 (up to 15 weeks)
September 2021
2-19
DRAFT: Do Not Cite or Quote
-------
Table 2-2 (Continued): Key evidence contributing to causal and likely to be causal causality determinations for
PM2.5 exposure and health effects evaluated in the 2019 Integrated Science Assessment
for Particulate Matter.
Key Evidence in
2019 PM ISA
Health Effect Category3 and Causality Determination
PM2.5 Concentrations Associated with Effects
Nervous System Effects and Long-Term PM2.5 Exposure (2019 PM ISA, Section 8.2): Likely to be Causal Relationship
Not Evaluated in the 2009 PM ISA; New Evidence Showing Brain Inflammation and Oxidative Stress, Neurodegeneration, Cognitive Effects, and
Neurodevelopmental Effects.
Section 8.2.9 There was evidence that long-term exposure to PM2.5 can modulate the
Table 8-20 autonomic nervous system leading to downstream consequences,
including cardiovascular effects (2019 PM ISA, Section 6.2.1). A
second pathway involving neuroinflammation and morphologic changes
in the brain indicative of neurodegeneration is well substantiated and
coherent across animal toxicological and epidemiologic studies. This
combination of evidence supported PIVh s-related reductions in brain
volume and cognitive effects in older adults. The evidence relating to
Parkinson's disease, and neurodevelopmental effects was more limited.
Consideration of copollutant confounding was generally lacking in the
epidemiologic studies, but the uncertainty in interpreting the study
findings was partly addressed by the direct evidence of effects provided
by animal toxicological studies.
Mean annual concentrations from epidemiologic studies for:
Brain volume:
11.1-12.2 |jg/m3
Cognition:
8.5 (5-yr avg)-14.9 |jg/m3
Autism:
14.0-19.6 |jg/m3
Concentrations from animal toxicological studies for:
Brain inflammation/oxidative stress:
65.7-441.7 |jg/m3
Neurodegenerative changes:
94.4 |jg/m3
Neurodevelopment:
92.7 |jg/m3
September 2021
2-20
DRAFT: Do Not Cite or Quote
-------
Table 2-2 (Continued): Key evidence contributing to causal and likely to be causal causality determinations for
PM2.5 exposure and health effects evaluated in the 2019 Integrated Science Assessment
for Particulate Matter.
Key Evidence in
2019 PM ISA Health Effect Category3 and Causality Determination PM2.5 Concentrations Associated with Effects
Cancer and Long-Term PM2.5 Exposure (2019 PM ISA, Section 10.2): Likely to be Causal Relationship
Change in Causality Determination From the 2009 PM ISA (Suggestive of a Causal Relationship) Due to Increased Evidence of Genotoxicity, Carcinogenicity,
and Epigenetic Effects for PM2.5 and Lung Cancer Incidence and Mortality.
Section 10.2.7 Primarily positive associations from multiple epidemiologic studies
Table 10-8 reported increases in the risk of lung cancer incidence and mortality.
This evidence was supported by analyses focusing on never smokers
and limited evidence of associations with histological subtypes of lung
cancer found in never smokers. Across studies that examined lung
cancer incidence and mortality, potential confounding by smoking
status and exposure to SHS was adequately controlled. A limited
number of studies examined potential copollutant confounding, but
associations were relatively unchanged in models with O3 with more
limited assessment of other gaseous pollutants and particle size
fractions. Experimental and epidemiologic studies provided evidence for
a relationship between PM2.5 exposure and genotoxicity, epigenetic
effects, and carcinogenic potential. Uncertainties exist due to the lack of
consistency in specific cancer-related biomarkers associated with PM2.5
exposure across both experimental and epidemiologic studies;
however, PM2.5 exhibits several characteristics of carcinogens, which
provided biological plausibility for PM2.5 exposure contributing to cancer
development. Additionally, there was limited evidence of cancer
occurring in other organ systems, but there was some evidence that
PM2.5 exposure may detrimentally affect survival from any type of
cancer.
Mean annual concentrations from epidemiologic studies for:
Lung cancer incidence and mortality.
U.S. and Canada:
6.3-23.6 |jg/m3
Europe:
6.6-31.0 |jg/m3
Asia:
33.7 |jg/m3
Concentrations from animal toxicological studies for:
Carcinogenic potential:
17.66 |jg/m3
September 2021
2-21
DRAFT: Do Not Cite or Quote
-------
Table 2-2 (Continued): Key evidence contributing to causal and likely to be causal causality determinations for
PM2.5 exposure and health effects evaluated in the 2019 Integrated Science Assessment
for Particulate Matter.
Key Evidence in
2019 PM ISA Health Effect Category3 and Causality Determination PM2.5 Concentrations Associated with Effects
Total Mortality and Short-Term PM2.5 Exposure (2019 PM ISA, Section 11.1): Causal Relationship
No Change in Causality Determination From the 2009 PM ISA; New Evidence Further Supports the Previous Determination.
Section 11.1.12 There was consistent epidemiologic evidence from multiple multicity
Table 11-4 studies conducted in the U.S., Canada, Europe, and Asia for increases
in total (nonaccidental) mortality at ambient concentrations, often below
20 |jg/m3. The associations observed were relatively unchanged in
copollutant models with gaseous pollutants and PM10-2.5, which was
consistent with copollutant analyses for cardiovascular and respiratory
mortality, but copollutant analyses were limited to studies conducted in
Europe and Asia. Biological plausibility for the epidemiologic evidence
for total mortality was provided by the strong cardiovascular morbidity
evidence, particularly for ischemic events and HF, while support for
biological plausibility was more limited from the respiratory morbidity
evidence, with the strongest evidence for exacerbations of COPD and
asthma. Although alternatives to linearity have not been systematically
evaluated, recent mortality studies continued to support a linear,
no-threshold C-R relationship.
Mean 24-h avg concentrations from epidemiologic studies for:
Total mortality.
U.S. and Canada:
4.37-17.97 |jg/m3
Europe:
13-27.7 |jg/m3
Asia:
11.8-69.9 |jg/m3
September 2021
2-22
DRAFT: Do Not Cite or Quote
-------
Table 2-2 (Continued): Key evidence contributing to causal and likely to be causal causality determinations for
PM2.5 exposure and health effects evaluated in the 2019 Integrated Science Assessment
for Particulate Matter.
Key Evidence in
2019 PM ISA Health Effect Category3 and Causality Determination PM2.5 Concentrations Associated with Effects
Total Mortality and Long-Term PM2.5 Exposure (2019 PM ISA, Section 11.2): Causal Relationship
No Change in Causality Determination From the 2009 PM ISA; New Evidence Further Supports the Previous Determination.
Section 11.2.7 There was consistent epidemiologic evidence from multiple studies
Table 11-8 reporting increases in the risk of total (nonaccidental) mortality from
extended follow-ups of the ACS cohort and HSC cohort, as well as
multiple studies focusing on a Medicare cohort, Canadian cohorts, and
North American employment cohorts. The consistent increases in total
mortality were observed across different exposure metrics based on
ambient measurements, models, remote sensing, or hybrid methods
that combine two or more of these methods, providing additional
support for the mortality associations due to long-term PM2.5 exposure
reported in the 2009 PM ISA that relied on exposure metrics from
ambient monitors. The consistent epidemiologic evidence for total
mortality was supported by positive associations for cardiovascular,
respiratory, and lung cancer mortality. Biological plausibility for total
mortality was provided by the strong cardiovascular morbidity evidence
particularly for CHD, stroke, and atherosclerosis, while there is more
limited evidence for biological plausibility from the respiratory morbidity
evidence, with some evidence for development of COPD. Extensive
epidemiologic evidence provides additional support for a linear,
no-threshold C-R relationship. A series of studies demonstrated that
decreases in long-term PM2.5 concentrations were associated with an
increase in life expectancy across the U.S. for multiple time periods
examined.
ACS = American Cancer Society; avg = average; BP = blood pressure; CAC = coronary artery calcification; CHD = coronary heart disease; cIMT = carotid intima-media thickness;
CO = carbon monoxide; COPD = chronic obstructive pulmonary disease; C-R = concentration-response; h = hour; HF = high frequency; HSC = Harvard Six Cities; IHD = ischemic
heart disease; |jg/m3 = micrograms per cubic meter; Ml = myocardial infarction; N02 = nitrogen dioxide; 03 = ozone; PM25 = particulate matter with a nominal mean aerodynamic
diameter less than or equal to 2.5 |jm; PM10-25 = particulate matter with a nominal mean aerodynamic diameter greater than 2.5 |jm and less than or equal to 10 |jm;
SHS = second-hand smoke; S02 = sulfur dioxide.
aA large spectrum of outcomes is evaluated as part of a broad health effect category including physiological measures (e.g., airway responsiveness, lung function), clinical outcomes
(e.g., respiratory symptoms, hospital admissions), and cause-specific mortality. Total mortality includes all nonaccidental causes of mortality and is informed by the nature of the
evidence for the spectrum of morbidity effects (e.g., respiratory, cardiovascular) that can lead to mortality. The sections and tables referenced in the 2019 PM ISA include a detailed
discussion of the available evidence that informed the causality determinations.
Mean annual concentrations from epidemiologic studies for:
Total mortality.
ACS/HSC cohorts:
11.4-23.6 |jg/m3
Medicare cohort:
8.12-12.0 |jg/m3
Canadian cohorts:
8.7-9.1 |jg/m3
Employment cohorts:
12.7-17.0 |jg/m3
September 2021
2-23
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
2.2
Policy-Relevant Considerations
In the process of evaluating the current state of the science with respect to the effect of short- and
long-term PM exposure on health, studies were identified and evaluated within the 2019 PM ISA that
conducted analyses focused on addressing some of the main policy-relevant questions of this review, as
detailed in the PM IRP (U.S. EPA. 2016). such as:
• Is there new evidence aimed at disentangling the effect of PM from the complex air pollution
mixture to inform a direct effect of PM on health, specifically the assessment of potential
copollutant confounding?
• Is there new evidence to inform the current indicators (i.e., PM2 5 for fine particles and PM10 for
thoracic coarse particles), averaging times (i.e., 24-hour avg, annual average), and levels of the
PM NAAQS?
• Is there new evidence on the shape of the C-R relationship and whether a threshold hold exists
between PM exposure and various health outcomes (e.g., mortality, hospital admissions, etc.),
especially for concentrations near or below the levels of the current PM NAAQS?
• Is there new evidence that individual PM component(s) or source(s) (e.g., industrial facilities,
roads, atmospheric formation), are more strongly associated with health effects than PM mass,
particularly for health effects for which there is sufficient evidence of a strong relationship
(e.g., cardiovascular effects, mortality) with PM exposure?
• Is there new evidence indicating that specific populations or lifestages are at increased risk of a
PM-related health effect compared with a referent population?
The following sections summarize the evidence that can inform consideration of these
policy-relevant questions, specifically: potential copollutant confounding (Section 2.2.1). timing of effects
(Section 2.2.2). C-R relationship (Section 2.2.3). PM components and sources (Section 2.2.4). and
populations potentially at increased risk of a PM-related health effect (Section 2.2.52.2.5).
2.2.1 Potential Copollutant Confounding
Studies evaluated in the 2019 PM ISA further examined the potential confounding effects of
copollutants, both gaseous and particulate, on the relationship between short- and long-term PM2 5
exposure and health effects. These studies built upon the evidence detailed in the 2009 PM ISA and
continued to provide evidence indicating that associations with PM2 5 are relatively unchanged in
copollutant models. Evidence from epidemiologic studies, in combination with experimental studies
detailed in multiple chapters of the 2019 PM ISA (i.e., "Respiratory Effects"—Chapter 5 and
"Cardiovascular Effects"—Chapter 6 within the 2019 PM ISA) that examined exposure to PM
(e.g., CAPs, resuspended PM, and whole mixtures in the presence and absence of a particle trap),
demonstrate a direct effect of PM on health.
September 2021
2-24
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
2.2.1.1 Short-Term PM2.5 Exposure
Building upon the studies evaluated in the 2009 PM ISA, epidemiologic studies evaluated in the
2019 PM ISA have further examined whether copollutants confound associations between short-term
PM2 5 exposure and respiratory and cardiovascular effects and mortality. These studies continued to
demonstrate that PM2 5-associations are relatively unchanged in copollutant models with both gaseous
(i.e., O3, NO2, SO2, and CO) and particulate (i.e., PM10-2.5) pollutants.
The examination of potential copollutant confounding on the relationship between short-term
PM2 5 exposure and respiratory effects has been assessed most extensively through studies examining
respiratory-related ED visits and hospital admissions, particularly for asthma, with more limited
assessments of COPD and respiratory infection, and studies examining respiratory mortality
(Section 5.1.10.1). Correlations between PM2 5 and gaseous and particulate pollutants varied across
studies, with low to moderate correlations (i.e., <0.7) observed for NO2, SO2, CO, and PM10-2.5, and
correlations spanning from low to high for O3. Across the studies that assessed copollutant confounding,
O3 was most commonly examined, followed by NO2 and PM2 5 results were relatively unchanged in
copollutant models. Although fewer studies focused on SO2 and CO, the results from copollutant analyses
were consistent with studies evaluated in the 2009 PM ISA, indicating that results are relatively
unchanged in copollutant models. Studies evaluated in the 2019 PM ISA that examined PM10-2.5 further
expanded upon the initial results detailed in the 2009 PM ISA, and although results are consistent with
observations from analyses of gaseous pollutants, there is greater uncertainty in these results due to the
different methods employed across studies to estimate PM10-2.5 concentrations.
For cardiovascular effects, moderate to strong correlations were reported for NO2 and CO, with
low to moderate correlations for O3, SO2, and PM10-2.5. Across studies of various cardiovascular-related
ED visits and hospital admissions and cardiovascular mortality, results were relatively unchanged in
copollutant models, but there were some instances of attenuation of the PM2 5 association in models with
NO2 and CO (2019 PM ISA, Section 6.1.14.1). Overall, there was no observed difference in the trend or
pattern of copollutant model results across cardiovascular endpoints (e.g., aggregate CVD endpoints,
IHD, HF, cardiovascular mortality). However, the few instances of attenuation were with traffic-related
pollutants (i.e., NO2, CO), which generally had higher correlations with PM2 5 than the other copollutants.
As a result, it was difficult to distinguish whether the instances of observed attenuation in PM2 5
associations were due to confounding or collinearity with other pollutants.
Most epidemiologic studies evaluated in the 2019 PM ISA that examined the potential
confounding effects of copollutants focused on respiratory and cardiovascular effects; only a few focused
on mortality (2019 PM ISA, Section 11.1.4). Recent multicity studies conducted in Europe and Asia
supported the results from single- and multicity studies examined in the 2004 PM AQCD and 2009 PM
ISA that reported limited evidence of confounding by copollutants. Across studies that examined both
gaseous and particulate (i.e., PM10-2.5) pollutants, low to moderate correlations were reported with PM2 5.
September 2021
2-25
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
Associations with PM2 5 were relatively unchanged in copollutant models across the various study
locations examined.
In addition to conducting traditional copollutant analyses, epidemiologic studies of respiratory
(2019 PM ISA, Section 5.1.10.1.1) and cardiovascular (2019 PM ISA, Section 6.1.14.1.1) effects have
also examined the role of PM within the broader air pollution mixture. These studies do not inform
whether PM is independently associated with a respiratory effect, but they can assess whether days with
higher PM2 5 concentrations are more closely related to health effects. Studies of respiratory effects
demonstrated that days when the air pollution mixture has high PM2 5 concentrations often represented the
days with the largest associations (in terms of magnitude) with a respiratory effect. Additionally, results
indicated that risk estimates for a mixture were often similar, but in some cases larger, than those reported
for PM2 5 alone. However, for cardiovascular effects in general, the evidence neither consistently or
coherently indicated a stronger or weaker effect of combined exposure to PM2 5 and another pollutant
compared to exposure to PM2 5 and other pollutants alone.
2.2.1.2 Long-Term PM2.5 Exposure
Epidemiologic studies focusing on long-term PM2 5 exposure and health effects have traditionally
provided a more limited assessment of the potential confounding effects of copollutants on PM2 5
associations. Studies evaluated in the 2019 PM ISA that provided an assessment of copollutant
confounding directly addressed a previously identified uncertainty in the scientific evidence.
Across the health effects evaluated within the 2019 PM ISA, relatively few studies examined the
potential confounding effects of copollutants on the relationship between long-term PM2 5 exposure and
respiratory (2019 PM ISA, Section 5.2.13), cardiovascular (2019 PM ISA, Section 6.2.18), and cancer
(2019 PM ISA, Section 10.2.7), with a general lack of studies assessing the role of copollutant
confounding on observed associations with nervous system effects (2019 PM ISA, Section 8.2.9). These
studies often did not examine the full suite of gaseous pollutants but tended to focus on traffic-related
pollutants (i.e., N02, NOx, and CO) and O3, with some studies also examining PMio-2.5. Across studies,
low to moderate correlations (i.e., r < 0.7) were often observed between copollutants and PM2 5.
Collectively, studies that examined the potential confounding effects of copollutants on the PM2 5
association with respiratory (i.e., lung function and asthma development) and cardiovascular effects
(i.e., cardiovascular mortality), along with lung cancer incidence and mortality, reported associations that
were relatively unchanged in copollutant models, but these assessments were conducted in a limited
number of studies.
Several studies of long-term PM2 5 exposure and mortality examined potential copollutant
confounding. Within studies that examined the potential confounding effects of copollutants on the
relationship between long-term PM2 5 exposure and mortality, the most extensive analyses occurred for
O3, with a limited number of studies examining N02, S02, PMio-2.5, and benzene. Studies that examined
September 2021
2-26
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
O3 reported correlations that were generally moderate (ranging from r = 0.49-0.73), with a few studies
reporting weak correlations (r < 0.4). Overall, associations remained relatively unchanged in copollutant
models for total (nonaccidental) mortality, cardiovascular, and respiratory mortality (2019 PM ISA,
Figure 11-18). Studies focusing on copollutant models with NO2, PM10-2.5, SO2, and benzene were
examined in individual studies, and across these studies the PM2 5-mortality association was relatively
unchanged (2019 PM ISA, Figure 11-19).
2.2.2 Timing of Effects
An important question to address when evaluating the scientific evidence demonstrating health
effects due to short-term PM2 5 exposure is the timing of observed effects. Studies have attempted to
address this question through two primary avenues: (1) examining various averaging times of the
exposure metric used to represent short-term PM2 5 exposure to determine whether PM2 5 concentrations
averaged over time periods other than 24-hours are more closely associated with health effects and
(2) assessing whether the relationship between exposure and effect is biologically plausible by examining
the lag days over which associations are observed.
2.2.2.1 Averaging Time
Most epidemiologic studies evaluated in the 2019 PM ISA that examined the relationship
between short-term PM2 5 exposures and health effects relied primarily on an exposure metric averaged
over 24-hours. Some recent studies, focusing on respiratory and cardiovascular effects and mortality, have
examined whether there is evidence that subdaily exposure metrics are more closely related to health
effects than the traditional 24-hour avg metric.
Epidemiologic studies that examined both respiratory-related ED visits and hospital admissions,
as well as subclinical markers of respiratory effects, explored associations with subdaily exposure metrics
(2019 PM ISA, Section 5.1.10.5). In studies of respiratory-related ED visits and hospital admissions,
positive associations were not consistently observed with subdaily exposure metrics, and often there was
no information on spatiotemporal variability of the subdaily metrics. Additionally, in a study that
examined multiple subdaily averaging times and compared them with the 24-hour avg exposure metric,
there was no difference in associations across metrics, but this result was limited to a single study
location. Panel studies also examined subdaily exposure metrics through personal monitoring, but
associations were not consistently observed at shorter averaging times for markers of pulmonary
inflammation and changes in lung function.
A more limited number of studies examined subdaily exposure metrics and cardiovascular effects
(2019 PM ISA, Section 6.1.14.3). Studies of ST elevation, myocardial infarction, out-of-hospital cardiac
arrest, and cerebrovascular disease ED visits and hospital admissions reported positive associations with
September 2021
2-27
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
subdaily exposure metrics, but the magnitude of the association tended to be larger when averaging over
multiple hours up to 1 day (i.e., 24-hour avg). These studies provided evidence that continues to support
the use of a 24-hour avg exposure metric.
A few studies examined subdaily PM2 5 exposure metrics and associations with mortality,
focusing on comparisons between the 24-hour avg and an hourly peak exposure metric (2019 PM ISA,
Section 11.1.8.2). In these studies, positive associations were reported for both the 24-hour avg and
hourly peak exposure metric, with the association often slightly larger in magnitude for the 24-hour avg
metric. Collectively, the available evidence did not indicate that subdaily averaging periods for PM2 5
were more closely associated with health effects than the 24-hour avg exposure metric.
2.2.2.2 Lag Structure of Associations
Often epidemiologic studies examined associations between short-term PM2 5 exposure and health
effects over a series of single-day lags, multiday lags, or by selecting lags a priori. Studies evaluated in
the 2019 PM ISA have expanded the assessment of the timing of effects by systematically examining lag
days by focusing on whether there is evidence of an immediate (e.g., lag 0-1 days), delayed (e.g., lag
2-5 days), or prolonged (e.g., lag 0-5 days) effect of PM on health.
Epidemiologic studies of respiratory effects have primarily focused on examining the lag
structure of associations for respiratory-related ED visits and hospital admissions, with most studies
examining asthma exacerbation with a more limited assessment for COPD exacerbation and respiratory
infection (2019 PM ISA, Section 5.1.10.3). Across the studies that examined asthma, COPD, respiratory
infections, and combinations of respiratory-related diseases, the strongest association reported, in terms of
magnitude and precision, was generally within a few days after exposure, but there was some evidence
demonstrating the potential for a prolonged effect of PM2 5 (i.e., lags ranging from 0-5 days). Recent
studies of respiratory mortality provided additional insight on the lag structure of associations for
respiratory-related effects due to short-term PM2 5 exposure. Studies of respiratory mortality tended to
support more immediate PM2.5 effects (i.e., lags of 0 to 2 days), but with initial evidence of stronger
associations, in terms of magnitude and precision, at lags of 0-5 days. Collectively, the studies of
respiratory morbidity and mortality that conducted systematic evaluations of PM2 5 associations across a
range of lags provided evidence of effects within the range of 0-5 days after exposure.
As with respiratory effects, the majority of epidemiologic studies examining the lag structure of
associations for cardiovascular effects focused on ED visits and hospital admissions. Studies of ED visits
and hospital admissions for IHD, MI, and cardiovascular-related outcomes reported stronger associations
for multiday lags, but these effects tended to be in the range of 0-1 or 0-2 days. When examining
cerebrovascular disease, there was no evidence of an association at any of the lag days examined;
however, when focusing on specific stroke types, particularly ischemic stroke, there was evidence of
immediate effects at lags of 0 and 1 day, which was consistent with other cardiovascular outcomes. The
September 2021
2-28
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
immediate effects of PM2 5 on cardiovascular morbidity outcomes, specifically those related to ischemic
events, were consistent with the lag structure of associations observed in studies of cardiovascular
mortality that reported immediate effects (i.e., lag 0-1 day). There was some evidence indicating
PM2 5-cardviovascular mortality associations with exposures over longer durations, but this was not
supported by studies examining single-day lags that encompassed the same number of days.
An evaluation of epidemiologic studies of short-term PM2 5 exposure and mortality found that
studies either conducted analyses of single-day lags over many days or various iterations of multiday lags
(e.g., 0-1, 0-2, 0-3, etc.; 2019 PMISA, Section 11.1.8.1). Across studies, associations were largest in
terms of magnitude and precision for total (nonaccidental) mortality at lags of 0 to 1 day, but there was
some evidence that associations remained positive at multiday lags up to 0-4 days. The combination of
the multi- and single-day lag analyses provided further support of an immediate effect of short-term PM2 5
exposure on mortality.
2.2.3 Concentration-Response Relationship
In assessing the relationship-between short- and long-term PM exposure and health effects, an
important consideration is whether the relationship is linear across the full range of ambient
concentrations and whether there is a threshold concentration below which there is no evidence of an
effect. As detailed in the 2004 AQCD and 2009 PM ISA, conducting C-R and threshold analyses is
challenging because of the "(1) limited range of available concentration levels (i.e., sparse data at the low
and high end); (2) heterogeneity of (at-risk) populations (between cities); and (3) influence of
measurement error" (U.S. EPA. 2004). Studies evaluated in the 2019 PM ISA that focused on the shape
of the C-R curve expanded upon the health effects evaluated in previous reviews and continued to provide
evidence of a linear, no-threshold relationship between both short- and long-term PM2 5 exposure and
several respiratory and cardiovascular effects, and mortality. Some evidence indicated a steeper slope
(i.e., supralinear curve) at lower concentrations for some outcomes (i.e., long-term PM2 5 exposure and
mortality). Cutpoint analyses that focused on whether risk changed at different concentration ranges
provided some evidence of nonlinearity, specifically in the relationship between short-term PM2 5
exposure and respiratory-related ED visits and hospital admissions. Although studies evaluated in the
2019 PM ISA have used many different statistical methods to examine the shape of the C-R relationship
and generally provided evidence for a linear, no-threshold relationship, many of these studies have not
systematically evaluated alternatives to a linear relationship.
2.2.3.1 Short-Term Exposure
Epidemiologic studies evaluated in the 2019 PM ISA that examined the C-R relationship between
short-term PM2 5 exposure and health were limited to studies of respiratory-related ED visits and hospital
September 2021
2-29
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
admissions (2019 PM ISA, Section 5.1.10.6), and mortality (2019 PM ISA, Section 11.1.10). Across
studies that examined respiratory effects, different analytical methods have been employed to examine the
C-R relationship, either explicitly examining the shape of the C-R curve and whether there was evidence
of linearity across the full range of PM2 5 concentrations, or through cutpoint analyses that examine
whether the risk of a PM25-related respiratory effect changed within specified ranges of PM2 5
concentrations. These studies primarily focused on asthma ED visits and hospital admissions, with some
studies examining combinations of respiratory-related ED visits and hospital admissions. Studies that
focused on the shape of the C-R curve provided initial evidence of a linear relationship for short-term
PM2 5 exposure and both respiratory disease and asthma ED visits and hospital admissions, with less
certainty at concentrations below 10 (ig/m3. However, cutpoint analyses provided some initial evidence
indicating nonlinearity in the relationship (i.e., larger risk estimates at various quintiles when compared
with the lowest quintile) between short-term PM2 5 exposure and asthma ED visits and hospital
admissions.
Studies that examined the C-R relationship for short-term PM exposure and mortality were
initially limited to those focusing on PMi0. Recent epidemiologic studies focused on PM2 5 and
specifically the shape of the C-R curve at the low end of the PM2 5 concentration distribution. Evidence
from U.S. studies conducted at lower PM2 5 concentrations compared with other countries, provided
evidence indicating a linear relationship at concentrations as low as 5 (ig/m3. The observations from C-R
analyses were further supported by cutpoint analyses examining associations at different PM2 5
concentrations, as well as analyses that reported no evidence of a threshold. Overall, studies evaluated in
the 2019 PM ISA focusing on short-term PM2 5 exposure and mortality supported a linear, no-threshold
relationship at ambient PM2 5 concentrations lower than those evaluated in the 2009 PM ISA.
2.2.3.2 Long-Term Exposure
The most extensive analyses of the C-R relationship between long-term PM exposure and a health
effect have generally been for PM2 5 and mortality. Recent studies further expanded and provided new
insights on the relationship between long-term PM2 5 exposure and mortality. In addition, these studies
provided initial examinations of the C-R relationship for respiratory and cardiovascular effects, as well as
for lung cancer incidence and mortality.
Although the C-R relationship for long-term PM2 5 exposure has not been assessed for most
health effects, it has been extensively examined in studies of mortality (2019 PM ISA, Section 11.2.4).
Across studies, a variety of statistical methods have been used to assess whether there is evidence of
deviations in linearity. Studies have also conducted cutpoint analyses that focus on examining risk at
specific ambient concentrations (2019 PM ISA, Table 11-7). These studies reported results that generally
support a linear, no-threshold relationship for total (nonaccidental) mortality, especially at lower ambient
PM2 5 concentrations, with confidence in the linear relationship as low as 5-8 (ig/m3 in some studies.
September 2021
2-30
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
Additionally, there was initial evidence indicating that the slope of the C-R curve may be steeper
(supralinear) at lower concentrations for cardiovascular mortality.
Few epidemiologic studies have examined the C-R relationship for long-term PM2 5 exposure and
respiratory effects (2019 PM ISA, Section 5.2.3.1.2). The ones that do have focused on asthma incidence
and childhood wheeze. Studies of asthma incidence that examined the shape of the C-R curve and
whether risk changes at different quartiles of PM2 5 concentrations did not find any evidence of deviations
in linearity and monotonically increasing risk, respectively. In an initial study of childhood wheeze,
specifically repeated wheeze events, there was evidence of a linear C-R relationship with confidence in
the linear relationship at long-term PM2 5 concentrations as low as 10 to 12 (ig/m3.
A limited number of studies reported initial assessments of the C-R relationship for long-term
PM2 5 concentrations and cardiovascular effects, specifically IHD incidence, CAC, and hypertension
(2019 PM ISA, Section 6.2.16). For IHD incidence, there was evidence of a linear C-R relationship at
concentrations below 15 (ig/m3, which was consistent with the shape of the curve when compared with
the full range of PM25 concentrations. Analyses of the relationship between long-term PM25 exposure and
CAC indicated both linear and nonlinear relationships, while there is preliminary evidence of a linear
relationship between long-term PM2 5 exposure and incidence of hypertension. A few studies that
examined the relationship between long-term PM2 5 exposure and lung cancer incidence and mortality
also examined the shape of the C-R curve by assessing its linearity and conducting cutpoint and threshold
analyses (2019 PM ISA, Section 10.2.5.1.4). These collective assessments provided initial evidence
supporting a no-threshold, linear relationship across the range of PM2 5 concentrations observed in the
U.S., with confidence in a linear relationship as low as 5-10 (ig/m3 in some studies.
2.2.4 PM Components and Sources
Building on the initial evaluation conducted in the 2004 PM AQCD, the 2009 PM ISA formally
evaluated the relationship between exposures to PM components and sources and health effects. This
evaluation found that many components and sources representative of combustion-related activities
(e.g., motor vehicle emissions, coal combustion, oil burning, vegetative burning) were associated with a
range of health effects. The 2009 PM ISA, therefore, concluded that "many [components] of PM can be
linked with differing health effects and the evidence is not yet sufficient to allow differentiation of those
components or sources that are more closely related to specific health outcomes."
Building upon the evaluation of PM sources and components in the 2009 PM ISA, and as detailed
in the Preface of the 2019 PM ISA, the 2019 PM ISA systematically evaluated whether specific PM
components or sources were more strongly associated with health effects than PM mass by focusing on
those studies that: (1) included a composite metric of PM (e.g., mass of PM2 5 and/or PM10-2.5, or in the
case of ultrafine particles [UFP] mass, particle number, etc.) and PM components; (2) applied some
approach to assess the particle effect (e.g., particle trap) of a mixture; or (3) conducted formal statistical
September 2021
2-31
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
analyses to identify source-based exposures (see 2019 PM ISA, Preface). Overall, these criteria allowed
for a thorough evaluation of whether there was evidence that an individual component(s) and/or source(s)
was more closely related to health effects than PM mass. Across the health effects categories evaluated in
the 2019 PM ISA, most studies that examined PM sources and components focused on PM2 5. Thus, the
following sections summarize the state of the science on PM2 5 components and sources for those health
effects categories for which it was concluded within the 2019 PM ISA that there was a causal or likely to
be causal relationship. Details on the PM2 5 components and sources evidence relevant to other health
effects categories (e.g., Reproductive and Developmental Effects) are covered in the health chapters of
the 2019 PM ISA.
Overall, recent studies continued to demonstrate that many PM2 5 components and sources were
associated with health effects ranging from subclinical (e.g., changes in heart function, such as HRV, or
circulating biomarkers) to the more overt (i.e., ED visits, hospital admissions, and mortality). The results
of these studies confirmed and further supported the conclusion of the 2009 PM ISA that many PM2 5
components and sources are associated with many health effects and that the evidence does not indicate
that any one source or component is consistently more strongly related with health effects than PM2 5
mass.
2.2.4.1 Respiratory Effects
The examination of PM2 5 components and sources and respiratory effects was limited to
epidemiologic studies (2019 PM ISA, Section 5.1.11). Epidemiologic studies that examined the
relationship between respiratory health effects and short-term exposure to both PM2 5 mass (n = 113) and
PM2 5 components, primarily focused on the components nitrate (n = 29), sulfate (n = 40), OC (n = 50),
and EC/BC (n = 95). Across these studies, the health effects examined range from inflammation and
changes in lung function to respiratory-related ED visits and hospital admissions. When examining the
pattern of associations for individual PM2 5 components with those observed for PM2 5 mass, all the
components examined (i.e., evaluated in at least three studies) were positively associated with a
respiratory effect in at least a few studies (2019 PM ISA, Section 5.1.11.7). For EC/BC, the most
extensively examined PM2 5 component, many studies reported positive associations, but some studies
also reported results indicating no association, which was consistent with the pattern of associations for
PM2 5 mass.
A more limited number of studies examined associations between long-term PM2 5 components
exposure and respiratory effects (2019 PM ISA, Section 5.2.12). Similar to short-term exposure studies,
most long-term exposure studies focused on EC/BC and did not observe a pattern of associations with
respiratory effects different from that observed for PM2 5 mass. Collectively, positive associations were
observed in studies examining short- and long-term PM2 5 component exposure and respiratory effects,
September 2021
2-32
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
but there was no evidence that any one component was more strongly associated with respiratory effects
than PM2 5 mass.
Few studies examined the relationship between PM2 5 sources and respiratory health effects.
Through analyses in which PM2 5 components were apportioned into source factors, positive associations
were reported for several respiratory effects, particularly asthma exacerbation, and sources representative
of combustion-related activities, such as traffic and biomass burning. No studies evaluated in the 2019
PM ISA examined long-term exposure to PM2 5 sources and respiratory effects.
2.2.4.2 Cardiovascular Effects
Both epidemiologic and experimental studies examined the relationship between exposure to
PM2 5 component and sources and cardiovascular effects (2019 PM ISA, Section 6.1.15). In short-term
exposure studies, the epidemiologic evidence focused on studies examining cardiovascular-related ED
visits and hospital admissions, with only a few studies examining other cardiovascular effects. Similar to
respiratory effects studies, the cardiovascular effects studies that examined both PM2 5 mass and
components (n = 14), focused most extensively on EC (n = 12), OC (n = 10), sulfate (n = 9), and nitrate
(n = 9). Across all components examined, most were positively associated with cardiovascular-related ED
visits and hospital admissions in at least one study (2019 PM ISA, Section 6.1.15). Although EC was
positively associated with cardiovascular-related ED visits and hospital admissions in many of the studies
evaluated, it was not possible to tell whether EC was independently associated or a marker of exposure to
PM2 5 mass.
Few studies examined long-term exposure to PM2 5 components and cardiovascular effects, and
those that did were consistent with the long-term exposure and respiratory effects studies that primarily
focused on EC/BC (2019 PM ISA, Section 6.2.17). These studies did not provide evidence that any one
component was more strongly associated with a cardiovascular effect. Collectively, studies examining
short- and long-term PM2 5 components exposure continue to support that there is no evidence that any
one component is more strongly associated with a cardiovascular effect than PM2 5 mass.
Epidemiologic and animal toxicological studies conducted source-based analyses using
mathematical methods to apportion PM2 5 components into source factors (2019 PM ISA, Section 6.1.15.6
and Section 6.1.15.8). Epidemiologic studies focused on cardiovascular-related ED visits and hospital
admissions and reported positive associations with sources representative of combustion-related activities
(e.g., industrial combustion, traffic), with more limited evidence for wildfires. Animal toxicological
studies, which focused on markers of heart function (e.g., HR, HRV), reported associations with a variety
of source categories, but the associations were dependent on the location of the study (i.e., where the
PM2 5 CAPs were collected). Additional studies focusing on long-term exposures to PM2 5 sources were
fewer, with epidemiologic studies only examining traffic sources and animal toxicological studies
reporting associations between a number of sources and various cardiovascular effects.
September 2021
2-33
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
2.2.4.3 Mortality
Epidemiologic studies that examined associations with PM2 5 components and sources and
mortality have primarily focused on examining short- and long-term exposures to components (2019 PM
ISA, Section 11.1.11 and Section 11.2.6). Both short- and long-term exposure studies reported consistent,
positive associations with PM2 5 mass across all studies that also examined a component. Although the
respiratory and cardiovascular effects studies focused mainly on EC/BC, the studies of mortality did not
examine any one component disproportionately to the others. Of the PM2 5 components examined, each
were found to be positively associated with mortality in at least a few studies, but overall, one component
was not found to be as consistently associated with mortality as PM2 5 mass.
Compared with the 2009 PM ISA, in which most epidemiologic studies of mortality conducted
formal source apportionment analyses, studies evaluated in the 2019 PM ISA have focused more
exclusively on PM2 5 components. Of the limited number of studies that examined associations between
short- and long-term source exposures and mortality, positive associations were observed for those
sources representative of combustion-related activities, including traffic, coal, and vegetative fires.
2.2.5 Populations and Lifestages at Potentially Increased Risk of a
PM-Related Health Effect
An important consideration in evaluating the scientific evidence for PM, and in determining the
extent to which the NAAQS provides public health protection, is whether specific populations or
lifestages are at increased risk of a PM-related health effect. As detailed in the preceding sections of this
chapter and in health effects chapters of the 2019 PM ISA, a large body of evidence shows that health
effects related to PM exposure, particularly PM2 5 exposure, occur across populations with diverse
characteristics (e.g., children, older adults, people with preexisting cardiovascular diseases, etc.).
Although this larger body of evidence provided information on the causal nature of the relationship
between PM exposure and health effects, this section focuses on answering the following question:
Are there specific populations and lifestages at increased risk of a PM-related health effect,
compared to a reference population? That is, is the magnitude of effect or exposure greater for some
populations or lifestages compared to a reference population, where applicable?
The evaluation of populations and lifestages potentially at increased risk builds off the approach
used in the 2009 PM ISA and involved application of a framework detailed in the 2013 O3 ISA to
characterize the evidence informing whether a population or lifestage is at increased risk (U.S. EPA.
2013). The focus of this evaluation was on determining the extent to which specific factors may increase
the risk of a PM-related health effect in a population or lifestage relative to a reference population, where
applicable. Importantly, this evaluation builds on the conclusions drawn elsewhere in the 2019 PM ISA,
taking into consideration the relationship between exposure to PM and health effects. As detailed in the
September 2021
2-34
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Preamble to the ISAs (U.S. EPA. 2015). the evaluation of the evidence includes (1) epidemiologic studies
that conducted stratified analyses, (2) evidence from animal toxicological studies using animal models of
disease and epidemiologic or controlled human exposure studies conducted in specific populations
(e.g., lung function growth in children, people with mild asthma), (3) information on the dosimetry of PM
within the body, and (4) consideration of information on differential exposure to PM within a population
or lifestage. Overall, the framework allows for a transparent characterization of the collective body of
evidence to draw conclusions on the degree to which the scientific evidence indicates that a specific
population or lifestage is at increased risk of a PM-related health effect (2019 PM ISA, Table 12-1).
Based on the causality determinations briefly summarized within this section, and more fully
detailed in the health effects chapters of the 2019 PM ISA, the strongest evidence indicating an effect of
short- and long-term PM exposure on health is for PM2 5 and the broad health categories of respiratory
and cardiovascular effects, nervous system effects, cancer, and mortality. Thus, the assessment of
populations and lifestages potentially at increased risk of a PM2 5-related health effect primarily focused
on studies that form the basis of these causality determinations that also conducted analyses to inform
whether there is differential risk in a specific population or lifestage. In evaluating studies, several factors
can influence the ability to observe an association, including, but not limited to, publication bias (i.e., not
reporting null findings when examining evidence of differential risk), variability in how indicators or
metrics are defined across studies (e.g., socioeconomic status, obesity, age), and variability in the
population as a whole, particularly with respect to behavioral differences, biological differences
(e.g., obese vs. nonobese), and adherence to treatment for preexisting diseases.
Of the factors evaluated (2019 PM ISA, Table 12-3 for a full list), children and race were the only
factors for which it was concluded that "adequate evidence " was available indicating that people of a
specific lifestage and race are at increased risk of PM2 5-related health effects (2019 PM ISA,
Section 12.5.1.1 and Section 12.5.4). Although stratified analyses do not indicate adifference in the risk
of PM-related health effects between children and adults, there was strong evidence from studies focusing
on children that demonstrated health effects only observable in growing children that were attributed to
PM2 5 exposure. Specifically, epidemiologic studies evaluated in the 2019 PM ISA of long-term PM25
exposure provided strong evidence of impaired lung function growth with additional evidence of
decrements in lung function and the development of asthma. The results of these longitudinal
epidemiologic studies were consistent with and extended the evidence that was available in the 2009 PM
ISA demonstrating health effects in children due to long-term PM25 exposure. The conclusion of
"adequate evidence " for race was based on studies that examined whether there was evidence of
increased risk for PM2 5-related health effects as well as studies that examined differential exposure by
race. Multiple studies reported that non-White populations across different geographical regions are
exposed to higher PM2 5 concentrations and were at increased risk for PM2 5-related mortality, particularly
due to long-term exposure. Collectively, the combination of evidence demonstrated that non-White
populations are at greater risk for both PM2 5-related health effects and PM2 5 exposure than are Whites.
September 2021
2-35
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
There was "suggestive evidence " that populations with preexisting cardiovascular (2019 PM ISA,
Section 12.3.1) or respiratory (2019 PM ISA, Section 12.3.5) disease, those that are overweight or obese
(2019 PM ISA, Section 12.3.3), those with particular genetic variants (2019 PM ISA, Section 12.4), those
that are of low SES (2019 PM ISA, Section 12.5.3), and those who are current or former smokers (2019
PM ISA, Section 12.6.1) are at increased risk for PM2 5-related health effects. Epidemiologic studies that
conducted analyses stratified by preexisting cardiovascular disease tended to focus on hypertension, one
of the most easily measurable cardiovascular conditions, and did not consistently indicate increased risk
for several outcomes examined (e.g., mortality, stroke, BP). However, the strong evidence supporting a
causal relationship between short- and long-term PM2 5 exposure and cardiovascular effects, which
included cardiovascular-related mortality and ischemic heart disease (2019 PM ISA, Section 6.1.16 and
Section 6.2.18) indicated that individuals with underlying cardiovascular conditions related to these
serious outcomes may be at increased risk of a PM2 5-related health effect. Similarly, there were few
studies that evaluated whether there is evidence of increased risk of a PM2 5-related health effect between
people with preexisting asthma (2019 PM ISA, Section 12.3.5) and COPD (2019 PM ISA, Section 12.3.5)
compared with people that do not have a preexisting respiratory disease. However, epidemiologic studies,
particularly those studies examining short-term PM2 5 exposure and asthma or COPD ED visits and
hospital admissions reported generally consistent positive associations (2019 PM ISA, Section 5.1.2.1 and
Section 5.1.4.1), which represent exacerbations that are only possible in people with asthma or COPD.
Therefore, there was limited evidence to support that people with preexisting respiratory diseases,
specifically asthma or COPD, are at increased risk for a PM2 5-related health effect, but there was
generally consistent evidence demonstrating these populations experience health effects due to a PM2 5
exposure.
Studies that examined whether being obese or overweight increased the risk of a PM2 5-related
health effect, reported evidence of increased risk for mortality associated with long-term exposures to
PM2 5, but inconsistent evidence was found for subclinical cardiovascular outcomes, when comparing
obese or overweight individuals with normal weight individuals. However, the evaluation of studies
focusing on differences in risk by weight were complicated by the different definitions of obesity used
across studies.
The examination of whether specific genetic characteristics dictate increased risk of a
PM2 5-related health effect involved studies of genetic variants. Across the large number of genetic
variants examined there was a consistent trend for increased risk of respiratory and cardiovascular effects
associated with PM2 5 exposure across gene variants involved in the glutathione transferase pathway.
These results were consistent with underlying mechanisms that provided biological plausibility for
PM2 5-related health effects and have shown that oxidative stress is an early response to PM2 5 exposure.
Epidemiologic studies have examined several measures of SES (e.g., income level, educational
attainment, etc.) in assessing whether populations are at increased risk of a PM2 5-related health effect. In
studies examining both differential exposure as well as increased risk of health effects, there was some
September 2021
2-36
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
evidence that low SES populations are more likely to have higher PM2 5 exposures and that low SES
populations, as measured by metrics for income, are at increased risk of PM2 5-related mortality when
compared with populations defined as higher SES. Finally, there was some epidemiologic evidence from
studies examining long-term PM2 5 exposure and lung cancer incidence and mortality, as well as total
mortality, that people who currently smoke or were former smokers may be at increased risk of a
PM2 5-related health effect compared with never smokers.
For the remaining factors evaluated, "inadequate evidence" exists to determine whether having
diabetes (2019 PM ISA, Section 12.3.2), being in an older lifestage (i.e., older adults; 2019 PM ISA,
Section 12.5.1.2), residential location (including proximity to source and urban residence; 2019 PM ISA,
Section 12.5.5), sex (2019 PM ISA, Section 12.5.2), or diet (2019 PM ISA, Section 12.6.2) increase the
risk of PM2 5-related health effects. Across these factors there was either limited assessment of differential
risk or exposure (i.e., residential location, diet), or inconsistency in results across studies to support
evidence of increased risk of a PM2 5-related health effect (i.e., diabetes and sex). Instead, the
inconsistency in the evidence makes the determination of disproportionately increased risk more difficult.
For example, for older adults (2019 PM ISA, Section 12.5.1.2) there was a relatively small number of
studies that examined whether there is evidence of differential risk between age groups. In the evaluation
of these studies there was limited evidence indicating that older adults are at increased risk of
PM2 5-related health effects compared with other age ranges; however, epidemiologic studies focusing
only on older adults demonstrated associations with respiratory-related ED visits and hospital admissions
with additional, but more limited, evidence of subclinical cardiovascular effects from epidemiologic panel
studies and controlled human exposure studies.
September 2021
2-37
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
2.3 Welfare Effects
Whereas the evaluation of the evidence for PM exposures and health effects was specific to
exposure duration (i.e., short- and long-term) and PM size fraction (i.e., PM2 5, PM10-2.5, and UFP), the
evaluation of the evidence for welfare effects focused generally on whether there was a causal
relationship between PM and visibility impairment, climate effects, and effects on materials. As detailed
below, the evidence continued to support a causal relationship between PM and visibility impairment
(2019 PM ISA, Section 1.6.1), climate effects (2019 PM ISA, Section 1.6.2), and materials effects (2019
PM ISA, Section 1.6.3).
2.3.1 Visibility Impairment
It is well known that light extinction from pollution is primarily due to PM2 5, resulting in the
2019 PM ISA concluding there is a causal relationship between PM and visibility impairment, which was
consistent with the conclusions of the 2009 PM ISA (Table 2-3). This conclusion was based on additional
characterization of the effect of PM size and composition on light extinction.
The relationship between PM and light extinction has been well documented (2019 PM ISA,
Section 13.2.2). Although reconstruction of light extinction is best achieved with detailed information on
the size and composition of PM measurements, empirical relationships between light extinction of PM
components are more practical and have been successfully evaluated and widely used (2019 PM ISA,
Section 13.2.3). Light extinction has been found to vary depending on the available PM species
monitoring data, with light extinction efficiencies varying by a factor of 10 between species. Additionally,
the variation in PM species by region and season, as well as urban and rural location, can affect light
extinction. The steep decline in PM2 5 sulfate of-4.6% per year in rural areas and -6.2% per year in urban
areas from 2002-2012 (2019 PM ISA, Section 1.2.1) has affected the apportionment of light extinction
among PM2 5 species. Although PM2 5 sulfate is still responsible for more light extinction than any other
single species, visibility in many areas has improved, and a smaller and less seasonally variable fraction
of light extinction can be attributed to PM2 5 sulfate, with an increasing share due to nitrate and organic
matter (2019 PM ISA, Section 13.2.4).
2.3.2 Climate Effects
Substantial evidence indicates that PM affects the radiative forcing of the climate system, both
through direct scattering and absorption of radiation, and indirectly, by altering cloud properties, resulting
in the conclusion that there is a causal relationship between PM and climate effects, which was consistent
with the conclusions of the 2009 PM ISA (2019 PM ISA, Table 1-3). This conclusion was based on
multiple studies evaluated in the 2019 PM ISA that have strengthened the evidence for the effects of PM
September 2021
2-38
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
on radiative forcing and have improved the characterization of major sources of uncertainty in estimating
PM climate effects, including the indirect radiative forcing effects associated with PM-cloud interactions,
and the additional climate effects and feedbacks involving atmospheric circulation and the hydrologic
cycle resulting from PM effects on radiative forcing.
Because of these radiative effects, the net effect of PM has been to cool the planet over the last
century, masking some of the effects of greenhouse gases on warming (2019 PM ISA, Section 13.3.3).
The decrease in PM concentrations in many developed countries over the last few decades has likely
contributed to the recent shift toward "global brightening," which may in turn have helped drive rapid
warming in North American and Europe because this greenhouse-gas warming was unmasked (2019 PM
ISA, Section 13.3.6). In developing countries in Asia, by contrast, there has been an increase in PM
concentrations over the last several decades, but the associated radiative forcing effects are highly
uncertain, due to uncertainties in emissions estimates and the lack of accurate information on the
proportion of reflecting versus absorbing species. Although uncertainties in the relationship between PM
and climate effects have been further studied and characterized since the 2009 PM ISA, there are still
substantial uncertainties with respect to key processes linking PM and climate, specifically the interaction
between clouds and aerosols. This is because of the small scale of PM-relevant cloud microphysical
processes compared with the coarse resolution of state-of-the-art models, and because of the complex
cascade of indirect effects and feedbacks in the climate system that result from a given initial radiative
perturbation caused by PM.
2.3.3 Materials Effects
Multiple studies evaluated in the 2019 PM ISA further characterized soiling and corrosion
processes associated with PM and add to the body of evidence of PM damage to materials. Approaches to
quantify pollutant exposure corresponding to perceived soiling and damage continue to indicate that
deposition can result in increased cleaning and maintenance costs and reduced usefulness of soiled
material. The combination of this evidence resulted in the conclusion that there is a causal relationship
between PM and effects on materials, which was consistent with the conclusions of the 2009 PM ISA
(Table 2-3).
Assessments of the relationship between PM and effects on materials have often focused on
quantitative assessments including the development of dose-response relationships and application of
damage functions to stone used for historic monuments and buildings. Recent studies provided additional
information on understanding soiling and corrosion process for glass and metals and have allowed for the
development of new dose-response curves (2019 PM ISA, Section 13.4.3), particularly for glass, as well
as new damage functions for materials (2019 PM ISA, Section 13.4.4). Additional evidence demonstrated
that atmospheric soiling can affect energy costs and climate control, energy consumption of large
buildings, and the efficiency of photovoltaic systems (2019 PM ISA, Section 13.4.2).
September 2021
2-39
DRAFT: Do Not Cite or Quote
-------
Table 2-3 Key evidence contributing to causal causality determinations for PM
exposure and welfare effects evaluated in the 2019 Integrated
Science Assessment for Particulate Matter.
Key Evidence
in 2019 PM ISA Welfare Effect Category3 and Causality Determination
Visibility Impairment and PM Exposure (2019 PM ISA, Section 13.2): Causal Relationship
No Change in Causality Determination From the 2009 PM ISA; New Evidence Further Supports the Previous
Determination.
Section 13.2.6 Visibility impairment by atmospheric PM, with the strongest effects in the size range from 0.1 to
1.0 |jm, was supported by numerous studies summarized in the 1969 PM AQCD (NAPCA,
1969), although the relationship between PM and atmospheric visibility impairment was well
established decades earlier. Additional studies supporting the relationship have been described
in subsequent documents, and additional new evidence is based on extensive simultaneous
network measurements of PM2.5 and light extinction.
Climate Effects and PM Exposure (2019 PM ISA, Section 13.3): Causal Relationship
No Change in Causality Determination From the 2009 PM ISA; New Evidence Further Supports the Previous
Determination.
Section 13.3.9 Effects of PM on radiative forcing of the climate system through both absorption and scattering
of radiation directly, as well as through indirect effects involving interactions between PM and
cloud droplets, with corresponding effects on temperature, precipitation, and atmospheric
circulation, was supported by numerous observational and modeling studies. Research since the
2009 PM ISA (U.S. EPA. 2009) has improved understanding of climate-relevant aerosol
properties and processes, as well as characterized key sources of uncertainty in estimating PM
climate effects, particularly with respect to PM-cloud interactions.
Materials Effects and PM Exposure (2019 PM ISA, Section 13.4): Causal Relationship
No Change in Causality Determination From the 2009 PM ISA; New Evidence Further Supports the Previous
Determination.
Section 13.4.5 Both soiling and corrosion associated with PM contribute to materials damage (U.S. EPA. 2009
2004. 1982). Deposition of PM can physically affect materials by promoting or accelerating the
corrosion of metals, by degrading paints, and by deteriorating building materials such as stone,
concrete, and marble. Further characterization of PM effects on glass and metals, along with
quantitative dose-response relationships and damage functions for stone and other materials
lend additional support to the causal relationship in the 2009 PM ISA. Studies evaluated in the
2019 PM ISA showed that deposition of PM reduces energy efficiency of photovoltaic systems.
AQCD = Air Quality Criteria Document; PM = particulate matter.
aThe sections referenced in the 2019 PM ISA include a detailed discussion of the available evidence that informed the causality
determinations.
September 2021
2-40
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
3.EVALUATION OF RECENT HEALTH EFFECTS
EVIDENCE
The following section focuses on the evaluation of recent health effects studies that fall within the
scope of this Supplement as detailed in Section 1.2.2. Within this section the evaluation of recent studies
is performed in the context of the studies evaluated and scientific conclusions presented in the 2019 PM
ISA. As a result, within each of the following sections, the summary and causality determination from the
2019 PM ISA is presented prior to the evaluation of recent studies published since the literature cutoff
date of the 2019 PM ISA that examine the relationship between short-term and long-term PM2 5 exposure
and cardiovascular effects (Section 3.1) and mortality (Section 3.2V6 This approach allows for a full
accounting of the evidence that formed the basis of the key scientific conclusions in the 2019 PM ISA and
the identification of specific sections of the 2019 PM ISA that provide additional details on the total
evidence base being considered in the process of reconsidering the PM NAAQS.
In addition to the evaluation of recent U.S. and Canadian epidemiologic studies that examine the
relationship between short-term and long-term PM2 5 exposure cardiovascular effects and mortality this
section also evaluates studies that address key scientific topics where the literature has evolved since the
2020 PM NAAQS review was completed, specifically since the literature cutoff date for the 2019 PM
ISA (Section 3.3). These topics that further inform the health effects attributed to PM2 5 exposure include
experimental studies conducted at near-ambient concentrations (Section 3.3.1). studies that examine the
role of PM25 exposure on COVID-19 infection and death (Section 3.3.2). and studies that examine
whether there are disparities in exposure to PM2 5 or the risk of PM2 5-related health effects by
race/ethnicity or socioeconomic status (SES).
The studies evaluated within the following sections represent only those studies most informative
in considering potential revisions to the PM NAAQS as defined by the scope of this Supplement
(Section 1.2.2). i.e., U.S. and Canadian epidemiologic studies and other studies that address key scientific
topics. Therefore, the scientific information presented within this section does not represent the full
multidisciplinary evaluation presented within the 2019 PM ISA, which would lead to the formation of a
causality determination. As a result, the summary sections for each health effects category convey how
the evidence from recent studies fits within the scientific conclusions of the 2019 PM ISA, and indicates
whether recent evidence supports (is consistent with), supports and extends (is consistent with and
reduces uncertainties), or does not support (is not consistent with) the causality determinations in the
2019 PM ISA.
6Throughout this supplement, as detailed in the Preface of the 2019 PM ISA (Section P.3.2.2), risk estimates from
epidemiologic studies examining short-term exposures are for a 10-|ig/m3 increase in 24-hour avg PM2 5
concentrations and long-term exposures are for a 5-|ig/m3 increase in annual concentrations, unless otherwise noted.
September 2021
3-1
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
3.1 Cardiovascular Effects
3.1.1 Short-Term PM2.5 Exposure
The following sections represent a summary of the evidence and the corresponding causality
determination for short-term PM2 5 exposure and cardiovascular effects presented within the 2019 PM
ISA (Section 3.1.1.1) along with an evaluation of recent epidemiologic studies that fall within the scope
of the Supplement (i.e., studies conducted in the U.S. and Canada) and were published since the literature
cutoff date of the 2019 PM ISA (Section 3.1.1.2). In addition, with the expansion of epidemiologic
studies using causal modeling methods, recent studies that employed such methods are also evaluated
(Section 3.1.1.3). which can further inform the relationship between short-term PM2 5 exposure and
cardiovascular morbidity. Lastly, a summary of the results of recent studies evaluated within the section is
presented in the context of the scientific conclusions detailed in the 2019 PM ISA (Section 3.1.1.4). The
evaluation of recent studies presented in this Supplement adds to the collective body of evidence
reviewed in the process of reconsidering the PM NAAQS.
3.1.1.1 Summary and Causality Determination from 2019 Integrated
Science Assessment for Particulate Matter
A large body of evidence evaluated in the 2019 PM ISA confirmed and extended the evidence
from the 2009 PM ISA (U.S. EPA. 2009) indicating a causal relationship between short-term PM2 5
exposure and cardiovascular effects. The strongest evidence in the 2009 PM ISA was from epidemiologic
studies of emergency department (ED) visits and hospital admissions for ischemic heart disease (IHD)
and heart failure, with supporting evidence from epidemiologic studies of cardiovascular mortality.
Changes in various measures of cardiovascular function in controlled human exposure studies provided
some biological plausibility for these associations. In addition, animal toxicological studies reporting
some evidence of reduced myocardial blood flow during ischemia, altered vascular reactivity, and ST
segment depression provided additional biological plausibility. In the 2019 PM ISA, evidence supporting
the causality determination included generally positive associations from epidemiologic studies of
hospital admissions and ED visits for cardiovascular-related effects, and in particular, for IHD and heart
failure. Results from these observational studies were supported by experimental evidence from
controlled human exposure and animal toxicological studies of endothelial dysfunction, as well as
endpoints indicating impaired cardiac function, increased risk of arrhythmia, changes in heart rate
variability (HRV), increases in blood pressure (BP), and increases in indicators of systemic inflammation,
oxidative stress, and coagulation. Additional results from observational panel studies, although not
entirely consistent, provided at least some evidence of increased risk of arrhythmia, decreases in HRV,
increases in BP, and ST segment depression. Thus, epidemiologic panel studies also provided some
September 2021
3-2
DRAFT: Do Not Cite or Quote
-------
1 support to the causality determination and to biological plausibility. Finally, epidemiologic studies of
2 cardiovascular-related mortality provided additional evidence that demonstrated a continuum of effects
3 from biomarkers of inflammation and coagulation, subclinical endpoints (e.g., HRV, BP, endothelial
4 dysfunction), ED visits and hospital admissions, and eventually death. The evidence evaluated in the 2019
5 PM ISA also reduced uncertainties from the previous review related to potential copollutant confounding
6 and limited biological plausibility for cardiovascular effects following short-term PM2 5 exposure.
7 Evidence supporting the causality determination for short-term PM2 5 exposure and cardiovascular effects
8 reached in the 2019 PM ISA is discussed below and summarized in Table 3-4. using the framework for
9 causality determinations described in the Preamble to the ISAs (U.S. EPA. 2015).
Table 3-1 Summary of evidence for a causal relationship between short-term
PM2.5 exposure and cardiovascular effects from the 2019 Integrated
Science Assessment for Particulate Matter.
Rationale for
Causality
Determination3
Key Evidence13
Key References
and Sections in
the 2019 PM ISAb
PM2.5 Concentrations
Associated with Effects0
(Hg/m3)
Consistent
epidemiologic evidence
from multiple studies at
relevant PM2.5
concentrations
Increases in ED visits and hospital
admissions for IHD and heart failure in
multicity studies conducted in the U.S.,
Canada, Europe, and Asia
Increases in cardiovascular mortality in
multicity studies conducted in the U.S.,
Canada, Europe, and Asia.
Section 6.1.2.1
Section 6.1.3.1
Section 6.1.9
5.8-18.6
5.8-18.0
Evidence from
controlled human
exposure studies at
relevant PM2.5
concentrations
Consistent changes in measures of
endothelial dysfunction
Generally consistent recent evidence for
small increases in measures of blood
pressure following CAPs exposure
Although not entirely consistent, there is
additional evidence of conduction
abnormalities, heart rate variability,
impaired heart function, systemic
inflammation/oxidative stress.
Section
6.1.13.2
Section
6.1.6.3
Section
6.1.4.3
Section
6.1.3.2
Section
6.1.10.2
Section
6.1.11.2
24-325
See Tables in identified
sections
Consistent evidence
from animal
toxicological studies at
relevant PM2.5
concentrations
Consistent changes in indicators of
endothelial dysfunction.
Additional evidence of changes in
impaired heart function, conduction
abnormalities/arrhythmia, heart rate
variability, blood pressure, systemic
inflammation/oxidative stress.
Section 6.1.13.3
Section 6.1.6.4
Section 6.1.4.4
Section 6.1.3.3
Section 6.1.11.3
168.7-510
See Tables in identified
sections
September 2021
3-3
DRAFT: Do Not Cite or Quote
-------
Table 3-1 (Continued): Summary of evidence for a causal relationship between
short-term PM2.5 exposure and cardiovascular effects from
the 2019 Integrated Science Assessment for Particulate
Matter.
Rationale for Key References „ PM2.5 Concentrations
Causality and Sections in Associated with Effects'
Determination3 Key Evidence13 the 2019 PM ISAb (|jg/m3)
Epidemiologic evidence
from copollutant
models provides some
support for an
independent PM2.5
association
The magnitude of PM2.5 associations
remain positive, but in some cases are
reduced with larger confidence intervals
in copollutant models with gaseous
pollutants. Further support from
copollutant analyses indicates positive
associations for cardiovascular mortality.
Recent studies that examined potential
copollutant confounding are limited to
studies conducted in Europe and Asia.
When reported, correlations with gaseous
copollutants were primarily in the low to
moderate range (r< 0.7).
Section 6.1.14.1
Consistent positive
epidemiologic evidence
for associations
between PM2.5
exposure and CVD ED
visits and hospital
admissions across
Positive associations consistently
observed across studies that used
ground-based (i.e., monitors), model
(e.g., CMAQ, dispersion models), and
remote sensing (e.g., AOD
measurements from satellites) methods,
including hybrid methods that combine
Klooa etal. (2014)
exposure measurement two or more of these methods,
metrics
Generally consistent
evidence for biological
plausibility of
cardiovascular effects
Strong evidence for coherence of effects
across scientific disciplines and biological
plausibility for a range of cardiovascular
effects in response to short-term PM2.5
exposure. Includes evidence for reduced
myocardial blood flow, altered vascular
reactivity, and ST segment depression.
Section 6.1.1
Figure 6-1
Uncertainty regarding
geographic
heterogeneity in PM2.5
associations
Multicity U.S. studies demonstrate
city-to-city and regional heterogeneity in
PIVhs-cardiovascular ED visit and hospital
admission associations. Evidence
supports the supposition that a
combination of factors including
composition and exposure factors may
contribute to the observed heterogeneity.
Section 6.1.2.1
Section 6.1.3.1
AOD = aerosol optical depth; CMAQ = Community Multiscale Air Quality; ED = emergency department; IHD = ischemic heart
disease; |jg/m3 = micrograms per cubic meter; PM2 5 = particulate matter with a nominal mean aerodynamic diameter less than or
equal to 2.5 |jm; ST = beginning of S wave to end of T wave.
aBased on aspects considered in judgments of causality and weight of evidence in causal framework in Table I and Table II of the
Preamble to the ISAs (U.S. EPA. 2015V
bDescribes the key evidence and references, supporting or contradicting, contributing most heavily to causality determination and,
where applicable, to uncertainties or inconsistencies. References to earlier sections indicate where the full body of evidence is
described in the 2019 PM ISA.
°Describes the PM25 concentrations with which the evidence is substantiated.
September 2021
3-4
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
The generally consistent, positive associations observed in numerous epidemiologic studies of ED
visits and hospital admissions for IHD, heart failure, and combined cardiovascular-related endpoints
contributed to the evidence supporting a causal relationship between short-term PM2 5 exposure and
CVD. Among this body of evidence, nationwide studies of older adults using Medicare reported positive
associations between PM2 5 concentrations and heart failure hospital admissions (2019 PM ISA,
Section 6.1.3.1). Consistent with the results of these large Medicare studies, additional multicity studies
conducted in the Northeast U.S. reported positive associations between short-term PM2 5 concentrations
and ED visits or hospital admissions for IHD (2019 PM ISA, Section 6.1.2.1), whereas studies conducted
in the U.S. and Canada reported positive associations between short-term PM2 5 concentrations and ED
visits for heart failure. Results from epidemiologic studies conducted in single cities contributed
additional support to the causality determination but are less consistent and reported both positive and
null associations between PM2 5 concentrations and these endpoints (2019 PM ISA, Section 6.1.2 and
Section 6.1.3). Overall, the body of IHD and heart failure epidemiologic evidence evaluated in the 2019
PM ISA agreed with the evidence from previous ISAs reporting mainly positive associations between
short-term PM2 5 concentrations and ED visits and hospital admissions. In addition, several controlled
human exposure, animal toxicological, and epidemiologic panel studies provided biologically plausible
evidence that PM2 5 exposure could result in IHD or heart failure through pathways that include
endothelial dysfunction, arterial thrombosis, and arrhythmia (2019 PM ISA, Section 6.1.1).
Epidemiologic panel studies evaluated in the 2019 PM ISA also supported biological plausibility for IHD
and heart failure endpoints by reporting some evidence of ST segment depression (2019 PM ISA,
Section 6.1.2.2), with a controlled human exposure study and animal toxicological study showing
decreased cardiac function following short-term PM25 exposure (2019 PM ISA, Section 6.1.3.2 and
Section 6.1.3.3).
Results from additional controlled human exposure studies published since the 2009 PM ISA also
support a causal relationship between short-term PM2 5 exposure and cardiovascular effects. The most
consistent evidence from these studies is for endothelial dysfunction as measured by changes in BAD or
FMD. More specifically, and in contrast to the 2009 PM ISA where a couple of studies did not find
changes in endothelial function, multiple studies evaluated in the 2019 PM ISA that examined the
potential for endothelial dysfunction reported an effect of PM2 5 on measures of blood flow (2019 PM
ISA, Section 6.1.13.2) relative to FA exposure. Nevertheless, all studies were not in agreement with
respect to the timing of the effect or the mechanism by which reduced blood flow occurred
(i.e., endothelial-independent vs. endothelial-dependent mechanisms). In addition to endothelial
dysfunction, controlled human exposure studies evaluated in the 2019 PM ISA that used CAPs, but not
filtered DE, generally reported evidence for small increases in blood pressure, although there were
inconsistencies across studies with respect to changes in SBP and DBP. It is notable however, that CAPs
studies evaluated in the 2019 PM ISA that reported increases in one measure of BP (e.g., SBP), but not
the other (e.g., DBP) was found to be statistically significant, that other measure of BP usually changed as
well, but the change was not found to be statistically significant (2019 PM ISA, Section 6.1.6.3). That
September 2021
3-5
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
said, the results of studies evaluated in the 2019 PM ISA are not in agreement with a couple of older
controlled human exposure studies that reported no appreciable changes in blood pressure following
short-term PM2 5 exposure. In addition, although not entirely consistent, there is further evidence from
controlled human exposure studies evaluated in the 2019 PM ISA for conduction
abnormalities/arrhythmia (2019 PM ISA, Section 6.1.4.3), changes in HRV (2019 PM ISA,
Section 6.1.10.2), changes in hemostasis that could promote clot formation (2019 PM ISA,
Section 6.1.12.2), and increases in inflammatory cells and markers (2019 PM ISA, Section 6.1.11.2).
Thus, although uncertainties remain, controlled human exposure studies are in coherence with
epidemiologic studies by demonstrating that short-term exposure to PM2 5 may result in the types of
cardiovascular endpoints that could lead to ED visits and hospital admissions.
Animal toxicological studies published since the 2009 PM ISA also support a causal relationship
between short-term PM2 5 exposure and cardiovascular effects. A study evaluated in the 2019 PM ISA
demonstrated decreased cardiac contractility and left ventricular pressure in mice which was coherent
with the results of epidemiologic studies that reported associations between short-term PM2 5 exposure
and heart failure (2019 PM ISA, Section 6.1.3.3). In addition, like in the controlled human exposure
studies, there was generally consistent evidence in animal toxicological studies for indicators of
endothelial dysfunction (2019 PM ISA, Section 6.1.13.3). Studies in animals also provided evidence for
changes in several other cardiovascular endpoints following short-term PM2 5 exposure. Although not
entirely consistent, these studies provided at least some evidence of conduction abnormalities and
arrhythmia (2019 PM ISA, Section 6.1.4.4), changes in HRV (2019 PM ISA, Section 6.1.10.3), changes
in BP (2019 PM ISA, Section 6.1.6.4), and evidence for systemic inflammation and oxidative stress (2019
PM ISA, Section 6.1.11.3). Finally, these toxicological studies also provided evidence indicating that
genetic background, diet, and PM composition may influence the effect of short-term PM2 5 exposure on
some of these health endpoints.
As outlined above, across the scientific disciplines, there is evidence for a continuum of
cardiovascular-related health effects following short-term exposure to PM2 5. These effects ranged from
relatively modest increases in biomarkers related to inflammation and coagulation, to subclinical CVD
endpoints such as endothelial dysfunction, to ED visits and hospital admissions for outcomes such as IHD
and heart failure. This continuum of effects is supported by epidemiologic studies that reported a
relatively consistent relationship between short-term PM2 5 exposure and CVD-related mortality. These
epidemiologic studies also reduced a key uncertainty from the 2009 PM ISA by providing evidence that
gaseous pollutants are not likely to confound the PM2 5-cardiovascular mortality relationship.
Taken together, the evidence described within the 2019 PM ISA extends the consistency and
coherence of the evidence base reported in the 2009 PM ISA and 2004 AQCD. Direct evidence for PM2 5
exposure-related cardiovascular effects can be found in several controlled human exposure and animal
toxicological studies. In coherence with these results are epidemiologic panel studies also finding that
PM2 5 exposure is associated with some of the same cardiovascular endpoints reported in controlled
September 2021
3-6
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
human exposure and animal toxicological studies. There is a limited number of studies evaluating some of
these endpoints, and there are some inconsistencies in results across some animal toxicological, controlled
human exposure, and epidemiologic panel studies, although this may be due to substantial differences in
study design, study populations, or differences in PM composition across air sheds. Nonetheless, the
results from these epidemiologic panel, controlled human exposure, and animal toxicological studies, in
particular those related to endothelial dysfunction, impaired cardiac function, ST segment depression,
thrombosis, conduction abnormalities, and BP, provide coherence and biological plausibility for the
consistent results from epidemiologic studies that reported positive associations between short-term PM2 5
concentrations and IHD and heart failure, and ultimately cardiovascular mortality. Overall, considering
the entire evidence base, there continues to be sufficient evidence to conclude that a causal
relationship exists between short-term PM2.5 exposure and cardiovascular effects.
3.1.1.2 Recent U.S. and Canadian Epidemiologic Studies
In addition to evaluating epidemiologic studies that examined the relationship between short-term
PM2 5 exposure and cardiovascular effects, the 2019 PM ISA characterized whether evidence supported
biologically plausible mechanisms by which short-term PM2 5 exposure could lead to cardiovascular
effects. This evaluation consisted of an assessment of animal toxicological, controlled human exposure,
and epidemiologic studies that examined a range of cardiovascular effects (2019 PM ISA, Section 6.1.1).
Plausible mechanisms were identified by which inhalation exposure to PM2 5 could progress from initial
events to apical events reported in epidemiologic studies (2019 PM ISA, Figure 6-1). The first proposed
pathway identified begins as respiratory tract inflammation leading to systemic inflammation. The second
proposed pathway identified involves activation of sensory nerve pathways in the respiratory tract that
lead to modulation of the autonomic nervous system. Once these pathways are initiated, there is evidence
from experimental and observational studies that short-term exposure to PM2 5 may result in a series of
pathophysiological responses that could lead to cardiovascular events such as emergency department
(ED) visits and hospital admissions for IHD and heart failure, and ultimately mortality (2019 PM ISA,
Figure 6-1).
Recent epidemiologic studies conducted in the U.S. and Canada build upon the strong
epidemiologic evidence base evaluated in the 2019 PM ISA, as well as previous assessments, that
provided the scientific rationale supporting a causal relationship between short-term PM2 5 exposure and
cardiovascular effects (Section 3.1.1.1). In addition to examining the relationship between short-term
PM2 5 exposure and specific cardiovascular outcomes (i.e., IHD and myocardial infarction
[Section 3.1.1.2.11. cerebrovascular disease and stroke [Section 3.1.1.2.21. heart failure
[Section 3.1.1.2.31. arrhythmia rSection 3.1.1.2.41. combined cardiovascular effects rSection 3.1.1.2.51.
and cardiovascular mortality rSection 3.1.1.2.61). analyses within these recent studies also further
examined issues relevant to expanding the overall understanding of the effect of short-term PM2 5
exposure on cardiovascular outcomes. Specifically, recent studies assessed potential copollutant
September 2021
3-7
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
confounding (Section 3.1.1.2.7) and the lag structure of associations (Section 3.1.1.2.8). The following
sections evaluate recent epidemiologic studies conducted in the U.S. and Canada that inform each of the
aforementioned topics within the context of the evidence base evaluated and summarized in the 2019 PM
ISA.
3.1.1.2.1 Ischemic Heart Disease and Myocardial Infarction
IHD is a chronic condition characterized by atherosclerosis and reduced blood flow to the heart.
Myocardial infarction (MI), more commonly known as a heart attack, occurs when heart tissue death
occurs that is secondary to prolonged ischemia. The effect of short-term PM2 5 exposure on acute MI,
complications from recent MI, and other acute or chronic IHD are generally evaluated using ICD codes
recorded when a patient is admitted or discharged from the hospital or ED (ICD9: 410-414 or ICD 10:
120-125). In experimental or epidemiologic panel studies, indicators of MI include ST segment
depression as measured by an electrocardiograph (ECG). The ST segment of an electrocardiogram
recorded by surface electrodes corresponds to the electrical activity of the heart registered between
ventricular depolarization and repolarization and is normally isoelectric.
The epidemiologic studies reviewed in the 2019 PM ISA (U.S. EPA. 2019) strengthened the
evidence characterized in the previous ISA (U.S. EPA. 2009). Most of the evidence for IHD and MI in the
2009 PM ISA was from multicity epidemiologic studies of ED visits and hospital admissions [i.e., the
U.S. Medicare Air Pollution Study (MCAPS) (Dominici et al.. 2006). a four-city study in Australia
(Barnett et al.. 2006). and a study among older adults in several French cities (Host et al.. 2008)1. The
positive associations reported in these studies were an important line of evidence in the 2009 PM ISA
concluding a causal relationship between short-term PM2 5 exposure and cardiovascular effects.
Uncertainties noted in the 2009 PM ISA with respect to exposure measurement error for those not living
near a PM2 5 monitor were reduced in the 2019 PM ISA with the consideration of studies that applied
hybrid exposure assessment techniques that combine land use regression data with satellite aerosol optical
depth (AOD) measurements and PM2 5 concentrations measured at fixed-site monitors to estimate PM2 5
concentrations. Further, compared to the 2009 PM ISA the evidence in the 2019 PM ISA was expanded to
include studies examining the association of short-term PM2 5 exposure ST segment depression in addition
to ED visits and hospital admissions for MI.
A recent study extends the evidence presented in the 2019 PM ISA through its examination of the
association between short-term PM2 5 exposure with hospital admissions for MI among the low income
and/or disabled Americans comprising the Medicaid population (deSouza et al.. 2021). deSouza etal.
(2021) reported a positive association between PM2 5 concentration (0-1 day average) and acute MI (OR:
1.1 [95% CI: 1.03, 1.7]). Recent studies have also addressed methodological challenges. Specifically,
Krall et al. (2018) conducted an analysis to elucidate the interpretation of potentially uncertain single-city
estimates. These authors used Poisson time-series regression to estimate the associations of 24-hour
average PM2 5 concentration (lag Day 0) with ED visits for IHD and other cardiovascular outcomes for
September 2021
3-8
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
each of the five cities included in their study. To estimate the association across all cities and the posterior
city-specific associations, they fit both traditional Bayesian hierarchical models in which associations
were estimated for each outcome separately, and multi-cause multicity (MCM) Bayesian hierarchical
models in which multiple cardiovascular outcomes were included in the model simultaneously so that a
shared between-city variation could be estimated. The authors also performed analyses to determine
whether their results were sensitive to the choice of exposure lag (Section 3.1.1.2.8) or the specification of
time trends. The associations between 24-hour PM2 5 concentration and IHD ED visits in the traditional
multicity model was 1.009 (95% Posterior Interval [PI]: 0.993, 1.025). The comparable association (1.009
[95% PI: 0.998, 1.022]) using MCM was more precise (i.e., narrower confidence intervals). As expected,
the city specific estimates were relatively uncertain and heterogeneous across cities when there were a
small number of daily ED visits. In another recent study, Leiseretal. (2019) designed an analysis to
examine the association between short-term PM2 5 exposure and IHD and MI hospital admissions in
which the competing risk of mortality was controlled and differences across sex and age categories were
examined. These authors used Medicare data for residents, 65 years and older, of the contiguous counties
of the Wasatch Front in Utah to examine the association between PM2 5 concentration and cardiac hospital
re-admissions within 30 days of an index hospitalization, controlling for competing mortality risk. These
authors found an association of 3-day average PM2 5 concentration (lag 0-2 day) with IHD (HR: 1.03
[95% CI: 0.96, 1.12]) and MI (HR: 1.03 [95% CI: 0.92, 1.16]). Confidence intervals for the age- and
sex-stratified results, which were conducted to evaluate potential modification of the association, were
generally overlapping.
In another analysis of older adults using Medicare data, Wei et al. (2019) estimated the
association of short-term PM2 5 exposure with MI hospital admissions and a range of other health
conditions, including some diseases that are rarely studied in relationship to PM2 5 exposure. Hospital
admission data were ascertained using discharge data recorded for Medicare inpatient hospital claims in
the continental U.S. (2000-2012). Rather than report a relative risk estimate, the authors reported the
absolute risk per 10 million person-days associated with each 1 unit increase in lag 0-1 PM2 5
concentration (i.e., 0.29 [95% CI: 0.17, 0.40]).
Recent single city studies also add to the evidence base presented in the 2019 PM ISA. Liu et al.
(2020) examined the modification of the association between short-term PM2 5 exposure and MI hospital
admissions by long-term NO2 exposure. These authors performed a case-crossover study to estimate the
association of short-term exposure to PM2 5 among individuals living in Calgary neighborhoods with
higher long-term N02 exposure (2004-2012). The OR for the association between 0-2-day average
PM2 5 concentration with hospital admissions for MI among the entire population was 1.03 (95% CI: 0.96,
1.12). The association was null in the lowest tertile of long-term NO2 concentration (OR: 0.94 [95% CI:
0.86, 1.18]), but increased with increasing NO2 concentration tertile (tertile 2, OR: 1.04 [95% CI: 0.94,
1.18]) and (tertile 3, OR: 1.10 [95% CI: 1.00, 1.19]). In addition, an extended analysis of a study reviewed
in the 2019 PM ISA supports previous results that found a positive association between short-term PM2 5
exposure and ST elevation myocardial infarction (STEMI) (Evans et al.. 2017). Specifically, Evans et al.
September 2021
3-9
DRAFT: Do Not Cite or Quote
-------
1 (2017) performed a case-crossover analysis to examine the relationship between short-term PM2 5
2 concentration and STEMI in acute coronary syndrome or unstable angina patients (n = 362) in Monroe
3 County, NY (2007-2012). The association between previous 1-hour PM2 5 concentration and STEMI
4 reported by these authors (OR: 1.25 [95% CI: 0.99, 1.59]) was virtually identical to the association (OR:
5 1.26 [95% CI: 1.01, 1.57]) reported in a previous analysis of this population conducted by Gardner et al.
6 (2014) that reported fewer patients (n = 338) and a shorter follow-up time (2007-2010).
7 Results of studies of IHD and MI included in the 2009 PM ISA, the 2019 PM ISA and recent
8 studies published since the literature cutoff date of the 2019 PM ISA are summarized in Figure 3-1.
9 Overall, recent studies support and extend the findings of the 2019 PM ISA with additional studies
10 reporting positive associations between short-term PM2 5 exposure and both IHD and MI hospital
11 admissions and ED visits.
September 2021
3-10
DRAFT: Do Not Cite or Quote
-------
Study
Outcome
Location
Mean (pg/m3)
Dominici et al. (2006)
IHD
204 U.S. Counties
13.4
Barnett et
al. (2006)
IHD
4 Australian Cities
8.1-9.7
Host et al.
(2007)
IHD
6 French Cities
13.8-18.6
TBell et al.
(2015)
IHD
213 U.S. Counties
12.3
TKIoog et al. (2014)
IHD
7 Mid-Atlantic States
11.9
Zanobetti et al. (2006)
Ml
Boston, MA
11.1
TBell et al.
(2015)
Ml
213 U.S. Counties
12.3
TZanobetti et al. (2009)
Ml
26 U.S. Cities
15.3
"DeSouza et al. (2019)
Ml
U.S. (National)
11.5
THaley et al. (2009)
8 New York Cities
5.8
'Krall et al
. (2018)
IHD
5 Cities, U.S.
10.8-15.4
tHsu et al
. (2017)
IHD
4 NY Regions
NR
tTalbott el
t al. (2014)
IHD
7 U.S. States
6.5-12.8
tOstro et i
al. (2016)
IHD
8 CA counties
16.5
tMilojevic etal. (2014)
IHD
15 Conurbations, UK
10
tSarnat et
al. (2015)
IHD
St. Louis, MO
18
tMilojevic
etal. (2014)
Ml
15 Conurbations, UK
10
tStieb et al. (2009)
Angina/Ml
6 Canadian Cities
6.7-9.8
tSzyskowicz et al. (2009)
Angina
6 Canadian Cities
8.3
tWeichenthal et al. (2016)
Ml
16 Cities Ontario
6.9
"Liu et al. i
(2020)
Ml
Alberta, Canada
9.79
tTalbott etal. (2014)
TOstro et al. (2016)
TRich et al. (2010)
tPope et al. (2015)
*Leiser et al. (2019)
'Evans et al. (2017)
tGardner et al. (2014)
Ml
IHD
7 U.S. States
8 CA counties
New Jersey state
4 counties, Utah, U.S.
Rochester, NY
Rochester, NY
7.6-12.3
9.9-10.6
10.96
17.1
Lag
2
0-1
0-1
0
0-1
0
0-1
0
Notes
Ages 65+
Ages65+
Ages 65+
Ages65+
Ages 65+
Ages65+
Ages 65+
Ages65+
Medicaid
Wffi!!cTldly
NYC, Long Island & Hudson
Adirondack & North.
& Binahamton
u
0-2
0
0-23h
Massachusetts
Mlw Hampshire
Ulster0
Washington
Florida
Massachusetts
New Jersey
New Hampshire
New Mexico
lfe$on
STEMI
M,
i
i
i ¦
o
lO
t
0.8
0.9
1.1
1.2
13
Relative Risk (95% CI)
Source: Update of Figure 6-2, 2019 PM ISA.
Note: tStudies published since the 2009 PM ISA. *U.S. and Canadian studies published since the literature cutoff date (~January
2018) for the 2019 PM ISA. IHD = ischemic heart disease, MCM = multi-cause multicity; Ml = myocardial infarction, NR = not
reported; NSTEMI = non-ST segment elevation Ml, STEMI = ST segment elevation Ml. Risk estimates are standardized to a
10 |jg/m3 increase in PM25 concentrations.
Figure 3-1 Results of studies of short-term PM2.5 exposure and hospital
admissions and emergency department visits for ischemic heart
disease.
3.1.1.2.2 Cerebrovascular Disease and Stroke
1 Cerebrovascular disease (CBVD) typically includes conditions classified under ICD10 codes
2 160-169 (ICD9: 430-438) such as hemorrhagic stroke (HS), cerebral infarction (i.e., ischemic stroke [IS])
3 and occlusion of the precerebral and cerebral arteries. IS results from an obstruction within a blood vessel
4 that supplies oxygen to the brain, potentially leading to infarction, and accounts for 87% of all strokes
5 (Goldberger et al.. 2008). Hemorrhagic stroke is less common but results in a disproportionate number of
6 fatalities. The HS subtype results from a brain aneurysm or leaking vessel in the brain and can be further
7 categorized by brain region (e.g., intracerebral, or subarachnoid). Older age, female sex, smoking,
8 obesity, and prior stroke are known risk factors for stroke and should be considered in epidemiologic
September 2021
3-11
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
analysis. Comorbidities that increase stroke risk but may also be associated with PM2 5 exposure include
hypertension, diabetes, CHD, and atrial fibrillation. The 2009 PM ISA and the 2019 PM ISA described
inconsistent results across epidemiologic studies that considered the relationship between short-term
PM2 5 exposure and ED visits and hospital admissions for CBVD, with most studies reporting a lack of an
association. Evidence relating to various stroke subtypes was also inconsistent. Results from recent
studies of the association between short-term PM2 5 concentration and stroke expand the evidence but
remain inconsistent overall. Specifically, a study of Medicaid recipients found a large magnitude positive
association, while several analyses of established cohorts (Health Professionals Follow-Up [HPFU] study,
Women's Health Initiative [WHI], REasons for Geographic and Racial Differences in Stroke
[REGARDS]) report null or inverse associations with stroke regardless of subtype.
Recent studies that analyze data from participants enrolled in several established cohort studies
expand the evidence pertaining to stroke subtype. Fisher et al. (2019) estimated the associations of
short-term PM2 5 exposure with several stroke types, which were ascertained through self-report and
expert medical record review, among men enrolled in the HPFU study. The authors reported no evidence
of positive associations between lag 0 to 3-day average PM2 5 concentration and total stroke (OR: 0.93
[95% CI: 0.78, 1.11]), ischemic (OR: 0.99 [95% CI: 0.81, 1.20]), hemorrhagic (OR: 0.80 [95% CI: 0.50,
1.27]), undetermined type (OR: 0.56 [95% CI: 0.26, 1.24]), and non-fatal stroke (OR: 1.05 [95% CI: 0.84,
1.27]). The authors also evaluated whether factors including age, BMI, smoking status, diabetes mellitus,
hypertension, hypercholesterolemia, and current aspirin use potentially modified the associations with IS
or HS; however, the number of stroke events within each strata was small and no statistical evidence of
heterogeneity between stratified estimates was reported based on chi-square tests of model homogeneity.
Sun et al. (2019) also estimated the association of short-term PM2 5 with total, hemorrhagic, and ischemic
stroke but studied a different population (i.e., post-menopausal women enrolled in the WHI study). Stroke
was ascertained through self-report and physician adjudication. Three-day average PM2 5 concentration
(lag 0-2) was not associated with total (OR: 0.98 [95% CI: 0.92, 1.02]), ischemic (OR: 0.96 [95% CI:
0.90, 1.02]), or hemorrhagic stroke (OR: 1.02 [95% CI: 0.90, 1.17]) in this study. The authors also
conducted stratified analysis to examine whether associations varied across categories of age at stroke
onset, U.S. census region, smoking status, body mass index, and prior history of diabetes mellitus,
hypertension, heart or circulation problems, or arterial fibrillation at enrollment. Across these different
stratified analyses, only when examining the stratum for obese women was there some evidence that the
association of total stroke with PM2 5 may be increased. Finally, McClure et al. (2017) performed a
case-cross over analysis among participants in the REGARDS study to determine the association between
PM2 5 exposure at single-day lags (1,2, and 3 day lags) and stroke ascertained through self-report
followed by a medical record. The REGARDS study oversampled participants in several southern states
where stroke risk is high among Black residents in order to study geographic and racial differences in
stroke. PM2 5 concentration was dichotomized (<12 (.ig/nr1 vs. 12 to 150.4 (.ig/nr1) and the odds of stroke in
the higher category was compared to the odds of stroke in the lower category. After adjustment for
temperature and relative humidity, no association was reported between PM2 5 exposure and stroke,
regardless of the lag examined (OR: 0.99 [95% CI: 0.83, 1.19], lag 1). This finding persisted regardless of
September 2021
3-12
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
stroke subtype or exposure lag. Overall, analyses from three established and diverse cohorts did not
present evidence of an association between short-term PM2 5 exposure and stroke.
Unlike the analyses described above, deSouza et al. (2021) examined the relationship between
PM2 5 concentration (0-1 day average) and hospital admissions for IS among the low income and/or
disabled Americans comprising the Medicaid population. The OR for the association between PM2 5 (lag
0-1 day average) and IS was 1.12 (95% CI: 1.04, 1.20). In a study designed to gain an understanding of
the heterogeneity in results across single-city studies, Krall et al. (2018) examined the associations of
24-hour PM2 5 concentration (lag Day 0) with ED visits for stroke. These authors estimated the
associations across five cities using a traditional Bayesian hierarchical approach and a MCM Bayesian
hierarchical model in which all outcomes were modelled simultaneously. The association of 24-hour
PM2 5 concentration (lag 0) with ED visits for stroke was 1.008 (95% CI: 0.984, 1.034) in the traditional
multicity model and more precise (i.e., narrower confidence intervals) 1.008 (95% PI: 0.995, 1.021) in the
MCM model. However, in a single city study of recurrent IS in Nueces county Texas, Wing et al. (2017)
did not report evidence of an association between PM2 5 concentration during the previous day (lag 1), and
the odds of recurrent stroke (OR: 0.95 (95% CI: 0.71-1.28]).
Results of studies of short-term exposure to PM2 5 and ED visits or hospital admissions for stroke
included in the 2009 PM ISA, the 2019 PM ISA and recent studies published since the literature cutoff
date of the 2019 PM ISA are summarized in Figure 3-2. Consistent with the evidence evaluated in the
2019 PM ISA, some recent studies report evidence of a positive association with stroke while others
report null or inverse associations.
September 2021
3-13
DRAFT: Do Not Cite or Quote
-------
Study
*deSouza etal. (2019)
'Fisher et al. (2019)
*Sun et al. (2019)
"McLure et al. (2017)
tKloog et al. (2012)
tKloog et al. (2014)
tMilojevic etal. (2014)
•Krall et al. (2018)
tVilleneuve etal. (2012)
tWellenius et al. (2012)
TLisabeth et al. (2008)
TWingetal. (2015)
•Winget al. (2017)
tO'Donnell et al. (2011)
Outcome
Ischemic Stroke
Stroke
Ischemic Stroke
Hemorrhagic Stroke
Undetermined
non-fatal
Stroke
Ischemic Stroke
Hemorrhagic Stroke
Stroke
Stroke
Stroke
Stroke
Stroke
Stroke
Hemorrhagic Stroke
Ischemic Stroke
Transient Ischemic Attacks
Acute Ischemic Stroke
Ischemic Stroke and
Transient Ischemic Attacks
Ischemic Stroke
Ischemic Stroke
Acute Ischemic Stroke
Location Mean
(pg/m3)
Medicaid, U.S. 11.5
REGARDS, U.S. NR
6 New England States 9.6
7 Mid-Atlantic States 11.9
15 Conurbations, UK 10 (Med.)
5 Cities, U.S. 10.8-15.4
Denver, CO 8.1
Boston, MA + 10.2
Nueces County, TX 7
Nueces County, TX 7.7 (Med)
BASIC-Nueces Co TX 7.7 (Med.)
8 Cities, Ontario Canada 6.9
Lag
0-1 d avg
0-3 d avg
10.8(Med) 3d avg
Lag d 1
0-47h
High v Low (<12 pg/m3)
Age 65+
Age 65+
Traditional Multicity
MCM Multicity
Age 45+
Recurrent
Subjects < 20km
from monitor
0.5 0.7 0.8 1 1.2
Relative Risk (95% CI)
1.4
Source: Update of Figure 6-5, 2019 P MISA.
Note: tStudies published since the 2009 PM ISA. *U.S. and Canadian studies published since the literature cutoff date (~January
2018) for the 2019 PM ISA. NR = not reported. Risk estimates are standardized to a 10 |jg/m3 increase in PM25 concentrations.
Figure 3-2 Results of studies of short-term PM2.5 exposure and hospital
admissions and emergency department visits for stroke.
3.1.1.2.3 Heart Failure
1 Heart failure (HF) refers to a set of conditions in which pumping action of the heart is weakened.
2 In congestive heart failure (CHF), the flow of blood from the heart slows, failing to meet the oxygen
3 demands of the body, and returning blood can back up, causing swelling or edema in the lungs or other
4 tissues (typically in the legs and ankles). The effect of short-term PM2 5 exposure on people with
5 CHF—which is a chronic condition—is generally evaluated using ICD codes recorded when a patient is
6 admitted or discharged from the hospital or ED. The relevant diagnostic codes for heart failure are ICD9
7 428 and ICD10 150. These codes encompass left, systolic, diastolic, and combined heart failure. Similar to
8 the other cardiovascular outcomes, the majority of the evidence in the 2009 PM ISA was from
9 epidemiologic studies of hospital admissions and ED visits [i.e., multicity studies in the U.S.(Dominici et
September 2021
3-14
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
al.. 2006) and Australia (Barnett et al.. 2006)1. Studies evaluated in the 2019 PM ISA strengthened this
line of evidence with additional multicity epidemiologic studies conducted in the U.S., Canada, and
Europe generally reporting positive associations between short-term PM2 5 exposure and hospital
admissions and ED visits for HF. Results from single-city studies tended to be less consistent. Several
recent studies add to the body of evidence providing additional support for a positive association between
short-term PM2 5 exposure and ED visits and hospital admissions for exacerbations of HF.
Recent studies conducted examined the association of short-term PM2 5 exposure with
readmission to the hospital for HF within 30 days of an index hospitalization. Leiseretal. (2019) used
Medicare data to examine the association of short-term PM2 5 exposure and cardiac hospital re-admissions
among older adults within 30 days of the index hospitalization. These authors reported an association of
3-day average PM25 concentration (lag 0-2 day) with readmission for HF (HR: 1.10 [95% CI: 1.03,
1.19]). Confidence intervals of the age and sex stratified results were generally overlapping. In another
study of 30-day hospital readmission, Wvatt et al. (2020c) characterized the association of short-term
PM2 5 exposure with CHF among end stage renal disease patients (i.e., those undergoing hemodialysis).
Both readmission within 1 to 7 days and readmission between 8 to 30 days was evaluated. The RR for the
association of 24-hour PM2 5 concentration (lag 0) with HF readmission within 1-7 days was 1.37 (95%
CI: 1.14, 1.60). The association with late readmission (8-30 days) was 1.41 (95% CI: 1.25, 1.58). In
another unique population, deSouza et al. (2021) estimated the association of PM2 5 concentration
(0-1 day average) with CHF hospital admissions among the low income and/or disabled Americans
comprising the Medicaid population. The OR for the association between PM2 5 (lag 0-1 day average)
and CHF was 1.10 (95% CI: 1.04, 1.16).
In addition to the studies described above that focus on 30-day hospital readmission, Krall et al.
(2018) examined the association of 24-hour PM2 5 concentration (lag Day 0) with ED visits for CHF in an
analysis in five cites that was designed to compare methods used for multicity studies. These authors
estimated the multicity associations using a traditional Bayesian hierarchical approach and a MCM
Bayesian hierarchical model in which all outcomes were modelled simultaneously. The association of
24-hour PM2 5 concentration (lag 0) with CHF was 1.003 (95% CI: 0.986, 1.021) in the traditional
multicity model and 1.003 (95% PI: 0.992, 1.016) in the MCM model. In another study, Wei et al. (2019)
used Medicare data (i.e., inpatient hospital claims) to estimate the association of short-term PM2 5
exposure with CHF hospital admissions in the continental U.S. between 2000 and 2012. Rather than
report a relative risk estimate, the authors reported the absolute increase in risk of admission to hospital
per 10 million person-days associated with each 1 (.ig/ni1 increase in lag 0-1 PM2 5 (i.e., 0.68 [95% CI:
0.52 to 0.84]).
Results of studies of HF included in the 2009 PM ISA, the 2019 PM ISA and recent studies
published since the literature cutoff date of the 2019 PM ISA are summarized in Figure 3-3. Overall,
these studies support and extend the limited evidence in the 2019 PM ISA, reporting positive associations
between short-term PM2 5 exposure and HF.
September 2021
3-15
DRAFT: Do Not Cite or Quote
-------
Study
Dominici et al. (2006)
Barnett et al. (2006)
Outcome
HF
HF
TBell etal. (2015) HF
TZanobetti et al. (2009) HF
"DeSouza etal. (2019) HF
TTalbott etal. (2014) HF
THsu etal. (2017)
THaley et al. (2009)
"Krall et al. (2018)
HF
tOstro etal. (2016) HF
tStieb et al. (2009) HF
TMilojevic etal. (2014) HF
tRodopoulou etal. (2015) HF & HHD
TSarnat et al. (2015) HF
Location
204 U.S. Counties
4 Australian Cities
213 U.S. Counties
26 U.S. Cities
Medicaid, U.S.
7 U.S. States
Mean
((jg/m3)
13.4
12.3
15.3
11.5
6.5-12.8
Lag
0
0-2
0-2
8 New York Cities 5.8 0
5 Cities, U.S. 10.8-15.4 0
8 CA counties 16.5 0
6 Canadian Cities 8.2 0
15 Conurbations, UK 10 (Med.) 0-4
Little Rock, AR 12.4 1
St. Louis, MO 18 0-2
*Leiseret al. (2019)
•Wyattetal. (2020)
30 d Re-admit HF Wasatch Front, Utah 10.96
30 d Re-admit HF ESRD Patient 9.3
0-2
Notes
Ages 65+
Ages 14-64
Ages 65+
Florida
Massachusetts
New Jersey
New Hampshire
New Mexico
New York
Washington
NYC, Long Island & Hudson
Adirondack & North
Mohawk Valley & Binghamton
Central & Western NY
Traditional Multicity
MCM Multicity
Early (0-7 d)
Late (8-30 d)
0.9 1 1.1
Relative Risk (95% CI)
1.2
1.3
1.4
1.5
Source: Update of Figure 6-3, 2019 PM ISA.
Note: tStudies published since the 2009 PM ISA. *U.S. and Canadian studies published since the literature cutoff date (~January
2018) for the 2019 PM ISA. ESRD = end-stage renal disease; HF = heart failure, HHD = hypertensive heart disease, NR = not
reported; re-HA = readmission to the hospital for heart failure. Risk estimates are standardized to a 10 |jg/m3 increase in PM25
concentrations.
Figure 3-3 Results of studies of short-term PM2.5 exposure and hospital
admissions and emergency department visits for heart failure.
3.1.1.2.4 Arrhythmia
1 In epidemiologic studies, the association between short-term PM2 5 exposure and arrhythmia is
2 generally evaluated using ICD codes (ICD9 427 or ICD10 149.9) for hospital admissions and ED visits.
3 Out-of-hospital cardiac arrests (OHCA) that typically result from ventricular arrhythmia were evaluated
4 with the body of evidence pertaining to arrhythmia. Overall, the evidence evaluated in the 2009 PM ISA
5 and the 2019 PM ISA was limited. However, in the 2019 PM ISA there were was some evidence from
6 epidemiologic panel studies of an association between short-term PM2 5 exposure and potential indicators
7 of arrhythmia (e.g., ectopic beats and tachycardia). The small number of recent studies support a positive
8 association of short-term PM2 5 exposure with arrythmias.
September 2021
3-16
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
Recent studies examined the association of short-term exposure to PM2 5 and dysrhythmia adding
to the limited evidence evaluated in the 2019 PM ISA. Wvatt et al. (2020c) examined 30-day hospital
re-admission among end stage renal disease patients (i.e., those undergoing hemodialysis). Both early
re-admission within 1 to 7 days and later readmission after 8 to 30 days was evaluated. The RR for the
association of 24-hour PM2 5 concentration (lag 0) with dysrhythmia and conduction disorder
readmissions within 1-7 days was 1.48 (95% CI: 1.23, 1.74). The association with late re-admission
(8-30 days) was 1.31 (95% CI: 1.13, 1.50). In another study of 30-day hospital re-admission, Leiser et al.
(2019) estimated the association of PM2 5 concentration and cardiac arrhythmia and examined differences
across sex and age in models that adjusted for the competing risk of mortality. These authors reported an
inverse association of 3-day average PM2 5 concentration (lag 0-2 days) with re-admission for
dysrhythmia or arrhythmia (HR: 0.88 [95% CI: 0.75, 1.02]). Confidence intervals of the age and sex
stratified results were generally overlapping and did not provide evidence of effect modification. KraUet
al. (2018) examined the association of 24-hour PM2 5 concentration (lag Day 0) with CVD ED visits in an
analysis designed to compare methods for multicity analyses. These authors estimated the associations
with CVD ED visits including visit for dysrhythmia across five cities using a traditional Bayesian
hierarchical approach and a MCM Bayesian hierarchical model in which all outcomes were modelled
simultaneously. The association between 24-hour PM2 5 concentration (lag 0) with ED visits for
dysrhythmia was 1.009 (95% CI: 0.993, 1.023) in the traditional Bayesian hierarchical model and 1.009
(95% PI: 0.998, 1.022) in the MCM Bayesian hierarchical model. Finally, Wei et al. (2019) estimated the
association of short-term PM2 5 exposure with arrhythmia hospital admissions in a study using discharge
data recorded for Medicare inpatient hospital claims in the continental the U.S. (2000-2012). Rather than
report a relative risk estimate, the authors reported the absolute increase in risk of admission to hospital
per 10 million person-days associated with each 1 (.ig/ni1 increase in lag 0-1 PM2 5 concentration
(i.e., 0.26 [95% CI: 0.13 to 0.38]).
Results of studies of arrythmia included in the 2009 PM ISA, the 2019 PM ISA and recent studies
published since the 2019 PM ISA are summarized in Figure 3-4. Overall, these studies support and extend
the limited evidence in the 2019 PM ISA.
September 2021
3-17
DRAFT: Do Not Cite or Quote
-------
Outcome
Arrhythmia
Arrhythmia
Arrhythmia
Study
Dominici et al. (2006)
tBell et al. (2015)
tTalbott eta I. (2014)
THsu et al. (2017)
t M ilojevic et a I. (2014) Arrhyth mia
Atrial Fibrillation
tStieb et a I. (2009)
tHaley et al. (2009)
tOstro et al. (2016)
*Krall et al. (2018)
Arrhythmia
Arrhythmia
Dysrhythmia
Dysrhythmia
tSarnatet al. (2015) Arrhythmia
¦fRodopoulou etal. (2015) Arrhythmia
*Wyatt et al. (2020)
204 U.S. Counties
Mean
(pg/m3)
13.4
213 U.S. Counties 12.3
7 U.S. States
Lag Notes
0 Age65+
0 Age 65+
Massachusetts
New Jersey
New York
NYC, Long Island & Hudson
Central & Western NY
15 Conurbations, UK 10 (Med.) 0-4
6 Canadian Cities
8 New York Cities 5.8
ED Visits 5 Cities
10.8-15.4 Lagdl Traditional Multicity
Lag d 1 MCM Multicity
Cold Season
Warm Season
ESRD Patients
0-2 Early Re-admission (1-7 d)
0-2 Late Re-admission (8-30 d)
0.8 0.9 1 1.1
Relative Risk (95% CI)
1.2
1.3
1.4
1.5
Source: Update of Figure 6-4, 2019 PM ISA.
Note: jStudies published since the 2009 PM ISA. *U.S. and Canadian studies published since the literature cutoff date (~January
2018) for the 2019 PM ISA. HF = heart failure, HHD = hypertensive heart disease, NR = not reported. Risk estimates are
standardized to a 10 |jg/m3 increase in PM25 concentrations.
Figure 3-4 Results of studies of short-term PM2.5 exposure and hospital
admissions and emergency department visits for arrhythmia.
3.1.1.2.5 Combinations of Cardiovascular-Related Outcomes
1 In addition to analyses of individual cardiovascular diseases (e.g., MI, stroke, and HF),
2 epidemiologic studies examined cardiovascular diseases (CVD) in aggregate (i.e., specific combination of
3 cardiovascular diseases). The 2009 PM ISA and the 2019 PM ISA reviewed multicity studies of adults
4 ages 65 years and older that provided strong evidence of an association lYBell et al.. 2008; Host et al..
5 2008; Barnett et al.. 2006); Table 6-19 of the 2019 PM ISA], Studies of aggregate CVD have larger case
6 counts than studies of specific CVDs, potentially providing statistical power needed to perform stratified
7 analyses. Several recent studies examine the association between short-term exposure to PM2 5 and CVD
8 hospital admissions and ED visits, and report results that are generally consistent with studies evaluated
9 in the 2019 PM ISA.
September 2021
3-18
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
In a study of low income and/or disabled Americans enrolled in Medicaid deSouza et al. (2021)
estimated the association of PM2.5 concentration (0-1 day average) with cardiovascular hospital
admissions. The association of PM2 5 concentration (0-1 day average) with all CVD hospital admissions
was 1.09 (95% CI: 1.06, 1.11). The association with all CVD hospital admissions was larger in magnitude
at PM2 5 concentrations <25 (ig/m3 (OR: 1.13 [95% CI: 1.09, 1.16]). The association was similar among
older and younger adults (OR: 1.09 [95% CI: 1.06, 1.13]) among those <65 years old 1.08 (95% CI: 1.06,
1.09) versus among those >65 years old. In another study, Wvatt et al. (2020c) examined hospital
admissions among end stage renal disease patients (i.e., those undergoing hemodialysis). Same-day PM2 5
concentration (lag 0) was associated with an increase in the risk of CVD hospital admissions in this
population (RR: 1.09 [95% CI: 1.02, 1.17]).
Some recent studies of ED visits report null associations between short-term PM2 5 concentration
and aggregated CVD outcomes in adjusted models. Krall et al. (2018) examined the associations of
24-hour PM2 5 concentration (lag Day 0) with CVD ED visits, estimating the association across five cities
using a traditional Bayesian hierarchical approach. A null association between 24-hour PM2 5
concentration (lag 0) with CVD ED visits was observed (1.0 [95% CI: 0.992, 1.009]) while positive
associations were reported for specific cardiovascular outcomes evaluated. Ye et al. (2018) performed a
study to estimate the association between short-term exposure to PM2 5 components that are not routinely
measured, including water soluble metals, and CVD ED visits for a five-county area of Atlanta during the
period 1998 to 2013. In a single-pollutant model, these authors reported a positive association of 24-hour
PM2 5 concentration (lag 0) with CVD ED visits; however, the association was null after adjustment for
water soluble iron (WS Fe), which may be an indicator for certain aspects of traffic pollution.
3.1.1.2.6 Cardiovascular Mortality
The limited assessment of cause-specific mortality in recent studies provides similar results to
previously evaluated studies demonstrating a consistent relationship between short-term PM2 5 exposure
and cardiovascular mortality. Consistent with studies evaluated in the 2019 PM ISA, recent studies
indicate that associations between short-term PM2 5 exposure and cardiovascular mortality are relatively
unchanged in copollutant models and may be larger in magnitude in the presence of some co-occurring
pollutants (i.e., oxidant gases). In addition, factors that have been shown to vary between cities and
regions of the U.S., such as housing characteristics, have been shown to explain some of the city-to-city
and regional variability observed in PM2 5-mortality associations in multi-city epidemiologic studies. The
continued assessment of the concentration-response (C-R) relationship between short-term PM2 5
exposure and mortality further supports a linear relationship, with less confidence in the shape at
concentrations below 5 (ig/m3.
September 2021
3-19
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
3.1.1.2.7
Potential Copollutant Confounding
In the examination of potential confounding of the relationship between short-term PM2 5
exposure and cardiovascular effects by exposure to copollutants, it is informative to evaluate whether
PM2 5 risk estimates are changed in copollutant models. The evidence reviewed in the 2019 PM ISA
represented an expanded set of studies that performed analyses using two-pollutant models. These studies
addressed a data gap, generally supporting an association of PM2 5 with cardiovascular related health
effects that was independent of copollutant exposures (i.e., O3, NO2, SO2, CO, and PM10-2.5). In addition
to copollutant models, a limited number of studies that examined the joint effects of multiple pollutants
provided information on the role of PM2 5 within the complex air pollution mixture. This limited evidence
neither consistently nor coherently indicated a stronger or weaker effect of combined exposure to PM2 5
and another pollutant compared to exposure to a single pollutant alone (Lubcn et al.. 2018). A small
number of recent studies expand upon the overall assessment of potential confounding of the relationship
between short-term PM2 5 exposure and cardiovascular effects by copollutants.
In the study of Medicaid recipients conducted by deSouzaetal. (2021) the positive
single-pollutant association (OR: 1.09 [95% CI: 1.06, 1.11]) between all CVD and short-term PM2 5
persisted in a two-pollutant model adjusted for ozone (OR: 1.10 [95% CI: 1.07, 1.12]). By contrast, Ye et
al. (2018) reported that the single-pollutant association between short-term exposure to PM2 5 was null
after adjustment for water soluble iron (WS Fe), which is a PM2 5 component that may be an indicator for
certain aspects of traffic pollution. Wing et al. (2017) reported no association between short-term PM2 5
exposure and recurrent stroke in both single- and two-pollutant model that were adjusted for ozone.
An examination of the association between short-term PM2 5 exposure and cardiovascular effects
across different lag days can inform whether PM2 5 elicits an immediate, delayed, or prolonged effect on
these endpoints, and whether the effect of PM2 5 is consistent across cardiovascular endpoints. The
evidence reviewed in the 2019 PM ISA supported an immediate effect of short-term PM2 5 exposure on
hospital admissions and ED visits for aggregate CVD outcomes, IHD, HF, and OHCA, as well as for
cardiovascular mortality. This evidence came from the evaluation of both single-day and multiday lags, as
well as studies that evaluated subdaily lag periods. By contrast, the studies evaluated in the 2019 PM ISA
did not provide evidence of a consistent lag period for the association of short-term PM2 5 exposure with
CBVD and arrhythmia. Overall, stronger associations in terms of magnitude and precision were reported
for immediate lags for most cardiovascular-related outcomes, and the associations tended to be stronger
for immediate multiday lag periods (i.e., 0-1, 0-2) compared to immediate single-day lag periods (i.e., 0,
Several recent studies conducted analyses to determine if results were sensitive to the choice of
exposure lag. Overall, the available studies continue to support an immediate effect of short-term PM2 5
3.1.1.2.8
Lag Structure of Associations
1).
September 2021
3-20
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
exposure on MI. In a case-crossover analysis of STEMI among unstable angina patients, Evans et al.
(2017) found that the association between previous 1-hour PM2 5 concentration and STEMI became less
precise (i.e., wider confidence intervals) at exposure lags up to 24 or 48 hours and null with an exposure
lag of 72 hours. In a multicity analysis of ED visits for a number of cardiovascular outcomes (i.e., IHD,
CHF, Dysrhythmia), Krall et al. (2018) found that lags longer than their a priori choice (i.e., same-day
exposure [lag 0]) did not produce substantially different results. Studies that examined the lag structure of
associations in relation to stroke reported null associations that were unchanged regardless of the choice
of lag (Fisher etal.. 2019; McClure et al.. 2017). In a study of recurrent IS in Nueces county Texas, Wing
et al. (2017) reported null associations with short-term PM2 5 concentrations on lag Day 1, 2 and 3 and an
inverse association with same day exposures.
3.1.1.3 Recent Studies Examining the PIVh.s-Cardiovascular Effects
Relationship through Accountability Analyses and Causal Modeling
Methods
As discussed in Section 3.1.1.1, the 2019 PM ISA reported that there was sufficient evidence to
conclude that a causal relationship exists between short-term PM2 5 exposure and cardiovascular effects.
However, the body of evidence that supported this causality determination did not include any
epidemiologic studies that conducted accountability analyses or employed causal modeling methods.
Since the literature cutoff date for the 2019 PM ISA several studies that conducted accountability
analyses or implemented causal modeling methods have been published, which further inform the
relationship between short-term PM2 5 exposure and cardiovascular effects, specifically cardiovascular
hospital admissions (Table A-2). The cardiovascular-related hospital admissions examined ranged from
specific cardiovascular endpoints such cardiac arrhythmia, hypertension, ST segment elevation
myocardial infarction to a broader assessment of all cardiovascular diseases.
Zhang et al. (2018) and Wang et al. (2019) both evaluated whether associations between
short-term PM2 5 exposures and cardiovascular hospital admissions differed before, during, or after the
implementation of environmental policies to improve air quality in cities in New York. Zhang et al.
(2018) estimated the rate of cardiovascular hospital admissions associated with short-term PM2 5
concentrations, and whether the rates differed before (2005-2007), during (2008-2013), or after
(2014-2016) the implementation of multiple national and state policies aimed at improving air quality in
multiple cities in New York. Using a time-stratified, case-crossover design, the authors employed
conditional logistic regression models to estimate the rate ratio for cardiovascular hospital admissions
associated with short-term PM2 5 exposure, from the previous day (lag 0) and averaged over the previous
7 days (lag 0-6), adjusting for temperature and relative humidity. The excess rate of hospital admissions
decreased over the entire time period for total cardiovascular disease (before: 1.4% [95% CI: 1.0, 1.8];
during: 1.1% [95% CI: 0.7, 1.5]; after: 1.0% [95% CI: 0.3, 1.17]), with the largest association, in terms of
magnitude, observed in the "before" period and weaker associations observed in the "after" period.
September 2021
3-21
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
Similar results were reported for cerebrovascular disease, ischemic stroke, chronic rheumatic heart
disease, hypertension, ischemic heart disease, and myocardial infarction. Conversely, there were increases
in the excess rate of hospital admissions for cardiac arrhythmia and congestive heart failure in the "after"
period compared to the "before" and "during" periods. Overall, across the endpoints examined, there were
notable differences (i.e., reductions in hospital admissions) after policies were implemented compared to
before.
Wang et al. (2019) also used a time-stratified case-crossover study design to examine whether the
rate of ST segment elevation myocardial infarction (STEMI) is associated with PM2 5 concentrations in
the previous few hours or days, and whether these associations were modified by periods in which there
were changes in environmental policies in Rochester, NY. The authors hypothesized that increases in the
rate of STEMI associated with short-term PM2 5 exposures would be smaller after the changes were
implemented (2014-2016), compared to the periods before (2005-2007) and during (2008-2013)
implementation. Within this study, referent days were selected as the same hour of the event on the same
day earlier and later than the case event within the same month and calendar year. The analyses examined
hourly exposures of 1 hour (lag Hour 1), 3-hour avg (lag Hours 0-2), 12-hour average (lag Hours 0-11),
24-hour average (lag Hours 0-23), 48-hour average (lag Hours 0-47), and 72-hour average (lag
Hours 0-71) prior to the onset of STEMI symptoms. To examine whether the rate of STEMI was
associated with different hourly average concentrations of PM2 5 and were modified by the period of when
changes were implemented, two interaction terms for the "period" (a categorical variable to distinguish
periods of before, during, and after implementation) and PM2 5 concentrations of the time period were
included in the conditional logistic regression model. Over the entire study period, there was a decrease of
approximately 30% in PM2 5 concentrations. Across the three periods, there was a decrease in the rate of
STEMI for the interquartile range (7.59 |ig/m3) increase in PM2 5 concentration in the previous hour (lag
Hour 0) (before: OR= 1.03 [95% CI: 0.91, 1.17]; during: OR= 1.07 [95% CI: 0.92, 1.24]; after:
OR = 0.99[95% CI: 0.81, 1.21]). However, in the previous 72-hour (lag Hours 0-71) period, the rate of
STEMI increased with an IQR increase in PM2 5 concentration across the three periods from 0.91
(95% CI: 0.79, 1.05) before, 0.98 (95% CI: 0.82, 1.18) during, and 1.11 (95% CI: 0.88, 1.41) after
implementation.
The use of causal modeling methods can further inform the causal nature of the relationship
between short-term PM2 5 exposure and cardiovascular effects, specifically in studies examining
cardiovascular hospital admission rates, through the use of statistical methods to reduce uncertainties with
respect to confounding. Inverse probability weighting (IPW) is a causal modeling method that analyzes
observational data in a way that approximates conducting a randomized experiment to make exposure
independent of all potential confounders, rather than to control for the confounders in the outcome
regression (Qiu et al.. 2020). To explore the relationship between short-term PM2 5 exposure and
cardiovascular disease hospital admissions, Qiu et al. (2020) used IPW propensity score methods in a
case-crossover study design to examine an unconstrained distributed lag (lag 0-5) for acute myocardial
infarction (AMI), congestive heart failure (CHF) and ischemic stroke (IS) hospital admissions among
September 2021
3-22
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
New England Medicare participants between 2000 and 2012. In the first step, a linear regression model
was fitted with the exposure lag of interest against the other five lags of PM2 5 exposure and six lags of
ozone along with linear and quadratic terms for temperature (lag 0 and 1) and linear terms for relative
humidity (lag 0 and 1) to control for potential confounding by meteorological conditions. In the second
step, under the assumptions that no important confounders are omitted and correct specification of the
propensity score models, conditional logistic regression models were then used to regress cardiovascular
hospital admissions against each exposure lag, using the weights generated from the propensity score
models to that lag, to estimate the marginal effect of that particular lag of that exposure independent of
covariates.
Based on the IPW method, Oiu et al. (2020) reported an increase of 4.3% (95% CI: 2.2, 6.4) in
AMI hospital admission rate, 3.9% (95% CI: 2.4, 5.5) in CHF rate, and 2.6% (95% CI: 0.4, 4.7) in IS
hospital admission rate for a 10 (.ig/nr1 increase in 24-hour average PM2 5 concentrations. While Oiu et al.
(2020) reported associations in a causal modeling framework that further confirm an association between
short-term PM2 5 exposure and cardiovascular-related hospital admissions, several assumptions were used
by the authors when applying the IPW methods that are important to recognize. First, the authors assumed
there was no unmeasured confounding, with the caveat that the authors had the resources to obtain all the
potential unmeasured confounders. This assumption was tested through a series sensitivity analyses of
temperature by including more lags of temperature and spline adjustments. The second assumption was
positivity, meaning that there are both exposed and non-exposed individuals at every level of the
confounders. The last assumption is consistency, or that the observed outcome is exactly the same as the
potential outcome the individual will have under the exposure assigned.
The recent accountability and causal modeling studies evaluated in this section provide additional
support for a relationship between short-term PM2 5 exposure and cardiovascular effects. These studies
reported consistent associations between cardiovascular hospital admissions with short-term PM2 5
exposures across different statistical methods and study designs, which further supports the conclusions
of the 2019 PM ISA.
3.1.1.4 Summary of Recent Evidence in the Context of the 2019 Integrated
Science Assessment for Particulate Matter Causality Determination
for Short-Term PM2.5 Exposure and Cardiovascular Effects
Recent epidemiologic studies published since the 2019 PM ISA support and extend the evidence
that contributed to the conclusion of a causal relationship between short-term PM2 5 exposure and
cardiovascular effects in the 2019 PM ISA. Multicity analyses of the relationship between short-term
exposure to PM2 5 and cardiovascular ED visits and hospital admissions were an important consideration
in this causality determination. Recent studies support the evidence characterized in the 2019 PM ISA,
extending the evidence relating to hospital admissions and ED visits for specific outcomes (i.e., IHD, MI,
HF, and arrhythmia) with positive association observed across diverse populations (i.e., older adults
September 2021
3-23
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
enrolled in Medicare, Medicaid recipients, and patient populations). With respect to stroke, the evidence
in the 2019 PM ISA was characterized as inconsistent. Recent studies of established cohorts (i.e., WHI,
REGARDS, and HPFU) extend this evidence with observations of null or inverse associations between
short-term PM2 5 exposure and stroke, regardless of stroke subtype. However, an association between
short-term PM2 5 exposure and IS was observed in the Medicaid population.
Uncertainty related to exposure assessment was generally reduced with consideration of studies
included in the 2019 PM ISA that applied hybrid exposure assessment techniques that combined land use
regression with satellite AOD measurements and PM2 5 concentrations measured at fixed site monitors.
The majority of recent studies also rely on exposure assessment strategies that characterize the temporal
and spatial variability of short-term PM2 5 concentrations. Recent studies also performed analyses to
address methodological challenges, including applying techniques to elucidate uncertainties related to the
observation of variable results across single-city studies and controlling for competing mortality risks in
studies of ED visits and hospital admissions.
The evidence in the 2019 PM ISA indicated that the associations between short-term PM2 5
exposure and cardiovascular effects generally persisted in models that were adjusted for copollutants. A
recent study that reports copollutant model results supports the evidence characterized in the 2019 PM
ISA that the effect of short-term PM2 5 exposure on the cardiovascular system is independent of ozone
exposure; however, another recent study introduces the possibility for water soluble metals to confound
the association between short-term PM2 5 exposure and cardiovascular effects. Recent studies continue to
support an immediate effect of short term PM2 5 exposure on the cardiovascular system that was described
in the 2019 PM ISA. Lastly, recent studies that employed causal modeling methods or conducted
accountability analyses when examining short-term PM2 5 exposure and cardiovascular-related hospital
admissions provide additional support for a relationship between short-term PM2 5 exposure and
cardiovascular effects.
3.1.2 Long-Term PM2.5 Exposure
The following sections represent a summary of the evidence and the corresponding causality
determination for long-term PM2 5 exposure and cardiovascular morbidity presented within the 2019 PM
ISA (Section 3.1.2.1) along with an evaluation of recent epidemiologic studies that fall within the scope
of the Supplement (i.e., studies conducted in the U.S. and Canada) and were published since the literature
cutoff date of the 2019 PM ISA (Section 3.1.2.2). In addition, with the expansion of epidemiologic
studies using causal modeling methods, recent studies that employed such methods are also evaluated
(Section 3.1.2.3). which can further inform the relationship between long-term PM2 5 exposure and
cardiovascular morbidity. Lastly, a summary of the results of recent studies evaluated within the section is
presented in the context of the scientific conclusions detailed in the 2019 PM ISA (Section 3.1.2.4V The
September 2021
3-24
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
evaluation of recent studies presented in this Supplement adds to the collective body of evidence
reviewed in the process of reconsidering the PM NAAQS.
3.1.2.1 Summary and Causality Determination from 2019 Integrated
Science Assessment for Particulate Matter
The evidence reviewed in the 2009 PM ISA provided the rationale to conclude that there is a
"causal relationship between long-term PM2 5 exposure and cardiovascular effects" (U.S. EPA. 2009).
Studies of mortality from cardiovascular causes provided the strongest evidence in support of this
conclusion. While several studies included in the 2009 PM ISA reported associations between long-term
PM10 exposure and morbidity outcomes such as post-MI CHF and deep vein thrombosis (DVT), studies
of PM2 5 were limited. One large prospective study of postmenopausal women reported an increased risk
of cardiovascular events, including CHD and stroke, in association with long-term exposure to PM2 5
(Miller etal.. 2007). Cross-sectional analyses provided supporting evidence and experimental studies
demonstrating enhanced atherosclerotic plaque development and inflammation following long-term
exposures to PM2 5 CAPs provided biological plausibility for the epidemiologic findings. In addition,
evidence from the limited number of toxicological studies reporting CAPs-induced effects on
hypertension and vascular reactivity were drawn upon to support the causality determination. With
respect to the 2019 PM ISA, the evidence for the relationship between long-term exposure to PM2 5 and
cardiovascular effects is described below and summarized in Table 3-2. using the framework for causality
determinations described in the Preamble to the IS As (U.S. EPA. 2015).
September 2021
3-25
DRAFT: Do Not Cite or Quote
-------
Table 3-2 Summary of evidence for a causal relationship between long-term
PM2.5 exposure and cardiovascular effects from the 2019 Integrated
Science Assessment for Particulate Matter.
Rationale for
Causality
Determination3
Key Evidence13
Key References and
Sections in the 2019 PM
ISAb
PM2.5 Concentrations
Associated with
Effects0
(H9/m3)
Consistent
epidemiologic
evidence from
multiple studies at
relevant PM2.5
concentrations
Positive associations between long-term
PM2.5 exposure and cardiovascular
mortality in U.S. and Canadian cohorts;
positive associations persisted after
adjustment for common confounders.
Section 6.2.10
Figure 6-19
Mean concentrations
ranged from
4.08 (CCHS)-17.9 CA
Teachers
Positive associations observed in studies Section 6.2.10
examining varying spatial scales and
across different exposure assessment
and statistical methods.
Evidence from
copollutant models
generally supports
an independent
PM2.5 association
Positive associations observed between
long-term PM2.5 exposure and
cardiovascular mortality remain relatively
unchanged after adjustment for
copollutants.
Correlations with ozone were generally
moderate to high (0.49-0.73).
When reported, correlations with SO2,
NO2, and PM10-2.5 ranged from weak to
moderate (r= 0.25-0.55).
Section 6.2.15
Figure 6-21
Figure 6-22
Epidemiologic
evidence supports
a linear
no-threshold
concentration
response (C-R)
relationship
Most analyses support a linear,
no-threshold relationship for
cardiovascular mortality, especially at
lower ambient concentrations of PM2.5.
Confidence in C-R relationship extends to
8 |jg/m3 in Harvard Six Cities study.
Section 6.2.16
Lepeule et al. (2012)
Inconsistent
Association with coronary events, CHD,
Section 6.2.2
Range: 13.4-17.8
evidence from
and stroke (mortality and morbidity
Section 6.2.3
epidemiologic
combined) that persist after adjustment for
studies of CHD or
SES reported in the WHI study.
stroke
Association with stroke but not CHD in the
CA Teachers cohort.
No association with CHD or stroke in the
NHS or HPFU.
September 2021
3-26
DRAFT: Do Not Cite or Quote
-------
Table 3-2 (Continued): Summary of evidence for a causal relationship between
long-term PM2.5 exposure and cardiovascular effects.
Rationale for
Causality
Determination3
Key Evidence13
Key References and
Sections in the 2019 PM
ISAb
PM2.5 Concentrations
Associated with
Effects0
(Hg/m3)
Mean:
15.5
Mean:
14.6
Mean:
23.9
Mean:
13.4
Generally
consistent
evidence of an
association with
CHD or stroke
among those with
preexisting disease
Consistent associations with Ml in patient Hartiala et al. (2016)
populations. Tonne et al. (2015)
Association among women with diabetes Koton et al (2013)
in NHS
Hart et al. (2015b)
Some, but not all, Longitudinal change in CAC observed in Section 6.2.4
epidemiologic
studies provide
evidence for effect
of long-term PM2.5
on CAC
MESA but not in Framingham Heart
Offspring study.
Kaufman et al. (2016)
Dorans et al. (2016)
Mean: 14.2
Median: 9.8
Consistent
evidence from
animal toxicological
studies at relevant
PM2.5
concentrations
Consistent changes in measures of
impaired heart function and blood
pressure.
Additional evidence of atherosclerosis,
systemic inflammation, changes in
endothelial function.
Section 6.2.5.2
Section 6.2.7.2
Section 6.2.4.2
Section 6.2.12.2
Section 6.2.14.2
-85-30
See Tables in identified
sections
Generally Strong evidence for coherence of effects
consistent across scientific disciplines and biological
evidence for plausibility for a range of cardiovascular
biological effects in response to long-term PM2.5
plausibility of exposure. Includes evidence for impaired
cardiovascular heart function, atherosclerosis, and
effects increased blood pressure.
Section 6.2.1
CAC = coronary artery calcification; CCHS = Canadian Community Health Survey; CHD = coronary heart disease; HPFU = Health
Professionals Follow-Up; MESA = Multi-Ethnic Study of Atherosclerosis; |jg/m3 = micrograms per cubic meter; NHS = Nurses'
Health Study; N02 = nitrogen dioxide; PM25 = particulate matter with a nominal mean aerodynamic diameter less than or equal to
2.5 |jm; PM10 = particulate matter with a nominal mean aerodynamic diameter less than or equal to 10 |jm; PM10-2.5 = particulate
matter with a nominal mean aerodynamic diameter greater than 2.5 |jm and less than or equal to 10 |jm; r= correlation coefficient;
SES = socioeconomic status; S02 = sulfur dioxide.
aBased on aspects considered in judgments of causality and weight of evidence in causal framework in Tables I and II of the
Preamble (U.S. EPA. 2015).
bDescribes the key evidence and references contributing most heavily to causality determination and, where applicable, to
uncertainties or inconsistencies. References to earlier sections indicate where the full body of evidence is described in the 2019
PM ISA.
°Describes the PM25 concentrations with which the evidence is substantiated.
September 2021
3-27
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
The studies of long-term exposure to PM2 5 and cardiovascular mortality evaluated in the 2019
PM ISA continue to provide strong evidence of a causal relationship between long-term exposure to
PM2 5 and cardiovascular effects. Results from U.S. and Canadian cohort studies demonstrated consistent,
positive associations between long-term PM2 5 exposure and cardiovascular mortality (2019 PM ISA,
Figure 6-19). Overall, the studies reporting positive associations examined the relationship at varying
spatial scales and employed different exposure assessment and statistical methods (2019 PM ISA,
Section 6.2.10). The studies were conducted in locations where mean annual average concentrations
ranged from 4.08-17.9 (ig/m3. Generally, most of the PM2 5 effect estimates relating long-term PM2 5
exposure and cardiovascular mortality remained relatively unchanged or increased in copollutant models
adjusted for ozone, NO2, PM10-2.5, or SO2. In addition, most of the results from analyses examining the
C-R function for cardiovascular mortality supported a linear, no-threshold relationship for cardiovascular
mortality, especially at lower ambient concentrations of PM2 5.
The body of literature examining the relationship between long-term PM2 5 exposure and
cardiovascular morbidity evaluated in the 2019 PM ISA had greatly expanded since the 2009 PM ISA,
with positive associations reported in several cohorts. The findings from the WHI cohort of
postmenopausal women (Miller et al.. 2007). reporting associations of long-term PM2 5 and coronary
events, were strengthened through a subsequent analysis, which considered potential confounding and
modification by SES and applied enhanced exposure assessment methods (Chi et al.. 2016). However,
analyses of the NHS and CTS, which are both cohorts of women and include extensive data on covariates
(i.e., hormone use, menopausal status, and SES), were not entirely consistent with the WHI findings.
Although the NHS cohort is comparable to WHI in that it is made of predominantly postmenopausal
women, no associations with CHD or stroke were observed in this population (Hart et al.. 2015b). An
association with stroke, but not CHD, that was stronger among postmenopausal women was observed in
the CTS (Lipsett et al.. 2011). Several studies conducted among cardiovascular disease patient
populations generally reported positive associations with MI (Hartiala et al.. 2016; Tonne et al.. 2015;
Koton et al.. 2013). and a sensitivity analysis of the NHS restricted to women with diabetes detected a
positive association with CHD. Although the evidence is not consistent across the populations studied,
heterogeneity is expected when the methods, or the underlying distribution of covariates vary across
studies (Higgins. 2008).
Longitudinal change in measures of atherosclerosis in relation to long-term exposure to PM2 5 add
to the collective evidence base (Hartiala et al.. 2016; Kaufman et al.. 2016; Gan et al.. 2014; Kiinzli et al..
2010). Findings were somewhat variable across cohorts and depended, in part, on the vascular bed in
which atherosclerosis was evaluated. Kaufman et al. (2016) reported an association of PM2 5 with CAC
among middle to older aged adults in the MESA study, while Dorans et al. (2016) reported no association
in the Framingham Heart Study. Associations of long-term exposure to PM2 5 with cIMT were not
consistently observed across cohorts or between analyses of the same cohort with variable methods.
Relationships between PM2 5 and cIMT at younger ages were not observed. However, a recent
September 2021
3-28
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
toxicological study adds to similar evidence from the 2009 PM ISA by demonstrating increased plaque
progression in ApoE" " mice following long-term exposure to PM2 5 collected from multiple locations
across the U.S. (2019 PM ISA, Section 6.2.4.2). Thus, this study provided direct evidence that long-term
exposure to PM2 5 may result in atherosclerotic plaque progression. This study was also coherent with the
epidemiologic studies discussed above reporting positive associations between long-term exposure to
PM2 5 and indicators of atherosclerosis.
A small number of epidemiologic studies also reported positive associations between long-term
PM2 5 exposure and heart failure (2019 PM ISA, Section 6.2.5), blood pressure, and hypertension (2019
PM ISA, Section 6.2.7). These heart failure studies are in agreement with animal toxicological studies
that demonstrated decreased cardiac contractility and function and increased coronary artery wall
thickness following long-term PM2 5 exposure (2019 PM ISA, Section 6.2.5.2). Similarly, a limited
number of animal toxicological studies demonstrated a relationship between long-term exposure to PM2 5
and consistent increases in BP in rats and mice are coherent with epidemiologic studies that reported
positive associations between long-term exposure to PM2 5 and hypertension.
Longitudinal epidemiologic analyses also supported the observation of positive associations with
markers of systemic inflammation (2019 PM ISA, Section 6.2.12), coagulation (2019 PM ISA,
Section 6.2.13), and endothelial dysfunction (2019 PM ISA, Section 6.2.14). These results were in
coherence with animal toxicological studies generally reporting increased markers of systemic
inflammation and oxidative stress (2019 PM ISA, Section 6.2.12.2), as well as with toxicological studies
that generally demonstrated endothelial dysfunction as evidenced by reduced vasodilation in response to
acetylcholine (2019 PM ISA, Section 6.2.14).
There was also consistent evidence from multiple epidemiologic studies that long-term exposure
to PM2 5 was associated with mortality from cardiovascular causes. Associations with CHD, stroke, and
atherosclerosis progression were observed in several additional epidemiologic studies, providing
coherence with the mortality findings. Results from copollutant models generally supported the
independence of the PM2 5 associations. Additional evidence of the direct effect of PM2 5 on the
cardiovascular system was provided by experimental studies in animals, which in part, demonstrate
biologically plausible pathways by which long-term inhalation exposure to PM2 5 could potentially result
in outcomes such as CHD, stroke, CHF, and cardiovascular mortality. Together, these epidemiologic
and experimental studies constitute strong evidence that a causal relationship exists between
long-term exposure to PM2.5 and cardiovascular effects.
3.1.2.2 Recent U.S. and Canadian Epidemiologic Studies
In addition to evaluating epidemiologic studies that examined the relationship between long-term
PM2 5 exposure and cardiovascular effects, the 2019 PM ISA characterized whether evidence supported
biologically plausible mechanisms by which long-term PM2 5 exposure could lead to cardiovascular
September 2021
3-29
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
effects. This evaluation consisted of an assessment of animal toxicological, controlled human exposure,
and epidemiologic studies that examined a range of cardiovascular effects (2019 PM ISA, Section 6.2.1).
Plausible mechanisms were identified by which inhalation exposure to PM2 5 could progress from initial
events to apical events reported in epidemiologic studies (2019 PM ISA, Figure 6-2). The first proposed
pathway begins as respiratory tract inflammation leading to systemic inflammation. The second proposed
pathway involves modulation of the autonomic nervous system. Once these pathways are initiated, there
is evidence from experimental and observational studies that long-term exposure to PM2 5 may result in a
series of pathophysiological responses that could lead to cardiovascular events such as IHD and heart
failure (2019 PM ISA, Figure 6-1).
Recent epidemiologic studies conducted in the U.S. and Canada build upon the strong
epidemiologic evidence base evaluated in the 2019 PM ISA, as well as previous assessments, that
provided the scientific rationale supporting a causal relationship between long-term PM2 5 exposure and
cardiovascular effects (Section 3.1.1.1). In addition to examining the relationship between long-term
PM2 5 exposure and specific cardiovascular outcomes (i.e., IHD and myocardial infarction
[Section 3.1.2.2.11. cerebrovascular disease and stroke [Section 3.1.2.2.21. atherosclerosis
[Section 3.1.2.2.31. heart failure and impaired heart function rSection 3.1.2.2.41. cardiac
electrophysiology and arrhythmia rSection 3.1.2.2.51. blood pressure and hypertension
rSection 3.1.2.2.61. and cardiovascular mortality rSection 3.1.2.2.71). analyses within these recent studies
also further examined issues relevant to expanding the overall understanding the effect of long-term PM2 5
exposure on cardiovascular outcomes. Specifically, recent studies assessed potential copollutant
confounding (Section 3.1.2.2.8) and the shape of the concentration-response (C-R) relationship
(Section 3.1.2.2.9V The following sections evaluate recent epidemiologic studies conducted in the U.S.
and Canada that inform each of the aforementioned topics within the context of the evidence base
evaluated and summarized in the 2019 PM ISA.
3.1.2.2.1 Ischemic Heart Disease and Myocardial Infarction
The terms ischemic heart disease (IHD), coronary artery disease (CAD), and coronary heart
disease (CHD) are generally interchangeable as they appear in the epidemiologic literature on the effects
of air pollution. The majority of IHD is caused by atherosclerosis, which can result in the blockage of the
coronary arteries and restriction of blood flow to the heart muscle. A myocardial infarction (MI) or heart
attack is an acute event that results in heart muscle tissue death secondary to coronary artery occlusion.
The epidemiologic studies included in the 2019 PM ISA represented a substantial expansion of the
literature compared to the few studies available for review in the 2009 PM ISA. Overall, findings from
these studies were not entirely consistent. The strongest evidence of an association with IHD was found in
populations with preexisting diseases such as diabetes or cardiac patients that are followed after an acute
event or procedure. Recent studies examine the association between long-term PM2 5 exposure and MI
with most reporting results that are consistent with those studies evaluated in the 2019 PM ISA.
September 2021
3-30
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
Recent analyses of the Canadian Ontario Population Health and Environment Cohort (ONPHEC)
also add to the available evidence on the relationship between long-term PM2 5 exposure and
cardiovascular effects. ONPHEC includes more than 5 million Canadian-born adults (35-85 years old at
enrollment in 1996) who were registered with the provincial health service and had resided in Ontario for
>5 years. In a prospective analysis, Bai et al. (2019) estimated the association between 3-year average
PM2 5 concentrations and incident cases of acute MI. The study reported a positive association (HR: 1.07
[95% CI: 1.06, 1.09]). In addition, stratified analyses showed patterns of associations that indicated
stronger effect estimates in the youngest (35-44 years) and oldest (75-85 years) age groups. Bai et al.
(2019) also examined effect modification by oxidant gases, which was estimated as the redox weighted
average of NO2 and O3 (Ox). A stronger association was observed in the highest tertile of (>38.97 ppm)
Ox concentrations (HR: 1.12 [95% CI: 1.09, 1.15]).
Chen et al. (2020) also analyzed data from the ONPHEC study but examined the association of
annual average PM2 5 in the previous year with the incidence of acute MI. The authors conducted single
pollutant analyses using both a traditional Cox proportional hazards model where PM2 5 is fit as a linear
term and a Cox proportional hazards model where PM2 5 was fit as a non-linear term. Model fit, which
was assessed based on the Akaike information criterion (AIC) value, did not vary across models. The risk
estimates were also virtually the same across models (i.e., HR =1.14 [95% CI: 1.12, 1.16] in both
models). In addition to conducting single-pollutant analyses, the authors introduced a new approach to
assess whether the association of PM2 5 with acute MI varied depending on the proportion of PM2 5
attributed to selected components (i.e., sulphate, nitrate, ammonium, black carbon, organic matter,
mineral dust, and sea salt). The study found that the model that adjusted for the proportion of each of the
seven selected components was a better predictor of acute MI based on lower AIC values. In addition,
Chen et al. (2020) reported that acute MI associations increased by an average of 10% when compared to
single-pollutant results across each of the five regions of Ontario when using the component proportion
adjusted approach. Overall, the component adjusted model provided some support that variability in the
proportion of individual components that comprise PM2 5, could explain regional variability in risk
estimates.
While the studies above focus on examining the relationship between long-term PM2 5 exposure
and MI in cohorts of diverse populations, some recent studies have analyzed data from a cohort of women
and cardiac catheterization patients. Elliott et al. (2020) examined the interaction between 24-month PM2 5
concentration and physical activity in association with MI among women enrolled in the Nurses' Health
Study (NHS). Unlike an earlier analysis of this cohort that examined IHD (Hart et al.. 2015b). the authors
found a positive association of PM2 5 with MI (HR: 1.06 [95% CI: 1.00, 1.12]), although no statistical
evidence of an interaction with physical activity was observed. In the previous analysis of the NHS
cohort, Hart et al. (2015b) reported no association between long-term PM2 5 exposure and incident CHD
(HR: 1.01 [95% CI: 0.96, 1.07]), although a positive association with IHD was observed among women
with diabetes (HR: 1.10 [95% CI: 0.99, 1.21]). Weaver et al. (2019) studied cardiac catheterization
patients residing in three counties in NC to determine the association of annual average PM2 5
September 2021
3-31
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
concentration with MI, CAD, and hypertension. Among the objectives of this study was to understand the
effect of sociodemographic characteristics on associations by assigning study participants to clusters
based on the census block group of their residence that indicated specific sets of sociodemographic
characteristics. Positive associations of annual average PM2 5 concentration with both CAD and MI were
observed. The association with MI was observed across all sociodemographic clusters (OR for all
clusters: 3.57 [95% CI: 2.10, 5.77]). The association with CAD was also observed across all clusters (OR:
1.40 [95% CI: 0.90, 2.19]) but was largely driven by one cluster (OR: 2.01 [95% CI: 1.00, 3.86]), which
was urban and characterized by low poverty, low unemployment, and composed of relatively highly
educated residents with managerial jobs.
In another study, Loop et al. (2018) conducted an analysis of the REasons for Geographic and
Racial Differences in Stroke (REGARDS) cohort, a nationwide study which oversampled participants
from states in the southern U.S. where there is known to be an increased risk of stroke. Participants who
were free from CHD at baseline were followed for an average of 6 years. Loop et al. (2018) reported an
inverse association between annual average PM2 5 concentration at baseline and non-fatal MI (HR: 0.74
[95% CI: 0.56, 0.98]). Loop et al. (2018) also examined associations for total CHD, i.e., CHD deaths and
non-fatal MI cases combined, and reported no evidence of an association (HR: 0.89 [95% CI: 0.71, 1.11]).
The evidence informing the relationship between long-term exposure to PM2 5 and IHD, including
the recent studies of ML, is summarized in Figure 3-5. Recent studies do not all report positive
associations; however, the strongest evidence of a relationship continues to be for those with preexisting
disease or patient populations that are followed after a cardiac event or procedure such as catheterization.
September 2021
3-32
DRAFT: Do Not Cite or Quote
-------
Study
Miller et al. 2007
tHart et al. 2015
fLipsettet al. 2011
fPuett et al. 2011
Cohort
WHI-Women (post-menopause),
36 Urban sites, U.S.
NHS-Women, 48 States, U.S.
CTS -Women, Los Angeles.
California, U.S.
HPFU, Men, 13 States, U.S.
Outcome
CHD
CHD
Ml
Nonfatal Ml
Years
1994-1998
1989-2006
1999-2005
1988-2002
tMadrigano et al. 2013
Worcester Heart Attack, MA, U.S.
Confirmed Ml
1995-2003
fHartiala etal. 2016
Cardiac Patients, Ohio, U.S.
Ml
1998-2010
tHoffman et al. 2015
HNR study, Ruhr region Germany
Coronary Event
2008-2009
fAtkinson et al. 2013
GP Database, U.K.
Ml
2003-2007
fCesaroni et al. 2014
ESCAPE- 11 Cohorts Europe
IHD
2008-2011
*Yazdi etal. 2019
Medicare southearstern U.S.
Ml
2000-2012
*Loop et al. 2018
REGARDS U.S.
Ml
2003-2012
*Bai et al. 2019
ONPHEC Ontario Canada
Ml
1998-2012
•Elliot et al. 2020
NHS-Women, 48 States, U.S.
Ml
1988-2007
*Weaveretal. 2019
CATHGEN, RTP NC
Ml
2000-2008
tTonne et al. 2015
MINAP, London, U.K.
Recurrent Ml/death
2003-2010
|Kotonetal. 2013
8 Treatment Centers, Israel
Recurrent Ml
2003-2005
Mean (|jg/m3)
13.4
13.4
15.6
17.8
15.5
18.4
12.9
7.3-31
NR
13.6
13.7
12.7
14.6
23.9
6
i
lO
I
!•
-i—•-
-+-
0.5 1 1.5 2
Relative Risk (95% CI)
2.5
3.5
Source: Update of Figure 6-17, 2019 PM ISA.
Note: tStudies published since the 2009 PM ISA. *U.S. and Canadian studies published since the literature cutoff date (~January
2018) for the 2019 PM ISA. Circles represent point estimates; horizontal lines represent 95% confidence intervals for PM2 5. Black
text and circles represent evidence included in the 2009 PM ISA; red text and circles represent evidence considered in 2019 PM
ISAs; and blue text and circles represent recent studies published since the 2019 ISA. Mean concentrations in |jg/m3. Hazard Ratios
are standardized to a 5 |jg/m3 increase in PM25 concentrations. CATHGEN = Catheterization Genetics Study; CHD = Coronary
Heart Disease; CTS = California Teachers Study; ESCAPE = European Study of Cohorts for Air Pollution; HNR = Heinz Nixdorf
Recall study; HPFU = Health Professionals Follow-up Study; IHD = Ischemic Heart Disease; Ml = Myocardial Infarction;
MINAP = Myocardial Ischemia National Audit Project; NHS = Nurses' Health Study; ONPHEC = Ontario Population Health and
Environmental Cohort; REGARDS = REasons for Geographic and Racial Differences in Stroke; WHI = Women's Health Initiative.
Figure 3-5 Associations between long-term PM2.5 exposure and ischemic
heart disease or myocardial infarction.
3.1.2.2.2 Cerebrovascular Disease and Stroke
1 Cerebrovascular disease typically includes the conditions hemorrhagic stroke, cerebral infarction
2 (i.e., ischemic stroke), and occlusion of the precerebral and cerebral arteries. The 2009 PM ISA identified
3 one study that indicated a positive association between PM2.5 and cerebrovascular morbidity and mortality
4 in post-menopausal women (Miller et al.. 2007). Although the results were not entirely consistent across
5 studies or stroke subtype, some studies reviewed in the 2019 PM ISA provided evidence to support a
6 positive association between long-term exposure to PM2.5 and stroke. Several recent studies that observe
7 positive associations add to this evidence base (Figure 3-6).
September 2021
3-33
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
Several studies examined the association between long-term PM2 5 concentration and stroke. In a
study of women enrolled in the NHS cohort, Elliott et al. (2020) reported an imprecise (i.e., wide
confidence intervals relative to the size of the HR) association between 24 month average PM2 5
concentration and stroke that overlapped the null value (HR: 1.02 [95% CI: 0.96, 1.09]). An earlier
analysis examining the association with annual average PM2 5 concentration in the NHS, reported an
increased risk among women with diabetes (HR: 1.29 [95% CI: 1.14, 1.45]) but not in the population,
overall (HR: 1.01 [95% CI: 0.96, 1.05]) (Hart et al.. 2015b). Rhinehart et al. (2020) estimated the
association of annual average PM2 5 concentration within 300 meters of the residence with stroke in a
prospective analysis of residents of Allegheny county Pennsylvania who were diagnosed with atrial
fibrillation but had no history of stroke. This study reported a positive association (HR: 1.62 [95% CI:
1.00, 2.55]). As opposed to examining annual or 24-month average PM2 5 exposures, Shin et al. (2019)
estimated the association between 5-year PM2 5 concentration and incident cases of stroke in a prospective
analysis of the Canadian ONPHEC study and reported a positive association (HR: 1.05 [95% CI: 1.03,
1.07]).
Studies that examined the relationship of long-term PM2 5 exposure with CBVD and stroke are
summarized in Figure 3-6. Recent studies support the evidence in the 2019 PM ISA and extend the
evidence relating to the observation of associations among patients that are followed after a cardiac event
or procedure.
September 2021
3-34
DRAFT: Do Not Cite or Quote
-------
Study
Miller et al. 2007
tHart et al. 2015
WHI-Women (post-menopause) 36 c. .
Urban Sites, US btroKe
NHS- Women, 48 States, U.S.
*Elliot et al. 2020 NHS- Women, 48 States, U.S.
tLipsettet al. 2011
tPuettet al. 2011
tPuettetal. 2011
*Yazdi et al. 2019
*Shin et al. 2019
CTS -Women, Los Angeles,
California, U.S.
HPFU, Men, 13 States, U.S.
HPFU, Men, 13 States, U.S.
Medicare U.S.
ONPHEC, Ontario Canada
1 st Admission Stroke
1 st Admission Stroke
*Rhinehart et al. 2020 Atrial Fibrillation Patients, U.S.
tHartiala et al. 2016
tStafoggiaet al. 2014
tHoffmann et al. 2015
tAtkinson etal. 2013
tKotonetal. 2013
IS
Stroke
Cardiac Patients, Ohio, U.S.
ESCAPE-11 Cohorts Europe Stroke
HNR study, Ruhr area, Germany Stroke
GP Database, U.K.
5 Centers, Israel
Stroke
Post Ml Stroke
Years
1994-1998
1989-2006
1988-2007
1999-2005
1988-2002
1988-2002
2000-2012
1998-2012
2007-20017
1998-2010
2008-2011
2008-2009
2003-2007
2003-2005
Mean(|jg/m3)
13.4
13.4
I
I
o
17.8
17.8
NR
9,8
7.3-31
18.4
12.9
23.9
0.5 1
3 4 5
Relative Risk (95% CI)
Source: Update of Figure 6-18, 2019 PM ISA.
Note: jStudies published since the 2009 PM ISA. *U.S. and Canadian studies published since the literature cutoff date (^January
2018) for the 2019 PM ISA. Circles represent point estimates; horizontal lines represent 95% confidence intervals for PM2.5- Black
text and circles represent evidence included in the 2009 PM ISA; red text and circles represent evidence included in the 2019 PM
ISA; blue text and circles represent evidence not included in the 2019 PM ISA. Mean concentrations in |jg/m3. Hazard Ratios are
standardized to a 5-pg/m3 increase in PM2.5 concentrations. (U.S. EPA, 2018). ESCAPE = European Study of Cohorts for Air
Pollution; GP = General Practitioner; HNR = Heinz Nixdorf Recall; HPFU = Health Professional's Follow-up; HS = hemorrhagic
stroke; IS = ischemic stroke; Ml = Myocardial Infarction; NHS = Nurses' Health Study; ONPHEC = Ontario Population Health and
Environment Cohort; WHI = Women's Health Initiative.
Figure 3-6 Associations between long-term PM2.5 exposure and the
incidence of stroke.
3.1.2.2.3 Atherosclerosis
1 Atherosclerosis is the process of plaque buildup that forms lesions on the walls of the coronary
2 arteries, which can lead to narrowing of the vessel, reduced blood flow to the heart and IHD.
3 Atherosclerosis can be assessed within large arterial vascular beds in distinct regions of the body
4 i.e., carotid intima-media thickness (cIMT), coronary artery calcification (CAC), ankle-brachial index
5 (ABI), and the presence of plaques. Findings from studies reviewed in the 2009 PM ISA were
6 inconsistent, reporting null or positive, but imprecise associations with cIMT, CAC, and ABI. Similarly,
7 findings from studies reviewed in the 2019 PM ISA were not entirely consistent across populations,
September 2021
3-35
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
exposure assessment methods, and measures of atherosclerosis. Notably, an extended MESA analysis
reported a longitudinal increase in CAC (4.1 Agatston unit increase per year [95% CI: 1.4, 6.8]) in
association with annual average PM2 5 exposure, but no association with cIMT (Kaufman et al.. 2016).
Exposure measurement error, variation in baseline measures of atherosclerosis as well as statistical power
were noted as possible explanations for the lack of association observed in these studies. Consideration of
copollutant confounding was generally limited across the evidence base reviewed in the 2019 PM ISA.
Several recent studies expand the evidence available to consider the association of long-term
PM2 5 exposure with atherosclerosis. Following the analysis by Kaufman et al. (2016). Keller etal. (2018)
estimated the association of PM2 5 concentration (i.e., multi-year average during the study period,
2000-2012) with CAC progression among participants in MESA air residing in Baltimore, MD. The
authors also assessed whether this association was modified by membership in clusters with different
traffic related air pollution (TRAP) component profiles. The authors reported a 23.0 Agatston unit per
year increase (95% CI: 14.2, 31.7) among participants overall. Keller etal. (2018) also reported a larger
magnitude association with CAC progression (42.6 Agatston unit per year increase [95% CI: 25.7, 59.4])
in the cold season among those belonging to a cluster that was characterized as downtown with above
average ratios of ultrafine and accumulation mode particles relative to NOx.
Among women enrolled in the Study of Women's Health Across the Nation (SWAN), a cohort of
U.S. women transitioning through menopause, Duan et al. (2019a) estimated the association of 5-year
average PM2 5 concentration with cIMT. The study reported a 27.95 |im (95% CI: -2.90, 58.75) thicker
mean cIMT in association with 5-year mean PM2 5 concentration in adjusted models. PM2 5 was also
associated with an increase in increased mean inter-adventitial diameter (IAD), which is a marker of
vascular remodeling and aging as well as a predictor of cardiovascular events, of 105.90 (95% CI:
-63.00, 274.80). No association was reported with plaque presence (OR: 0.90 [95% CI: 0.50, 1.61]) or
plaque severity index (plaque index 0-2, OR: 1.05 [95% CI: 0.53, 8.95] and plaque index >2, OR: 0.62
[95% CI: 0.25, 1.47]) in the SWAN study. In an analysis of a subset of SWAN participants (Pittsburgh
and Chicago only), Duan et al. (2019b) estimated the association of the same measures of atherosclerosis
as Duan et al. (2019a) with annual average PM2 5 concentration reporting a 11.25 |im per year increase
(95% CI: -3.05, 25.60) in mean cIMT. The authors also reported associations with plaque presence (OR:
2.10 [95% CI: 0.66, 6.63]) and plaque index progression (OR: 2.70 [95% CI: 0.77, 9.24]).
Recent studies support and extend the evidence characterized in the 2019 PM ISA with
observations of associations with cIMT among women transitioning into menopause and potential effect
modification by TRAP in the MESA study.
3.1.2.2.4 Heart Failure and Impaired Heart Function
Heart failure (HF) refers to a set of conditions including congestive heart failure (CHF) in which
the heart's pumping action is weakened. With CHF the blood flow from the heart slows, failing to meet
September 2021
3-36
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
the oxygen demands of the body, and returning blood can back up, causing swelling or edema in the lungs
or other tissues (typically in the legs and ankles). Risk factors for HF include IHD, high blood pressure,
atrial fibrillation, and diabetes. The small number of epidemiologic studies reviewed in the 2019 PM ISA
provided evidence supporting a possible relationship between heart failure and long-term exposure to
PM2 5. In addition, an association with increased right ventricular (RV) mass was observed among MESA
participants. Right sided HF is typically a consequence of left-sided HF but can also result from damage
to the pulmonary vasculature, which can result in RV mass, reduced flow to the left ventricle and reduced
left ventricular (LV) mass. Recent studies provide consistent evidence of an association between
long-term PM2 5 and HF.
A recent study examining the association between long-term PM2 5 exposure and HF was
conducted among participants in the Canadian ONPHEC study. In this prospective analysis, Bai et al.
(2019) examined the relationship between 3-year moving average PM2 5 concentration with new cases of
CHF. A positive association was reported, overall (HR: 1.07 [95% CI: 1.06, 1.07]), and a larger
magnitude association was reported in the highest tertile of Ox concentrations in a stratified analysis
examining potential effect modification (HR: 1.12 [95% CI: 1.10, 1.13]).
3.1.2.2.5 Cardiac Electrophysiology and Arrhythmia
Electrical activity in the heart is typically measured using surface electrocardiography (ECG).
ECGs measure electrical activity in the heart due to depolarization and repolarization of the atria and
ventricles. Atrial fibrillation (AF) is the most common type of arrhythmia. Despite being common,
clinical and subclinical forms of AF are associated with reduced functional status, quality of life and is
associated with downstream consequences such as ischemic stroke (Prvstowskv et al.. 1996; Laupacis et
al.. 1994) and CHF (Roy et al.. 2009). contributing to both cardiovascular disease (CVD) and all-cause
mortality (Kannel et al.. 1983). Ventricular fibrillation is a well-known cause of sudden cardiac death and
commonly associated with MI, HF, cardiomyopathy, and other forms of structural (e.g., valvular) heart
disease.
In an analysis of the WHI, which was reviewed in the 2009 PM ISA, Liao et al. (2009) found no
association of long-term PM2 5 concentrations with supraventricular or ventricular ectopy, which are the
most frequent forms of arrhythmia in the general population. A limited number of studies reviewed in the
2019 PM ISA found associations of long-term PM2 5 exposure with premature atrial contractions and
ventricular conduction abnormalities, but not arrhythmias recorded on implantable cardioverter
defibrillators. In a recent prospective analysis of the Canadian OPHEC study, Shin et al. (2019) estimated
the association between 5-year average PM2 5 concentration and incident cases of atrial fibrillation (AF)
and reported a positive association (HR: 1.03 [95% CI: 1.01, 1.04]).
September 2021
3-37
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
3.1.2.2.6
Blood Pressure and Hypertension
High blood pressure is typically defined as a systolic blood pressure above 140 mm hg or a
diastolic blood pressure above 90 mm Hg (U.S. EPA. 2019). Hypertension, the clinically relevant
consequence of chronically high blood pressure, typically develops over years. The body of literature
reviewed in the 2019 PM ISA was substantially larger than the 2009 PM ISA with longitudinal analyses
generally showing small magnitude increases in SBP, PP, and MAP in association with long-term
exposure to PM2 5. In addition, the expanded body of literature provided evidence of associations between
long-term PM2 5 exposure and hypertension. Recent studies add to the evidence providing support for
positive associations among post-menopausal women enrolled in the WHI study and in cardiac
catheterization patients, but not among Black women enrolled in the Jackson Heart Study (JHS).
Honda et al. (2017) estimated the association of PM2 5 concentration with incident hypertension
among post-menopausal women enrolled in the WHI. Annual average PM2 5 concentration was associated
with incident hypertension. (1.17 [95% CI 1.10, 1.22]). The association with PM2 5 concentration
increased among non-White participants and those who lived in the Northeast U.S. By contrast, no
association of 1-year or 3-year average PM2 5 concentration with hypertension was observed in a recent
prospective analysis conducted by Weaver et al. (2021) of African American women enrolled in the
Jackson Heart Study (JHS) (1-year, RR: 1.00 [95% CI: 0.52, 2.03] and 3-year, RR: 1.10 [95% CI: 0.39,
2.84]). Further adjustment for diabetes did not change these findings. A cross-sectional analysis of PM2 5
concentration and prevalent hypertension conducted by Weaver et al. (2021) yielded similar results.
An association between long-term PM2 5 exposure and hypertension was also observed among
cardiac catheterization patients in three counties in North Carolina (Weaver et al.. 2019). In this study
Weaver etal. (2019) estimated the association of annual average PM2.5 concentration with hypertension.
No association between long-term PM2 5 exposure and hypertension was observed in the study population
overall (OR: 0.90 [95% CI: 0.59, 1.34). The pattern of associations between long-term PM2 5
concentration and hypertension indicated larger magnitude associations among study participants who
lived in two sociodemographic clusters, the first characterized as urban, having a high proportion of Black
individuals and individuals in non-managerial occupations (denoted as Cluster 1) and the second
characterized as urban, impoverished, having a high proportion of individuals who are unemployed, work
in non-managerial occupations, are Black, and live in single parent homes (Cluster 2). The OR for Cluster
1 was 2.70 (95% CI: 0.95, 7.59) and the OR for Cluster 2 was 11.86 (95% CI: 2.10, 67.21).
Multiple epidemiologic studies (Section 3.2.2.2.2) reviewed in the 2009 PM ISA and in the 2019
PM ISA found consistently positive associations between long-term PM2 5 exposure and cardiovascular
mortality. Generally, these studies had extensive control for a wide range of potential confounders and the
observed effect estimates remained relatively unchanged or increased in copollutant models adjusted for
3.1.2.2.7
Cardiovascular Mortality
September 2021
3-38
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
ozone, NO2, PM10-2.5, or SO2. Recent cohort studies, which are reviewed in detail in Section 3.2.2.2.2
provide additional evidence for associations with cardiovascular mortality outcomes across the
distribution of PM2 5 concentrations (Haves et al.. 2020). the potential implications of a comorbidity on
the PM2 5-cardiovascular mortality relationship (Pinault et al.. 2018). and associations with individual
cardiovascular mortality outcomes including IHD (Crouse et al.. 2020; Wang et al.. 2020; Cakmak et al..
2018; Pinault et al.. 2017) and stroke (Crouse et al.. 2020; Haves et al.. 2020; Wang et al.. 2020; Pope et
al.. 2019; Pinault et al.. 2017). Overall, these recent studies support the conclusions in the 2019 PM ISA
of consistent positive associations of long-term PM2 5 exposure with cardiovascular mortality, and
specifically with IHD- and stroke-related mortality. Although Pope et al. (2014) reported positive
associations of long-term PM2 5 exposure with congestive heart failure (CHF) mortality in a study of the
ACS cohort evaluated in the 2019 PM ISA, a recent analysis of the Medicare cohort Wang et al. (2020)
reported a null association. Recent studies also indicate that the combination of cardiovascular disease
and diabetes together has a greater mortality risk than cardiovascular mortality alone and that
cardiovascular diseases such as heart failure or previous MI may increase the risk of PM2 5-related
all-cause mortality (Ward-Caviness et al.. 2020; Malik et al.. 2019).
3.1.2.2.8 Copollutant Confounding
The independence of the association between long-term exposure to PM2 5 and cardiovascular
health effects can be examined through the use of copollutant models. A change in the PM2 5 risk
estimates, after adjustment for copollutants, may indicate the potential for confounding. A limited number
of studies were available in the 2019 PM ISA to assess copollutant confounding of the association
between long-term exposure to PM2 5 and cardiovascular morbidity. Considering these few available
studies, risk estimates remained largely unchanged after adjustment for PM10-2.5, NO2, and PM2 5 from
traffic sources. The limited number of recent analyses report some attenuation of risk estimates in models
adjusted for O3 and NO2.
Several recent analyses of the ONPHEC study add to the evidence pertaining to copollutant
confounding (Figure 3-7). Shin et al. (2019) reported that the association of long-term PM2 5 exposure
with stroke persisted after adjustment for NO2 but was attenuated in the models with O3 and oxidant gases
(Ox) represented by the redox weighted average ofN02 and O3 (HR: 1.05 [95% CI: 1.03, 1.06] and HR:
1.03 [95% CI: 1.02, 1.04], and HR: 1.02 [95% CI: 1.00, 1.05], respectively). In an analysis of AF, these
authors found that the association was slightly attenuated, but remained positive, in two-pollutant models
that adjusted for NO2, O3 and redox weighted average of NO2 and O3 (Ox) (HR: 1.03 [95% CI: 1.02, 1.04]
and HR: 1.02 [95% CI: 1.01, 1.03], and HR: 1.02 [95% CI: 1.01, 1.03], respectively). In addition, a study
of atherosclerosis in the SWAN cohort, Duan et al. (2019a) reported that the estimate for the association
of long-term PM2 5 exposure with cIMT was slightly attenuated in a two-pollutant model that adjusted for
O3 (24.45 (im [95% CI: -18.35, 67.25]). In a separate analysis of a subset of this cohort, however, Duan
et al. (2019b) reported that associations with plaque presence and plaque index progression persisted in
September 2021
3-39
DRAFT: Do Not Cite or Quote
-------
1 models adjusted for ozone (2.29 [95% CI: 0.70, 7.59]) for plaque presence and (OR 3.05 [95% CI: 0.86,
2 10.82] for plaque index progression). Overall, the limited evidence indicates that associations between
3 PM2 5 and cardiovascular health effects persist, but may be slightly attenuated, in models that are adjusted
4 for copollutants.
Study Cohort
tPuett et al. 2011 HPFU Study
Outcome
Ml
tMadrigano et al. 2013 Worcester Heart Attack Ml
tPuett et al. 2011 HPFU Study Ischemic Stroke
Hemmoragic Stroke
*Shin et al. 2019 ONPHEC Stroke
•Shin etal. 2019 ONPHEC
fFuks et al. 2014 ESCAPE
*Duanetal. 2019 SWAN
Atrial Fibrillation
Hypertension
BPLM
Plaque Presence
Plaque Index
Copolllutant (r)
Single
PM10-2.5 (NR)
Single
PM2.5 local (NR)
Single
PM10-2.5 (NR)
Single
PM10-2.5 (NR)
Single
N02 (0.65)
03 (0.28)
Ox (0.67)
Single
N02 (0.65)
03 (0.28)
Ox (0.67)
Single
N02 (0.19-0.88)
Single
N02 (0.19-0.88)
Single
03 (NR)
Single
03 (NR)
•+0—i
!•
0
1
0
v
1
©
i
.
!o
y
o
0.5
1.5
2.5
3.5
Relative Risk (95% CI)
Source: Update of Figure 6-20, 2019 PM ISA.
Note: tStudies published since the 2009 PM ISA. *U.S. and Canadian studies published since the literature cutoff date (~January
2018) for 2019 PM ISA. Circles represent point estimates; horizontal lines represent 95% confidence intervals for PM25. Solid circles
represent single pollutant results and open circles represent copollutant results. Hazard Ratios are standardized to a 5 |jg/m3
increase in PM25 concentrations.
Figure 3-7 Associations between long-term exposure to PM2.5 and
cardiovascular morbidity in single pollutant models and models
adjusted for copollutants.
September 2021
3-40
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
3.1.2.2.9
Examination of the Concentration-Response (C-R) Relationship
between Long-Term PM2.5 Exposure and Cardiovascular Effects
An important consideration in characterizing the association between long-term PM2 5 exposure
and cardiovascular effects is whether the C-R relationship is linear across the full concentration range that
is encountered, or if there are concentration ranges where there are departures from linearity. A limited
number of studies evaluated in the 2019 PM ISA examined the shape of the C-R relationships for
cardiovascular morbidity outcomes with the majority of studies lacking thorough evaluations of
alternatives to linearity (2019 PM ISA, Table 6-51). Several recent studies expand the evidence pertaining
to the shape of the C-R relationship for the incidence of MI, AF, stroke, and CHF. A number of these
studies use statistical techniques that allow for departures from linearity (Table 3-3) generally supporting
and extending the evidence characterized in the 2019 PM ISA showing linear, no-threshold C-R
relationship for most CVD outcomes. However, there is some evidence for a sublinear or supralinear C-R
relationship for some outcomes.
Table 3-3. Summary of studies examining the concentration-response (C-R)
relationship or conducted threshold analyses for long-term exposure
to PM2.5 and cardiovascular morbidity.
Study
Exposure
PM2.5 Mean:
Outcome (Range) in |jg/m3
Location—Cohort
(Table/Figure from Reference)
Statistical Analysis
Summary
Baietal. (2019)
Figure 3-8
Ontario, Canada
ONPHEC
Acute Ml
incidence
Mean (IQR):9.6
(3.5)
using SCHIF** (Nasari etal.. 2016).
A linear concentration response
Identified the shape of the C-R
function for fully adjusted Cox models
relationship between acute Ml and
PM2 5 concentration was observed.
Chen et al. (2020)
Figure 3-9
Ontario, Canada
ONPHEC
Acute Ml
Incidence
Mean: 8.61
Identified the shape of the C-R
function for fully adjusted Cox models
using SCHIF** (Nasari etal.. 2016).
Restricted cubic splines with 4 df to
assess linearity used in sensitivity
analysis.
Approximately linear relationship
observed with both methods.
Danesh Yazdi et al. (2019)
Figure 3-10
Medicare
Southeastern, U.S.
1st hospital NR
admission for
Ml
Penalized spline to estimate the
shape of the C-R relationship, with
degrees of freedom chosen based on
corrected AIC values.
September 2021
3-41
DRAFT: Do Not Cite or Quote
-------
Table 3-3 (Continued): Summary of studies examining the
concentration-response (C-R) relationship or conducted
threshold analyses for long-term exposure to PM2.5 and
cardiovascular morbidity.
Study
Location—Cohort
(Table/Figure from Reference)
Exposure
PM2.5 Mean:
Outcome (Range) in |jg/m3
Statistical Analysis
Summary
C-R relationship continued down to
low exposure levels and persisted
when the dataset was restricted
<12 |jg/m3. The relationship was
generally linear at concentrations
below 14 |jg/m3.
Loop etal. (2018)
Figure 3-11
U.S. Nationwide
REGARDS
Non-fatal Ml
incidence
Median (IQR): 13.6
(2.7)
Predicted log hazard modelled as a
linear function (non-linear relationship
tested using restricted cubic splines).
Sensitivity analyses to test for
interactions of PM2.5 with gender,
race, and urbanicity were conducted
to elucidate discrepant findings
(i.e., inverse relationship). No
statistically significant interactions
observed (p = 0.05 level).
Inverse relationship between annual
average PM2.5 exposure and nonfatal
Ml.
Shin etal. (2019)
ONPHEC
Ontario, Canada
Atrial Mean (IQR): 9.8 Identified the shape of the C-R
Fibrillation (4.0) function for fully adjusted Cox models
using SCHIF (Nasari et al.. 2016).
Sublinear relationship observed with
some evidence of potential threshold
at PM2.5 concentrations <6 |jg/m3.
Shin etal. (2019)
ONPHEC
Ontario, Canada
Stroke Mean (IQR): 9.8 Identified the shape of the C-R
(4.0) function for fully adjusted Cox models
using SCHIF (Nasari et al.. 2016).
Linear association was observed with
no evidence of a threshold.
Danesh Yazdi et al.
Medicare
Southeastern, U.S.
(2019)
1st hospital
admission for
stroke
NR
Penalized spline to estimate the
shape of the C-R relationship, with
degrees of freedom chosen based on
corrected AIC values.
C-R relationship continued down to
low exposure levels and persisted
when the data set was restricted
<12 |jg/m3. The relationship was
generally linear at concentrations
below 14 |jg/m3.
September 2021
3-42
DRAFT: Do Not Cite or Quote
-------
Table 3-3 (Continued): Summary of studies examining the
concentration-response (C-R) relationship or conducted
threshold analyses for long-term exposure to PM2.5 and
cardiovascular morbidity.
Study
Location—Cohort
(Table/Figure from Reference)
Bai etal. (2019)
Ontario, Canada
ONPHEC
Exposure
PM2.5 Mean:
Outcome (Range) in |jg/m3
CHF Mean (IQR):9.6
(3.5)
Statistical Analysis
Summary
Identified the shape of the C-R
function for fully adjusted Cox models
using SCHIF (Nasari et al.. 2016).
A supralinear concentration response
relationship between CHF and PM2.5
concentration was observed.
AIC = Akaike information criterion; C-R = Concentration-Response; CAC = coronary artery calcification; CHF = congestive heart
failure; ESCAPE = European Study of Cohorts for Air Pollution Effects, HR = hazard ration, IHD = ischemic heart disease,
IQR = interquartile range; Ml = myocardial infarction; REGARDS = REasons for Geographic and Racial Differences in Stroke;
SCHIF = Shape Constrained Health Impact Function.
Note: "SCHIF models various shapes including supra-linear, near-linear, and sublinear forms and permits different shapes of the
pollutant-outcome association in a monotonically nondecreasing manner but limits the amount of curvature in the shape.
tStudies included in the 2019 Integrated Science Assessment for Particulate Matter.
'Recent studies published since the literature cutoff date (~January 2018) for the 2019 Integrated Science Assessment for
Particulate Matter.
1 Several studies evaluated the shape of the C-R function for the relationship between long-term
2 PM2 5 exposure and MI, including two analyses of the ONPHEC study (Figure 3-8 and Figure 3-9). an
3 analysis of the U.S. Medicare population (Figure 3-10). and an analysis of the REGARDS cohort (Figure
4 3-11). Approximately linear relationships were observed in the ONPHEC analyses (Chen et al.. 2020; Bai
5 et al.. 2019) using Shape Constrained Health Impact Function (SCHIF) method (Nasari et al.. 2016).
6 which is described as a new class of variable coefficient risk functions that can capturing potentially
7 non-linear associations, and in the Medicare analysis using penalized splines, which is described in
8 Section 3.1.2.1 (Danesh Yazdi etal.. 2019). Both methods allow for deviations from linearity. By
9 contrast, Loop et al. (2018) found an inverse relationship between annual average PM2 5 exposure and
10 nonfatal MI.
September 2021
3-43
DRAFT: Do Not Cite or Quote
-------
PM2.5
-------
PM2.5 vs. First Admission for Mi
o I
1 1 r
8 10 12 14
Ainuul PM2.0 (mcg/m3)
Source: Danesh Yazdi et al. C2019;. copyright permission pending.
Note: The grey shaded area represents the 95% confidence interval.
Figure 3-10 Concentration-response relationship for the association of PM2.5
concentration with first admissions for myocardial infarction.
03
N
CO
X
a>
>
1.5-
1.0-
0.5-
0.0-
CD
-0.5 H
CO
-9 -1.0 H
Nonfatal Ml
10 15
PM2.5(ng/m3)
20
Source: Loop et al. (2018). copyright permission pending.
Note: Grey bands are 95% prediction intervals.
Figure 3-11 Predicted log hazard for incident non-fatal myocardial infarction
versus previous 1-year mean ambient PM2.5 concentration.
September 2021
3-45
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
Several additional analyses evaluated the shape of the C-R relationship for atrial fibrillation and
stroke. In analyses of ONPHEC using SCHIF, a sublinear relationship was observed for atrial fibrillation
with some evidence of potential threshold at PM2 5 concentrations <6 (ig/m3 (Shin et al.. 2019). and a
linear relationship with no evidence of a threshold was observed for stroke. Danesh Yazdi etal. (2019)
also found a C-R relationship that was generally linear (i.e., at PM25 concentrations <14 |ig/m3) among
Medicare recipients. One study assessed the shape of the C-R function for CHF, which was observed to
be supralinear, flattening at higher concentrations around 14 |ig/m3 (Bai et al.. 2019).
3.1.2.3 Recent Studies Examining the PM2.5-Cardiovascular Effects
Relationship through Accountability Analyses and Causal Modeling
Methods
Several studies in the 2019 PM ISA were assessed and in general, supported an association
between long-term PM2 5 exposure and a variety of cardiovascular hospital admissions (2019 PM ISA,
Section 6.2). However, the assessment of this outcome did not include any epidemiologic studies that
conducted accountability analyses or employed causal modeling methods. Since the literature cutoff date
of the 2019 PM ISA, a few recent studies conducted accountability analyses or employed causal modeling
methods to evaluate the relationship of long-term PM2 5 exposure and cardiovascular hospital admissions
(Table A-4).
Henneman et al. (2019) utilized a difference-in-difference (DID) approach to conduct an
accountability analysis of emissions reductions from coal-fueled power plants in the U.S. between
2005-2012 on cardiovascular hospital admission rates for acute myocardial infarction, cardiovascular
stroke, heart failure, heart rhythm disorders, ischemic heart disease, peripheral vascular disease, and all
cardiovascular diseases among Medicare beneficiaries. DID methods are used to estimate the effect of a
specific treatment or intervention, such as reductions in coal-fueled power plant emissions, by comparing
the changes in outcomes overtime prior to the treatment/intervention and after. For each 1 (ig/m3
decrease in PM2 5 concentrations, the authors reported the change in hospital admissions per
10,000 person-years and found overall reductions for all cardiovascular diseases of-8.4 (95% CI: -12.67,
-4.14), -0.01 (95% CI: -0.93, 0.91) for acute myocardial infarction, -1.95 (95% CI: -3.20, -0.70) for
stroke, -4.26 (95% CI: -6.09, -2.43) for heart failure, and -3.87 (95% CI: -5.67, -2.08) for ischemic
heart disease. However, an increase was reported for heart rhythm disorders 0.96 (95% CI: -0.21, 2.12).
Overall, Henneman et al. (2019) found that reductions in annual PM2 5 concentrations from coal-fueled
power plants resulted in corresponding reductions in a number of cardiovascular-related hospital
admissions.
To examine the relationship between annual average PM2 5 concentrations and
cardiovascular-related hospital admissions including myocardial infarction, stroke, and atrial fibrillation
and flutter among Medicare beneficiaries, Danesh Yazdi et al. (2021) used a doubly-robust additive
model (DRAM). The steps for the approach used by Danesh Yazdi et al. (2021) is depicted in Figure
September 2021
3-46
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
3-12. PM2 5 concentrations were derived from a spatiotemporal ensemble model. To control for potential
confounding from individual, socioeconomic, access to care variables, and copollutants (ozone and NO2),
an inverse probability weighting approach was applied through linear propensity score models to generate
weights. The weights were then stabilized by taking the probability of the exposure as the numerator and
the denominator as the probability density of the exposure as defined on the basis of the linear regression
with PM2 5 as the outcome and covariates and other pollutants (ozone and NO2) as the predictors. If the
inverse probability weights for the exposure and the adjustment for the weights in the outcome regression
are correctly specified, it can be assumed that the estimated coefficient is unbiased. A weighted linear
probability model showed that long-term exposure to PM2 5 was associated with increased admissions
across all cardiovascular hospitalization outcomes. For each 1 (ig/m3 increase in annual average PM2 5
concentrations, the authors estimated 2,536 (95% CI: 2,383, 2,691) additional admissions for ischemic
stroke, 637 (95% CI: 483, 814) additional admissions for myocardial infarction, and 1,575 (95% CI:
1,426, 1,691) additional admissions for atrial fibrillation and flutter.
• Weighted linear
probability model
for each outcome
• Control for
individual,
socioeconomic, and
access to care
variables as well as
other pollutants
| • Obtain coefficients
• Obtain confidence
intervals
• Run secondary
analyses
• Run EMM analyses
V J
Source: From Danesh Yazdi et al. (2021), copyright permission pending.
Figure 3-12 Analysis steps used by Danesh Yazdi et al. (2021) to examine
long-term PM2.5 exposure and cardiovascular-related hospital
admissions.
The addition of these recent studies further supports the findings from the studies of the 2019 PM
ISA. Overall, these studies reported consistent findings that long-term PM2.5 exposure is associated with
increased hospital admissions for a variety of cardiovascular disease outcomes among large nationally
representative study populations. The addition of studies that utilize methods to reduce uncertainties
related to potential confounding bias with statistical methods and/or study design approaches, like DRAM
used by Danesh Yazdi et al. (2021) or the DID approach used by Henneman et al. (2019). further increase
confidence in the relationship between long-term PM2.5 exposure and cardiovascular effects.
September 2021 3-47 DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
3.1.2.4 Summary of Recent Evidence in the Context of the 2019 Integrated
Science Assessment for Particulate Matter Causality Determination
for Long-Term PM2.s Exposure and Mortality
Recent epidemiologic studies published since the 2019 PM ISA support and extend the evidence
that contributed to the conclusion of a causal relationship between long-term PM2 5 exposure and
cardiovascular effects. Numerous U.S. and Canadian cohort studies conducted in locations where the
long-term PM2 5 concentration are less than 13 (ig/m3 add to the strong evidence base that was
characterized in the 2019 PM ISA describing the relationship between long-term PM2 5 and cardiovascular
mortality, and specifically IHD- and stroke-related mortality. Overall, these recent cardiovascular
mortality studies reported positive associations at varying spatial scales and across different exposure
assessment and statistical methods. The associations between long-term PM2 5 exposure and
cardiovascular mortality generally persisted in models that were adjusted for ozone, NO2, PM10-2.5, or
SO2, and most analyses of the C-R function supported a linear, no-threshold relationship for
cardiovascular mortality, especially at lower ambient concentrations of PM2 5 (Section 3.2.2.2.2).
Although results were not entirely consistent, evidence of positive associations between
long-term PM2 5 exposure and cardiovascular morbidity (i.e., CHD, stroke, and atherosclerosis
progression) were observed in epidemiologic studies reviewed in the 2019 PM ISA, providing coherence
with the mortality findings described above. Recent studies support and the extend findings characterized
in the 2019 PM ISA, providing additional evidence of positive associations between long-term PM2 5
exposure and cardiovascular outcomes including MI, stroke, arrhythmias, atherosclerosis, HF, and
hypertension. Although positive associations are not reported in all studies, these recent studies also
support and extend the most consistent evidence of cardiovascular effects reviewed in the 2019 PM ISA,
which described positive associations among those with preexisting diseases and among patients that are
followed after a cardiac event or procedure (Rhinchart et al.. 2020; Ward-Caviness et al.. 2020; Malik et
al.. 2019; Weaver etal.. 2019). Recent studies also support and extend the evidence in the 2019 PM ISA
regarding effect measure modification by income and SES (Section 3.3.3). Together these recent studies
examining effect measure modification may explain inconsistency observed across cardiovascular
morbidity studies by identifying factors that determine the heterogeneity.
The limited number of studies reviewed in the 2019 PM ISA found that risk estimates remained
largely unchanged after adjustment for PM10-2.5, NO2, and PM2 5 from traffic sources. The few recent
analyses report some attenuation of risk estimates in models adjusted for O3 and NO2. Recent studies also
support and extend the evidence in the 2019 PM ISA pertaining to the joint effects of multiple pollutants
indicating that associations may be modified by oxidant gases, PM2 5 composition and long-term exposure
to NO2. Further, recent studies support and extend the evidence in the 2019 PM ISA pertaining to the
shape of the C-R function for cardiovascular morbidity effects. Though still limited in number, recent
studies characterizing the C-R relationship provide a more thorough examination of potential for
departures from linearity. Evidence from these studies is generally consistent with that presented in the
2019 PM ISA, and shows a linear, no-threshold C-R relationship for most CVD outcomes. However,
September 2021
3-48
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
there is some evidence for a sublinear or supralinear C-R relationship for specific outcome (i.e., CHF and
AF). Lastly, a few recent epidemiologic studies that employed causal modeling methods to reduce
uncertainties related to potential confounding bias provide additional support for a relationship between
long-term PM2 5 exposure and cardiovascular effects.
3.2 Mortality
3.2.1 Short-Term PM2.5 Exposure
The following sections represent a summary of the evidence and the corresponding causality
determination for short-term PM2 5 exposure and mortality presented within the 2019 PM ISA
(Section 3.2.1.1) along with an evaluation of recent epidemiologic studies that fall within the scope of the
Supplement (i.e., studies conducted in the U.S. and Canada) and were published since the literature cutoff
date of the 2019 PM ISA (Section 3.2.1.2). In addition, with the expansion of epidemiologic studies using
causal modeling methods, recent studies that employed such methods are also evaluated (Section 3.2.1.3).
which can further inform the relationship between short-term PM2 5 exposure and mortality. Lastly, a
summary of the results of recent studies evaluated within the section is presented in the context of the
scientific conclusions detailed in the 2019 PM ISA (Section 3.2.1.4). The evaluation of recent studies on
short-term PM2 5 exposure and mortality presented in this Supplement adds to the collective body of
evidence reviewed in the process of reconsidering the PM NAAQS.
3.2.1.1 Summary and Causality Determination from 2019 Integrated
Science Assessment for Particulate Matter
Multicity studies evaluated since the completion of the 2009 PM ISA continue to provide
evidence of primarily positive associations between short-term PM2 5 exposures and total (nonaccidental)
mortality from studies conducted mostly in urban areas using traditional exposure assignment approaches
(i.e., average of all available monitors) as well as studies with a larger spatial coverage (i.e., urban and
rural areas) employing new methods using multiple types of PM25 data (i.e., combination of monitoring,
satellite, and LUR). Additionally, the evidence from studies evaluated in the 2019 PM ISA further
substantiated the relationship between short-term PM2 5 exposure and mortality by providing additional
information on potential copollutant confounding; effect modification (e.g., stressors, pollutants, season);
geographic heterogeneity in associations; and the shape of the C-R relationship, which collectively
reaffirmed that a causal relationship exists between short-term PM2 5 exposure and mortality. The body of
evidence for total mortality was supported by generally consistent positive associations with
cardiovascular and respiratory mortality. Although there was coherence of effects across the scientific
disciplines (i.e., animal toxicological, controlled human exposure studies, and epidemiologic) and
September 2021
3-49
DRAFT: Do Not Cite or Quote
-------
1 biological plausibility for PIVb s-related cardiovascular (2019 PM ISA, Chapter 6) and respiratory (2019
2 PM ISA, Chapter 5) morbidity, there was strong evidence indicating biological plausibility for
3 PM2 5-related cardiovascular mortality with more limited evidence for respiratory mortality. This section
4 describes the evaluation of evidence for total (nonaccidental) mortality conducted in the 2019 PM ISA,
5 with respect to the causality determination for short-term exposures to PM2 5 using the framework
6 described in Table II of the Preamble to the ISAs (U.S. EPA. 2015). The key evidence, as it relates to the
7 causal framework, is summarized in Table 3-4.
Table 3-4 Summary of evidence for a causal relationship between short-term
PM2.5 exposure and total mortality from the 2019 Integrated Science
Assessment for Particulate Matter.
PM2.5
Concentrations
Rationale for Key References and Associated with
Causality Sections in the 2019 PM Effects
Determination3 Key Evidence13 ISAb (|jg/m3)c
Consistent
epidemiologic
evidence from
multiple studies at
relevant PM2.5
concentrations
Increases in mortality in multicity studies
conducted in the U.S., Canada, Europe, and
Asia.
Total mortality associations, further
supported by increases in cardiovascular
and respiratory mortality in multicity studies
conducted in the U.S., Canada, Europe, and
Asia.
Section 11.1.2
Figure 11-1
Figure 11-2
Section 5.1.9
Section 6.1.9
Mean 24-h avg:
U.S. and Canada:
4.37-17.97
Europe:
13-27.7d
Asia:
11.8-69.9
Table 11-1
Epidemiologic
evidence from
copollutant models
provides some
support for an
independent PM2.5
association
The magnitude of PM2.5 associations remain
positive, but in some cases are reduced with
larger confidence intervals in copollutant
models with gaseous pollutants and
PM10-2.5, supporting the limited evidence
from the 2009 PM ISA. Further support
comes from copollutant analyses indicating
positive associations for cardiovascular and
respiratory mortality. Recent studies that
examined potential copollutant confounding
are limited to studies conducted in Europe
and Asia.
When reported, correlations with gaseous
copollutants were primarily in the low
(.r < 0.4) to moderate (r s 0.4 or < 0.8) range.
Section 11.1.4
Figure 11-3
Section 5.1.10.1
Section 6.1.14.1
Epidemiologic
evidence supports a
linear, no-threshold
C-R relationship
Recent multicity studies conducted in the
U.S. and Europe provide direct evidence of
a linear, no-threshold C-R relationship at
lower PM2.5 concentrations with initial
evidence of a steeper slope, but extensive
systematic evaluations of alternatives to
linearity have not been conducted.
Section 11.1.10
Shietal. (2015)
Lee etal. (2015)
Pi etal. (2017a)
September 2021
3-50
DRAFT: Do Not Cite or Quote
-------
Table 3-4 (Continued): Summary of evidence for a causal relationship between
short-term PM2.5 exposure and total mortality.
Rationale for
Causality
Determination3
Key Evidence13
Key References and
Sections in the 2019 PM
ISAb
PM2.5
Concentrations
Associated with
Effects
(Hg/m3)c
Biological plausibility
from cardiovascular
morbidity evidence
Strong evidence for coherence of effects
across scientific disciplines and biological
plausibility for a range of cardiovascular
effects in response to short-term PM2.5
exposure, specifically for ischemic events
and heart failure, which is supported by
experimental evidence and epidemiologic
studies examining hospital admissions and
ED visits. The collective body of
cardiovascular morbidity evidence provides
biological plausibility for a relationship
between short-term PM2.5 exposure and
cardiovascular mortality, which comprises
-33% of total mortality.
Section 6.1.16
Table 6-33
Limited biological
plausibility from
respiratory morbidity
evidence
Limited evidence for coherence of effects
across scientific disciplines and biological
plausibility, with the strongest evidence for
exacerbations of COPD and asthma. The
collective body of respiratory morbidity
evidence provides limited biological
plausibility for a relationship between
short-term PM2.5 exposure and respiratory
mortality, which comprises -9% of total
mortality.
Section 5.1.12
Table 5-18
Uncertainty
regarding
geographic
heterogeneity in
PM2.5 associations
Multicity U.S. studies demonstrate
city-to-city and regional heterogeneity in
PM2 5-mortality associations. Evidence
supports that a combination of factors,
including composition and exposure factors
may contribute to the observed
heterogeneity.
Section 11.1.6.3
avg = average; COPD = chronic obstructive pulmonary disease; C-R = concentration-response; ED = emergency department;
h = hour; PM = particulate matter; PM25 = particulate matter with a nominal mean aerodynamic diameter less than or equal to
2.5 |jm; PM10-2 5 = particulate matter with a nominal mean aerodynamic diameter greater than 2.5 |jm and less than or equal to
10 |jm; r= correlation coefficient.
aBased on aspects considered in judgments of causality and weight of evidence in causal framework in Table I and Table II of the
Preamble to the ISAs U.S. EPA (2015V
bDescribes the key evidence and references, supporting or contradicting, contributing most heavily to the causality determination
and, where applicable, to uncertainties or inconsistencies. References to earlier sections indicate where the full body of evidence
is described in the 2019 PM ISA.
°Describes the PM25 concentrations with which the evidence is substantiated.
dMedian concentration from Samoli et al. (2013V
statistics taken from NHLBI (2017V
September 2021
3-51
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Collectively, the evidence from multicity studies of short-term PM2 5 exposures and mortality
evaluated in the 2019 PM ISA generally demonstrated positive associations with total (nonaccidental)
mortality, with increases ranging from 0.19% (Lippmann et al.. 2013) to 2.80% (Kloog et al.. 2013) at
lags of 0 to 1 days in single-pollutant models. These results were further supported by initial studies
employing causal inference and quasi-experimental statistical approaches (2019 PM ISA,
Section 11.1.2.1). Whereas most studies relied on assigning exposure using data from ambient monitors,
some of the studies evaluated also employed approaches that used all available PM2 5 data (i.e., monitor,
satellite, and LUR), allowing for the inclusion of less urban and rural locations in the analyses (Lee et al..
2015; Shi et al.. 2015; Kloog et al.. 2013). The studies evaluated expanded the assessment of potential
copollutant confounding on the PM2 5-mortality relationship, and provided additional evidence supporting
the conclusion that PM2 5 associations remain positive and relatively unchanged in copollutant models
with both gaseous pollutants and PM10-2.5, but this conclusion was based on a limited number of multicity
studies conducted in Europe and Asia where mean 24-hour avg PM2 5 concentrations are higher (2019 PM
ISA, Table 3-1). However, the low (r < 0.4) to moderate correlations (r = 0.4 < 0.7) between PM2 5 and
gaseous pollutants and PM10-2.5 increased the confidence in PM2 5 having an independent effect on
mortality.
The positive associations for total (nonaccidental) mortality reported across the majority of
studies evaluated was further supported by analyses focusing on cause-specific mortality that continue to
provide evidence of generally consistent positive associations with both cardiovascular and respiratory
mortality, except in the case of a multicity study conducted in Europe (Lanzingcr et al.. 2016). Risk
estimates for cardiovascular mortality ranged from 0.09% (Lippmann et al.. 2013) to 2.32% (Lee et al..
2015). while those for respiratory mortality ranged from 0.09% (Lee et al.. 2015) to 2.30% (Janssen et al..
2013). but overall associations tended to be larger in magnitude for respiratory mortality. For both
cardiovascular and respiratory mortality there was a limited assessment of potential copollutant
confounding, but for both outcomes, initial evidence indicated that associations remained positive and
relatively unchanged in models with gaseous pollutants and PM10-2.5, which further supported the
copollutant analyses conducted for total (nonaccidental) mortality. The strong evidence for ischemic
events and heart failure detailed in the assessment of cardiovascular morbidity (2019 PM ISA, Chapter 6),
provided strong biological plausibility for PM2 5-related cardiovascular mortality, which comprises the
largest percent of total mortality [i.e., -33%; NHLBI (2017)1. Although there was evidence for
exacerbations of COPD and asthma, the collective body of respiratory morbidity evidence provided
limited biological plausibility for PM2 5-related respiratory mortality (2019 PM ISA, Chapter 5).
In addition to examining potential copollutant confounding, a number of studies also assessed
whether statistical models adequately accounted for temporal trends and weather covariates. Across
studies that evaluated model specification, PM2 5-mortality, associations remained positive, although in
some cases were attenuated, when using different approaches to account for temporal trends or weather
covariates (2019 PM ISA, Section 11.1.5). Seasonal analyses continued to provide evidence that
September 2021
3-52
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
associations were larger in magnitude during warmer months, but it remained unclear whether
copollutants confound the associations observed. In addition to seasonal analyses, some studies also
examined whether temperature modified the PM2 5-mortality relationship. Initial evidence indicated that
the PM2 5-mortality association may be larger in magnitude at lower and higher temperatures, but this
observation has not been substantiated by studies conducted in the U.S. (2019 PM ISA, Section 11.1.6.2).
At the completion of the 2009 PM ISA, one of the main uncertainties identified was the regional
and city-to-city heterogeneity in PM2 5-mortality associations observed in multicity studies. Studies
evaluated in the 2019 PM ISA examined both city specific as well as regional characteristics to identify
the underlying factors that contribute to this heterogeneity (2019 PM ISA, Section 11.1.6.3). Analyses
focusing on effect modification of the PM2 5-mortality relationship by PM2 5 components, regional
patterns in PM2 5 components, and city-specific differences in composition and sources indicated some
differences in the PM2 5 composition and sources across cities and regions, but these differences did not
fully explain the heterogeneity observed. Additional studies examined whether exposure factors play a
role in explaining the heterogeneity in PM2 5-mortality associations and found that some factors related to
housing stock and commuting, as well as city-specific factors (e.g., land use, port volume, and traffic
information), also explain some of the observed heterogeneity. Collectively, the studies evaluated
indicated that the heterogeneity in PM2 5-mortality risk estimates cannot be attributed to one factor, but
instead to a combination of factors, including, but not limited to, compositional and source differences, as
well as exposure differences.
A number of studies evaluated conducted systematic evaluations of the lag structure of
associations for the PM2 5-mortality relationship by examining either multiday lags or a series of
single-day lags, and these studies continued to support an immediate effect (i.e., lag 0 to 1 days) of
short-term PM2 5 exposures on mortality (2019 PM ISA, Section 11.1.8.1). Studies also conducted
analyses comparing the traditional 24-hour avg exposure metric with a subdaily metric (i.e., 1-hour max).
These initial studies provided evidence of a similar pattern of associations for both the 24-hour avg and
1-hour max metric, with a larger association for the 24-hour avg metric. Additionally, some studies
examined alternative exposure metrics representing size fractions smaller than PM2 5 and reflecting
number concentration (NC) and surface-area concentration (SC). The generally positive associations
reported with mortality for these smaller PM size fractions supported the larger body of PM2 5-mortality
evidence, but it is difficult to compare NC and SC metrics with the traditional mass-based metric.
Building off the initial analysis of the C-R relationship between short-term PM exposure and
mortality that focused on PM10, multicity studies conducted in the U.S. and Europe examined the shape of
the C-R relationship and whether a threshold exists specifically for PM2 5 (2019 PM ISA,
Section 11.1.10). These studies used different statistical approaches and consistently demonstrated a
linear relationship with no evidence of a threshold. Additionally, recent analyses conducted at lower
PM2 5 concentrations (i.e., 24-hour avg PM2 5 concentrations <30 (.ig/ni') provided initial evidence
indicating that PM2 5-mortality associations persist and may be stronger (i.e., a steeper slope) at lower
September 2021
3-53
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
concentrations. However, to date, extensive analyses have not been conducted to systematically explore
alternatives to linearity when examining the shape of the PM2 5-mortality C-R relationship.
Overall, epidemiologic studies evaluated in the 2019 PM ISA built upon and further reaffirm the
conclusions of the 2009 PM ISA for total mortality. The evidence particularly from the assessment of
PM2 5-related cardiovascular morbidity, with more limited evidence from respiratory morbidity, provided
biological plausibility for mortality from short-term PM2 5 exposures. In conclusion, the primarily positive
associations observed across studies conducted in various locations was further supported by the results
from copollutant analyses that indicated robust associations, along with evidence from analyses of the
C-R relationship. Collectively, this body of evidence is sufficient to conclude that a causal
relationship exists between short-term PM2.5 exposure and total mortality.
3.2.1.2 Recent U.S. and Canadian Epidemiologic Studies
In addition to evaluating epidemiologic studies that examined the relationship between short-term
PM2 5 exposure and mortality, the 2019 PM ISA characterized whether evidence supported biologically
plausible mechanisms by which short-term PM2 5 exposure could lead to mortality. This evaluation
consisted of an assessment of animal toxicological, controlled human exposure, and epidemiologic studies
of morbidity effects that are the largest contributors to total (nonaccidental) mortality, specifically,
cardiovascular and respiratory morbidity (2019 PM ISA, Section 6.1.1 and Section 5.1.1, respectively).
Plausible mechanisms were identified by which inhalation exposure to PM2 5 could progress from initial
events to endpoints relevant to the cardiovascular system and to population outcomes such as emergency
department (ED) visits and hospital admissions due to cardiovascular disease, particularly ischemic heart
disease and congestive heart failure (2019 PM ISA, Section 6.1.1). Similarly, available evidence was
characterized by which inhalation exposure to PM2 5 could progress from initial events to endpoints
relevant to the respiratory system (2019 PM ISA, Section 5.1.1). However, the evidence for how the
initial events and subsequent endpoints could lead to the observed increases in respiratory ED visits and
hospital admissions, in particular for chronic obstructive pulmonary disease (COPD) and asthma was
limited. Collectively, the progression demonstrated in the available evidence for cardiovascular morbidity
(and to a lesser extent, respiratory morbidity) supported potential biological pathways by which
short-term PM2 5 exposures could result in mortality (2019 PM ISA, Section 11.1).
The few recent multicity studies conducted in the U.S. and Canada build upon the strong
epidemiologic evidence base evaluated in the 2019 PM ISA, as well as previous assessments, that
provided the scientific rationale supporting a causal relationship between short-term PM2 5 exposure and
mortality (Section 3.1.1. IV In addition to examining the relationship between short-term PM2 5 exposure
and all-cause or nonaccidental mortality (Section 3.1.1.2.1) and cause-specific mortality
(Section 3.1.1.2.2). additional analyses within these recent studies also further examined issues relevant to
expanding the overall understanding of the effect of short-term PM2 5 exposure on mortality. Specifically,
September 2021
3-54
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
recent studies have assessed potential copollutant confounding (Section 3.1.1.2.3). examined effect
modification of the PM2 5-mortality relationship (Section 3.1.1.2.4). the lag structure of associations
(Section 3.1.1.2.5). and assessed the shape of the concentration-response (C-R) relationship
(Section 3.1.1.2.6). The following sections evaluate recent multicity studies that inform each of the
aforementioned topics within the context of the evidence base evaluated and summarized in the 2019 PM
ISA.
3.2.1.2.1 All Cause and Total (Nonaccidental) Mortality
Since the literature cutoff date for the 2019 PM ISA, a limited number of multicity studies have
been conducted within the U.S. and Canada (Shin et al.. 2021a; Liu et al.. 2019; Lavigne et al.. 2018).
Although few in number, these recent studies add to the extensive number of multicity studies evaluated
in the 2019 PM ISA that were conducted worldwide, specifically in locations where mean 24-hour
concentrations were generally <20 (ig/m3 (2019 PM ISA, Section P.3.1). Taken together, these studies
provide consistent evidence of positive associations between short-term PM2 5 exposure and mortality
across diverse geographic locations; in populations with a wide range of demographic characteristics; and
using a variety of statistical models, approaches to confounder adjustment, and exposure assessment
approaches (Figure 3-13).
September 2021
3-55
DRAFT: Do Not Cite or Quote
-------
Study
Burnett and Goldberg (2003)
Klemmand Mason (2003)
Burnett et al. (2004)
Zanobettiand Schwartz (2009)
Dominici et al. (2007)
Franklin et al. (2007)
Franklin et al. (2008)
Ostro et al. (2006)
tLippmann et al. (2013)
tBaxteretal. (2017)
tDai et al. (2014)
tKrall etal. (2013)
tKloog et al. (2013)
+Lee et al. (2015)a
tjanssen et al. (2013)
tsamoli etal (2013)
tstafoggia et al. (2017)
tLanzinger et al. (2016)b
tpascal etal.(2014)
tLeeetal. (2015)
*Liu et al. (2019)c
*Liu et al. (2019)c
'Lavigne et al. (2018)
*Shin etal. (2021)
tDi etal. (2017)c
tZanobetti etal.(2014)c
tshi et al. (2015)c
tYoung etal. (2017)
tueda et al. (2009)f
t Atkinson et al (2014)
tAdar etal. (2014)
Location Lag
8 Canadian cities 1
6 U.S. cities 0-1
12 Canadian cities 1
112 U.S. cities 0-1
96 U.S. cities (NMMAPS) 1
27 U.S. cities 1
25 U.S. cities 0-1
9 CA counties 0-1
148 U.S. cities 0
77 U.S. cities 0-1
75 U.S. cities 0-1
72 U.S. cities 1
New England, U.S. 0-1
3 Southeast states, U.S. 0-1
Netherlands 0
10 European Med cities 0-1
8 European cities 1
5 Central European cities (UFIREG) 0-1
9 French cities 0-1
11 East Asian cities 0-1
107 U.S. cities 0-1
25 Canadian cities 0-1
24 Canadian cities 0-2
22 Canadian cities 0
U.S. - Nation 0-1
121 U.S. cities 0-1
New England, U.S. 0-1
8 CA air basins 0-ld
8 CA air basins 0-3e
20 Japanese areas 1
Meta-analysis —g
Meta-analysis —h
All Ages
> 1
65+
All Ages
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
% Increase (95% Confidence Interval)
Source: Update of Figure 11-1, 2019 PM ISA.
avg = average; |jg/m3 = microgram per cubic meter; NMMAPS = National Morbidity, Mortality, and Air Pollution Study;
PM = particulate matter; PM25 = particulate matter with a nominal mean aerodynamic diameter less than or equal to 2.5 pm;
UFIREG = Ultrafine Particles—An Evidence-Based Contribution to the Development of Regional and European Environmental and
Health Policy.
Note: tStudies published since the 2009 PM ISA. *U.S. and Canadian studies published since the literature cutoff date (~January
2018) 2019 PM ISA. Black circles = U.S. and Canadian multicity studies evaluated in the 2004 PM AQCD and 2009 PM ISA. Red
circles = multicity studies and meta-analyses published since the completion of the 2009 PM ISA. Blue circles = multicity U.S. and
Canadian studies published since the literature cutoff date of the 2019 PM ISA. Risk estimates are standardized to a 10 pg/m3
increase in PM2.5 concentrations.
aResults are from modeled PM2.s analysis, analysis focusing on measured PM2.5 reported 1.21 % (95% CI: 0.94, 1.47).
bOnly four of the five cities measured PM2 5.
cShi et al. (2015). Zanobetti et al. (2014). and Liu et al. (2019) only had data for all-cause mortality including accidental mortalities.
dMain model used in Young et al. (2017) included current and average of 3 previous days daily maximum temperature, daily
minimum temperature, and maximum daily relative humidity.
eSensitivity analysis in Young et al. (2017 focusing on only the San Francisco Bay air basin, dropping out the maximum daily
relative humidity term, where the shortest duration of lag days examined was 0-3 days.
fUeda et al. (2009) presented results for three different modeling approaches, which are presented here: generalized additive model
(GAM), generalized linear model (GLM), and case-crossover.
gAtkinson et al. (2014) primarily focused on single-day lag results.
hAdar et al. (2014) focused on single-day lag results, specifically lag 0, 1, or 2.
Figure 3-13 Summary of associations between short-term PM2.5 exposure and
total (nonaccidental) mortality in multicity studies.
1 Recent studies that conducted multicity analyses in the U.S. and Canada include a large
2 international study that performed a worldwide multicity analysis (Liu et al.. 2019) and a few studies in
3 Canada that relied on data from over 20 cities (Shin et al.. 2021a; Lavigne et al.. 2018). Liu et al. (2019)
September 2021
3-56
DRAFT; Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
established a Multi-City Multi-Country (MCC) Collaborative Research Network that allowed the
investigators to collect data globally, resulting in air pollution and mortality data from 652 urban areas in
24 countries from 1986-2015. Although the goal of the study was to estimate a global estimate of the
association between short-term PM2 5 exposure and mortality, the authors presented country specific
estimates as well, including for the U.S. and Canada. The authors applied a uniform statistical model
across all of the cities within the study consisting of a quasi-Poisson general additive model that
controlled for temporal trends and weather covariates. In a second-stage analysis, the authors used a
random effects models to pool city-specific estimates into a country-specific estimate. All analyses relied
on PM2 5 data where the highest and lowest 5% of data was trimmed to remove outliers. In analyses of 25
Canadian cities from 1986-2011 and 107 U.S. cities from 1987-2006, the authors reported a 1.70% (95%
CI: 1.17, 2.23) and 1.58% (95% CI: 1.28, 1.88) increase in mortality, respectively, at lag 0-1 days.
Recent studies conducted in Canada by Lavigne et al. (2018) and Shin et al. (202 la) focused on
more recent years of data, 1998-2011 and 2001-2012, respectively, in comparison to Liu et al. (2019).
Lavigne et al. (2018) focused on examining whether oxidant gases modified the association between
short-term PM2 5 exposure and mortality in 24 Canadian cities (discussed in more detail in
Section 3.1.1.2.4). In a time-stratified case-crossover analysis that adjusted for both mean temperature and
location-specific temperature distributions and relative humidity, the authors reported a 0.76% (95% CI:
0.15, 1.21) increase in mortality at lag 0-2 days. The results of Lavigne et al. (2018) are consistent with
those reported by Shin et al. (2021a) in a time-series study of 22 Canadian cities. The authors examined
single-day lags ranging from 0 to 2 days using a two-stage hierarchical model consisting of a Poisson
model in the first stage to examine city-specific associations and a Bayesian random effects model in the
second stage to combine city-specific effects into a national estimate. Shin et al. (2021a) reported
associations with mortality similar in magnitude at lag 0 (0.94% [0.43, 1.46]) and 1 day (0.90% [95% CI:
0.33, 1.41]) with no evidence of an association at lag 2. Although it is unclear as to why the magnitude of
associations reported in Lavigne et al. (2018) and Shin et al. (2021a) differ from those reported by Liu et
al. (2019). even though both are using a similar subset of cities, it could be attributed to the more recent
years of data used in Lavigne et al. (2018) and Shin et al. (2021a) where there has been a decreasing trend
in PM2 5 concentrations (Shin et al.. 2021a).
Other nonaccidental mortality
The 2009 PM ISA reviewed a handful of small studies examining the association between PM2 5
exposure and out-of-hospital cardiac arrest (OHCA). No evidence of an association was reported.
Section 6.1.4.1 of the 2019 PM ISA evaluated studies published since the 2009 PM ISA, which provided
evidence for an association between short-term PM2 5 exposure and OHCA. This association was typically
observed with PM2 5 concentrations averaged over the past 0 to 2 days, although associations with PM2 5
concentrations as far back as 4 days before the event have been reported. Additionally, all of the studies
assessed in the 2009 and 2019 ISAs relied on a single monitor or an average of fixed-site monitors to
estimate PM2 5 exposure, which restricts the study population to people living near monitors. While the
September 2021
3-57
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
previously evaluated studies focused on a cardiovascular outcome, i.e., OHCA, and as a result where
discussed within the evidence for short-term PM2 5 exposure and cardiovascular effects, the results of
these previous studies are summarized here as they can inform a recent study by Rappazzo et al. (2019)
that examined out-of-hospital sudden unexpected deaths.
Rappazzo et al. (2019) conducted a time-stratified case-crossover analysis to examine the
relationship between short-term PM2 5 exposure and out-of-hospital nonaccidental sudden unexpected
deaths in a small population of individuals (n = 399) over a 2-year period that resided in Wake County,
NC. In analyses examining both single-day lags ranging from 0 to 3 days and a 0-1 day lag, the authors
reported a positive association at lag 1 (OR = 1.39 [95% CI: 0.96, 1.99]) that was smaller in magnitude
when using a 0-1 day lag (OR =1.18 [95% CI: 0.79, 1.78]). However, due to the small sample size
within this study confidence intervals are large. In addition to single-pollutant models, the authors
examined copollutant models across the single-day lags and reported that PM2 5 associations are relatively
unchanged in models with NO2, SO2, ozone, and CO. This initial study focusing out-of-hospital sudden
unexpected deaths from all nonaccidental causes, provides evidence consistent with the relatively limited
number of previous studies examining OHCA.
3.2.1.2.2 Cause-specific Mortality
Single and multicity studies evaluated in the 2009 PM ISA that examined cause-specific mortality
reported consistent positive associations with both cardiovascular and respiratory mortality. The
magnitude of the association was larger for respiratory mortality, but these associations also had wider
confidence intervals due to the smaller number of respiratory-related deaths than cardiovascular-related
deaths. Studies evaluated in the 2019 PM ISA added to this body of evidence but provided more
consistent evidence of associations for cardiovascular mortality compared to respiratory mortality. Recent
multicity studies conducted in Canada provide additional support for an association with cardiovascular
mortality (Lavigne et al.. 2018). Consistent with the evidence assessed in previous ISAs, recent studies
report more variable results with wider confidence intervals for respiratory mortality (Shin et al.. 2021b:
Lavigne et al.. 2018).
September 2021
3-58
DRAFT: Do Not Cite or Quote
-------
Study
Location
Lag
Zanobetti et al. (2009)
112 U.S. cities
0-1
Ostro et al. (2006)
9 CA counties
0-1
Franklin et aL (2008)
25 U.S. cities
0-1
Franklin et aL (2007)
27 U.S. cities
1
tLippmann etaL (2013)
148 U.S. cities
1
tDai et aL (2014)
75 U.S. cities
0-1
tLeeetal. (2015)
3 Southeast states, U.S.
0-1
*Lavigne etaL (2018)
24 Canadian cities
0-2
tSamoH etaL (2013)
10 European Med cities
0-1
fPascal et aL (2014)
9 French cities
0-1
tLanzinger et al. (2016)a
5 Central European cities (UFIREG)
0-1
tJanssen et al. (2013)
Netherlands
0
fLee et al. (2015)
11 Asian cities
0-1
tChen etaL (2011)
3 Chinese cities (CAPES)
0
tAtkinson et al. (2014)
Meta-analysis
—a
tAdar et al. (2014)
Meta-analysis
—b
Zanobetti et al. (2009)
112 U.S. cities
0-1
Ostro et al. (2006)
9 CA counties
0-1
Franklin et al. (2008)
25 U.S. cities
1-2
Franklin et al. (2007)
27 U.S. cities
1
tLippmann etaL (2013)
148 U.S. cities
1
fDai etaL (2014)
75 U.S. cities
0-1
tLee et al. (2015)
3 Southeast states, U.S.
0-1
*Shin etaL (2021)
24 Canadian cities
1
*Lavigne et aL (2018)
22 Canadian cities
0-2
tSamoK etaL (2013)
10 European Med cities
0-5
jPascal et aL (2014)
9 French cities
0-1
tLanzinger et al. (2016)a
5 Central European cities (UFIREG)
2-5
tJanssen et al. (2013)
Netherlands
3
tLee et al. (2015)
11 Asian cities
0-1
tChen etaL (2011)
3 Chinese cities (CAPES)
0
tAtkinson et al. (2014)
Meta-analysis
—b
tAdar et aL (2014)
Meta-analysis
—c
Cardiovascular
Respirator*"
-2.0 -1.0 0.0 1.0 2.0 3.0
% Increase (95% Confidence Interval)
Source: Update of Figure 11-2, 2019 PM ISA.
avg = average; |jg/m3 = microgram per cubic meter; PM = particulate matter; PM25 = particulate matter with a nominal mean
aerodynamic diameter less than or equal to 2.5 pm; UFIREG = Ultrafine Particles—An Evidence-Based Contribution to the
Development of Regional and European Environmental and Health Policy.
Note: tStudies published since the 2009 PM ISA. *U.S. and Canadian studies published since the literature cutoff date (~January
2018) for the 2019 PM ISA. Studies organized by lag structure, therefore, cardiovascular and respiratory mortality results are not in
the same order. Black circles = U.S. and Canadian multicity studies evaluated in the 2004 PM AQCD and 2009 PM ISA. Red
circles = multicity studies and meta-analyses published since the literature cutoff date of the 2009 PM ISA. Risk estimates are
standardized to a 10 pg/m3 increase in PM25 concentrations. All ages examined for all studies except Lanzinger et al. (2016 and
Shin et al. (2021 b) which focused on ages >1 year old.
aOnly four of the five cities measured PM2 5.
bAtkinson et al. (2014) primarily focused on single-day lag results.
cAdaret al. (2014) focused on single-day lag results, specifically lag 0, 1, or 2.
Figure 3-14 Summary of associations between short-term PM2.5 exposure and
cardiovascular and respiratory mortality in multicity studies.
3.2.1.2.3 Potential Copollutant Confounding of the PM2.5-Mortality
Relationship
1 At the time of the 2009 ISA, only a few studies had assessed the potential for confounding of the
2 PM2 5-mortality association by co-occurring pollutants. In contrast, the 2019 ISA included a number of
3 multicity studies that used copollutant models to evaluate this issue, including studies that examined both
4 gaseous pollutants and other particle size fractions. These studies reported that associations were
September 2021
3-59
DRAFT; Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
relatively unchanged in copollutant models, albeit with wider confidence intervals than single pollutant
models (2019 PM ISA, Figure 11-3).
Of the recent multicity studies conducted in the U.S. and Canada, only Lavigne et al. (2018)
conducted an assessment of copollutant confounding, with a focus on oxidant gases. Within this study,
oxidant gases were defined as the daily combined oxidant capacity of ozone and NO2 based on the
redox-weighted averages of both pollutants. In a copollutant model with oxidant gases, the association
between short-term PM2 5 exposure and mortality is unchanged compared to the single pollutant model
with both reporting a 0.76% increase in mortality at lag 0-2 days.
3.2.1.2.4 Effect Modification of the PM2.5-Mortality Relationship
Multicity epidemiologic studies evaluated in the 2009 PM ISA and 2019 PM ISA provided
evidence of city-to-city and regional heterogeneity in PM2 5-mortality associations. Within the 2019 PM
ISA, studies were evaluated that examined factors that could modify the PM2 5-mortality association and
potentially explain some of the observed heterogeneity in associations, including season (2019 PM ISA,
Section 11.1.6.1), temperature (2019 PM ISA, Section 11.1.6.2), and city and regional characteristics
(Section 11.1.6.3) such as composition/mixtures (2019 PM ISA, Section 11.1.6.3.1) and exposure factors
(2019 PM ISA, Section 11.1.6.3.2). Recent multicity studies provide additional insight into some of these
factors that could modify the PM2 5-mortality association.
The 2009 PM ISA reported some evidence that PM2 5-mortality associations are larger in
magnitude during the warm season, specifically the spring, with the majority of this evidence coming
from U.S. multicity studies (Zanobetti and Schwartz. 2009; Franklin et al.. 2008). As discussed in
Section 11.1.6.1 of the 2019 PM ISA, across recent multicity studies, there was general agreement that
PM2 5-mortality associations were larger in magnitude during warmer months. However, it remained
unclear whether copollutants confound the seasonal patterns in the associations observed. Across most
studies, the pattern of seasonal associations persisted using different methods to examine whether there
was evidence of seasonal differences in associations, with some studies relying on stratified analyses (Dai
et al.. 2014; Samoli et al.. 2013) and others incorporating interaction terms between PM2 5 and season
(Pascal et al.. 2014; Lippmann et al.. 2013). The recent studies conducted by Shin et al. (2021a) and Shin
et al. (2021b) further inform seasonal analyses, but do not address the uncertainties identified in the 2019
PM ISA. Both studies assessed associations by season through stratified analyses where the warm season
is defined as April-September and the cold season as October-March. In Shin et al. (202 la), when
focusing on lag 0, which had the largest magnitude of an association in all-year analyses, there is a clear
pattern of the warm season driving the overall association; however, the reverse pattern is reported when
focusing on lag 1, complicating the overall interpretation of results from this study. However, in Shin et
al. (2021b). which focused on respiratory mortality a slight larger association, with wide confidence
intervals, is reported for the warm season (1.0% [95% CI: -1.6, 3.5]) compared to the cold season (0.6%
[95% CI: -2.2, 4.1]) at lag 1, the main lag examined for PM2 5 and mortality within the study. Across
September 2021
3-60
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
these recent studies there continues to be some evidence indicating larger associations during the warm
season, but there are inconsistencies across the individual lags examined.
Within the 2019 PM ISA, an assessment of composition and mixtures (2019 PM ISA,
Section 11.1.6.3.1) focused on whether differences in the pollutant mixture across cities could explain
heterogeneity in the PM2 5-mortality association across cities and regions of the U.S. In the process of
examining the association between short-term PM2 5 exposure and mortality across 24 Canadian cities,
Lavigne et al. (2018) did not focus on whether effect modification by other pollutants could explain
heterogeneity, but broadly whether oxidant gases modify the PM2 5-mortality association. In the process
of evaluating effect modification, the authors conducted stratified analyses across tertiles of oxidant gases
based on the distribution of oxidant gases across all cities. For nonaccidental mortality, in analyses
examining both lag 0 and lag 0-2 days for PM2 5, there was a consistent pattern of the PM2 5-mortality
association being larger in magnitude for the 3rd tertile of oxidant gases. However, there is some
variability in the PM2 5-mortality association depending on the lag structure used to represent oxidant
gases, with some evidence indicating that as the exposure for oxidant gases increased in length, there are
larger PM2 5-mortality associations for both the 2nd and 3rd tertiles (Figure 3-15). The pattern of
associations for nonaccidental mortality is similar for cardiovascular mortality, but there is no evidence of
effect modification for respiratory mortality.
September 2021
3-61
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
Ox lag 0
PM2.5lag0 PM2.53-day mean PM2.5lagO
Ox 3-day mean
PM2.5 3-day mean
Ox 5-day mean
PM2.5lagO PM2.5 3-day mean
it
I II III
Ox lag 0
tertiles
I II III
Ox lag 0
tertiles
I II III
Ox lag 0
tertiles
I II III
Ox lag 0
tertiles
Ox lag 0
tertiles
Source: Copyright permission pending from Laviane et al. (20181.
Figure 3-15 Odds ratio and 95% confidence intervals for lag 0 and lag
0-2 days of nonaccidental mortality across tertiles of lag 0, lag
0-2, and lag 0-4 oxidant gases across 24 Canadian cities.
A recent study by Baxter et al. (2019) expands upon studies evaluated in the 2019 PM ISA
(Baxter et al.. 2017; Baxter and Sacks. 2014). which provided evidence indicating that combinations of
exposure factors representative of residential infiltration (i.e., prevalence of central AC, mean year home
was built, and mean size of home) explained some of the heterogeneity in the PM2 5-mortality association.
In a time-series analysis, Baxter et al. (2019) examined the association between short-term PM2 5
exposure and nonaccidental mortality in 312 core-based statistical areas (CBSAs) within the U.S. from
1999-2005. In a two-stage analysis, the authors first examined associations with mortality in each CBSA
in a time-series analysis and then conducted a meta-regression using a fixed-effects inverse variance
weighted linear regression to examine whether individual exposure factors or combinations of exposure
factors explained observed heterogeneity. The variables examined within the meta-regression fall within
five categories representative of housing characteristics, commuting, household heating, meteorological
factors, and poverty measures. In the first-stage analysis, Baxter et al. (2019) reported a 0.95% (IQR of
September 2021
3-62
DRAFT: Do Not Cite or Quote
-------
1 2.25)' increase in mortality across all CBSAs, but as depicted in Figure 3-16 there is extensive city-to-city
2 variability in associations across the U.S.
Fffert Mfl „n/m3V B < "1-5% D "1-5 t0 "10% D "1 t0 "0-5% D "°'5 t0 0%
Effect Estimate (10 Mg/m ) Q Q tQ 0_5o/o n 0 5 to 1M ~ 1,0 to 1.5% ¦ > 1.5%
Source; Copyright permission pending from Baxter et ai. (2019).
Figure 3-16 Associations between short-term PM2.5 exposure and
nonaccidental mortality at lag 1 for the 312 core-based statistical
areas examined in Baxter et al. (2019).
3 In the second-stage analysis, the authors conducted both a univariate and multivariate
4 meta-regression. In the univariate regression, mortality associations larger in magnitude were observed
5 for CBSAs with larger homes, more heating degree days, and a higher percentage of homes heating with
6 oil, while cities with more gas heating had smaller associations. Across all univariate analyses, no
7 individual factor explained much of the heterogeneity as reflected by R2 < 1%. For the multivariate
8 model, a backwards selection approach was used to develop the final model that included variables for
795% CIs were not presented in this study.
September 2021
3-63
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
gas heating use, heating degree days, cooling degree days, and variables for home size and age. Compared
to the univariate models, the multivariate models explained a larger amount of the heterogeneity in
mortality associations across the CBSAs examined, ranging from 11-13%. Overall, the results of Baxter
et al. (2019) further support studies evaluated in the 2019 PM ISA, which indicated that a combination of
factors that influence exposure to PM2 5, not an individual factor, explains some of the observed
city-to-city and regional heterogeneity reported in multicity epidemiologic studies.
3.2.1.2.5 Lag Structure of Associations
Within the 2009 PM ISA, the studies evaluated indicated that the effect of short-term PM2 5
exposure on mortality was immediate, occurring within the first few days after exposure, with the
strongest evidence, in terms of magnitude and precision of the associations, in the range of 0 to 1 day.
However, these studies defined the lags to examine a priori, often in accordance with the l-in-3 or
l-in-6-day sampling schedule of ambient PM25 monitors. As detailed in Section 11.1.8.1 of the 2019 PM
ISA, some studies published since the completion of the 2009 PM ISA conducted more extensive
examinations of the lag structure of associations for short-term PM2 5 exposures and mortality and
continue to support associations being largest in terms of magnitude and precision primarily within the
first few days of exposure (i.e., lags of 0 to 1 day) as depicted in Figure 3-14.
Of the recent studies evaluated, Lavigne et al. (2018) and Shin et al. (2021a) in multi-city studies
conducted in Canada examined the lag structure of associations primarily through examining single-day
lags ranging from 0 to 2 days, with Lavigne et al. (2018) also examining a multi-day lag of 0-1 days. In
the single-day lag analyses, Lavigne et al. (2018) and Shin et al. (2021a) both reported positive
associations relatively similar in magnitude at lag 0 and 1 with no evidence of an association at lag 2. In
the multi-day lag analysis focusing on lag 0-1 day, Lavigne et al. (2018) reported results similar in
magnitude to the single-day lag analysis of lag 0 and 1 day. The single-day lag analyses in combination
with the multi-day lag analysis conducted by Lavigne et al. (2018) supports the conclusions of previously
evaluated studies in the 2009 and 2019 PM ISA that indicated associations largest in magnitude at lags of
0 to 1 day.
3.2.1.2.6 Examination of the Concentration-Response (C-R) Relationship
between Short-Term PM2.5 Exposure and Mortality
In the 2009 PM ISA, the examination of the PM-mortality C-R relationship was limited to studies
of PM10. Within the multicity studies examined, there was evidence of a linear, no-threshold C-R
relationship between short-term PM exposures and mortality with some evidence of differences in the
shape of the C-R curve across cities. Studies evaluated in the 2019 PM ISA, focused specifically on
examining the C-R relationship between short-term PM2 5 exposure and mortality. Although difficulties
remain in assessing the shape of the PM2 5-mortality C-R relationship, as identified in the 2009 PM ISA,
September 2021
3-64
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
and studies had not conducted systematic evaluations of alternatives to linearity, the studies evaluated in
the 2019 PM ISA continued to provide evidence of a no-threshold linear relationship, with less
confidence at concentrations lower than 5 |ig/nr\ Additionally, those studies that conducted analyses
focused on examining associations at lower PM2.5 concentrations provided initial evidence indicating that
associations persist and may be larger (i.e., have a steeper slope) at lower PM2.5 concentrations.
Liu et al. (2019) in the worldwide analysis examining the association between short-term PM2.5
exposure and mortality in 652 cities, also examined country-specific C-R relationships. For each country,
a linear term for PM2.5 was added to the main model with a B-spline function with knots at the 25th and
75th percentiles of the mean PM2.5 concentration across all cities. In C-R analyses consisting of 107 U.S.
cities (Figure 3-17a) and 25 Canadian cities (Figure 3-17b). analyses indicated a linear, no-threshold
relationship at concentrations often experienced within the U.S. and Canada, with less certainty in the
shape of the curve at concentrations less than approximately 8 (ig/m3 and greater than 30 (ig/m3.
0
1
United States
B
0 10 20 30 *0 50
PMj L Conoenifationi inaW)
Canada
5i o -
m
I
J
4
*
/
rfr
^ m. ¦
/
1 ^
J*
^ ¦¦
II f
u
\I —
v F
t
*
(F
*
Jf
**¦
V
\
\
\
J
1
10 2D W *0
PMj j Ccncentratans
Source: Adapted from Liu et al. (20191. copyright permission pending.
Figure 3-17
Concentration-response curves for the U.S. (A) and Canada (B)
using a B-spline function with knots at the 25th and 75th
percentiles of PM2.5 concentrations across all cities in each
location.
While Liu et al. (2019) focused on nonaccidental mortality, Lavigne et al. (2018) examined the
C-R relationship for nonaccidental mortality as well as cardiovascular- and respiratory-related mortality
in an analysis of 24 Canadian cities. The authors used the same model for each mortality outcome,
consisting of natural cubic splines with 3 df focusing on 0-2-day PM2 5 exposures. Across mortality
outcomes, C-R curves support a linear relationship at PM2.5 concentrations often experienced in the U.S.
September 2021
3-65
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
and Canada, with less certainty in the shape of the curve for nonaccidental mortality at concentrations
below approximately 5 (ig/m3 and some evidence of nonlinearity in the respiratory mortality C-R
relationship occurring around 7 (ig/m3 (Figure 3-18). However, compared to nonaccidental and
cardiovascular mortality, confidence intervals were much wider for respiratory mortality resulting in less
confidence in the overall shape of the C-R curve.
Source: Copyright permission pending from Laviane et al. (20181.
Figure 3-18 Concentration-response curves for nonaccidental,
cardiovascular, and respiratory mortality using natural cubic
splines with 3 degrees of freedom for associations with 0-2-day
PM2.5 across 24 Canadian cities.
Overall, recent studies, although limited in number continue to provide evidence of a linear,
no-threshold relationship between short-term PM2.5 exposure and mortality. Additionally, analyses of
nonaccidental mortality support previous studies evaluated that indicated confidence in the shape of the
C-R relationship down to concentrations in the range of 5-8 (ig/m3. However, consistent with studies
evaluated in previous assessments, neither study conducted systematic evaluations of alternatives to
linearity.
September 2021
3-66
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
3.2.1.3 Recent Studies Examining the PIVh.s-Mortality Relationship through
Accountability Analyses and Causal Modeling Methods
Within the 2019 PM ISA, in assessing the relationship between short-term PM2 5 exposure and
mortality several studies were evaluated that employed causal modeling methods (2019 PM ISA,
Section 11.1.2.1). These studies, which were limited to single-city analyses and used different statistical
approaches provided evidence that further confirmed the consistent positive association between
short-term PM2 5 exposure and mortality reported in numerous multi-city studies and further supported the
conclusion of a causal relationship. Since the literature cutoff date of the 2019 PM ISA, additional
epidemiologic studies have been identified that implemented causal modeling methods, which further
inform the relationship between short-term PM2 5 exposure and mortality (TableA-6).
Causal modeling methods seek to mimic randomized experiments through the use of study design
and statistical methods, which reduce the potential bias of effects due to confounding. One such method,
generalized propensity score (GPS), estimates the conditional probability of an individual being exposed
to the observed ambient concentration, accounting for all measured potential confounders. In other words,
the probability of being exposed is the same as the probability of being unexposed and the exposure can
be considered "random" (Wei et al.. 2020). To assess the associations between short-term PM2 5 exposure
and mortality, recent studies by Wei et al. (2020) and Wei et al. (2021b) used different GPS approaches.
Wei et al. (2020) evaluated the association between short-term PM2 5 exposure and all-cause
mortality among Medicare beneficiaries residing in Massachusetts during 2000-2012. In the design stage,
to construct the GPS, an ordinary least squares model was used to regress PM2 5 against a linear
combination of covariates including the copollutants ozone and NO2 (Wei et al.. 2020). In the analysis
stage, an ordinary least squares regression was used to fit a linear probability model relating the outcome
(death) with the observed exposures and the estimated GPS. Wei et al. (2020) reported 3.04 (95% CI:
2.17, 3.94) excess deaths per 10 million person-days for each 1 (.ig/ni3 increase in short-term PM2 5
exposure. When the analysis was restricted to a range of PM2 5 concentrations, the number of excess
deaths associated with a 1 (.ig/ni3 increase in short-term PM2 5 exposure increased from 3.33 (PM2 5
concentrations <35 ug/m3; 95% CI: 2.41, 4.11) to 14.56 (PM25 concentrations <5 ug/m3; 95% CI: 3.96,
24.59), per 10 million person-days.
In a subsequent study, Wei et al. (2021b) used three GPS approaches (linear probability model,
weighted least squares, and m-out-of-n random forests [moonRF]), for assessing additive effects of
short-term exposures to PM2 5 and the copollutants O3, and NO2 on mortality rates among Medicare
beneficiaries residing in Massachusetts between 2000 and 2012. To reduce the computational burden of
the linear probability model GPS approach, weighted least squares and moonRF GPS approaches were
proposed as alternatives. Consistent with Wei et al. (2020). for the linear probability model, the authors
had both a design stage and an analysis stage. In the design stage, the GPS for PM2 5 concentrations was
constructed by fitting a linear regression of the predicted PM2 5 concentration against a column vector of
September 2021
3-67
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
covariates. In the analysis stage, a linear probability model was fitted with the outcome of death, against
the predicted PM2 5 concentration and the GPS.
For the weighted least squares method, there was also a design stage and an analysis stage. In the
design stage, the person-days that had the same sex, race, age, Medicaid eligibility, ZIP code of residence,
and date were aggregated as a single record and assigned the numbers of person-days for that record as
weight. The GPS was constructed by fitting a weighted linear regression of the predicted PM2 5
concentration against all the covariates from this aggregated data set, with continuous covariates modeled
with cubic polynomials. The average outcome for each aggregated person-day group was calculated and
assigned to the person-day in the aggregated data set. A weighted linear regression was then fitted for the
averaged outcome against the predicted PM2 5 concentration and the estimated GPS.
The moonRF method is based on the random forest method, which is a non-parametric machine
learning approach of classification for possible non-linear relationships and interactions through building
individual decision trees through resampling. The m-out-of-n bootstrapping method resamples the m
observations out of an original data set (1,... ,n) without replacement, where m « n (Wei et al.. 2021b).
In the design stage, the number of person-days aggregated for each record in the aggregated data set was
used as the frequency weight and sampled 62,000 person-days without replacement. With this sample,
trees were built for PM2 5 to make predictions of the exposure for each person-day in the aggregated data
set, which was repeated 100 times. The final predicted PM2 5 concentration for each person-day was
obtained by averaging the predictions of the 100 trees. The GPS was constructed by using the averaged
predictions of the 100 trees as the predicted PM2 5 concentration and covariates for each person-day in the
aggregated data set. In the analysis stage, the authors fit a weighted regression of the averaged outcome
against the predicted PM2 5 concentration and the estimated GPS using the aggregated data set to obtain
the estimate for the additive effect of short-term PM2 5 exposure on mortality rate.
Wei et al. (202 lb) reported that the linear probability model and the weighted least squares model
produced identical results with the estimated annual number of early deaths associated with a 1 (.ig/ni1
increase in 24-hour avg PM2 5 concentrations being 92 (95% CI: 67, 117). However, the moonRF
approach estimated a smaller number of annual deaths 69 (95% CI: 44, 95). When restricting the analysis
to person-days with 24-hour average PM2 5 concentrations below <35 |ig/m3. the authors estimated a
larger annual number of early deaths for each method (101 [95% CI: 74, 127] forthe linear probability
model and weighted least squares approaches and 78 [95% CI: 52, 105] forthe moonRF approach).
In another recent causal modeling study, using data from the National Center for Health Statistics
in 135 U.S. cities, Schwartz et al. (2018a) utilized three causal methods: instrumental variable analysis, a
negative exposure control, and marginal structural models to estimate the association between local
pollution, including PM2 5, and daily deaths. Instrumental variable analysis constructs a single or set of
instrument variables that represent variations in the exposure that are randomized with respect to both
measured and unmeasured confounders. The instrument variables considered were planetary boundary
layer, wind speed, and sea level pressure. Negative exposure control identifies a negative exposure
September 2021
3-68
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
variable, which is likely to be correlated with unmeasured potential confounders but could not be a cause
of the outcome of interest. Negative exposure controls serve as instruments for the unmeasured
confounders. If such confounders exist, control for the negative exposure would be expected to reduce or
eliminate the estimated effect of the exposure of interest. Marginal structural models estimate the
marginal effects of exposure by using inverse probability weights of time-varying exposures to render the
exposure independent of the measured covariates. If the exposure is independent of covariates, its effect
on the outcome cannot be confounded by them and resulting estimates do not depend on the distributions
of confounders. The instrumental variable approach estimated that mortality increased by 1.54% (95% CI:
1.12, 1.97) at lag 0-1 for a 10 |ig/m3 increase in 24-hour avg PM2 5 concentrations. When restricted to
days with 24-hour average PM2 5 concentrations below 25 |ig/nr\ the instrument for PM2 5 was associated
with an increase of 1.70% (95% CI: 1.11, 2.29). With the negative control exposure method, there was
-0:1% (95% CI: -0.5, 0.3) change in mortality. For the marginal structural models, there was an
estimated 0.75% (95% CI: 0.35, 1.15) increase in mortality. When restricted to days with 24-hour average
PM2 5 concentrations below 25 |ig/nr\ the marginal structural model also reported a 0.83% (95% CI: 0.39,
1.27) increase in mortality, albeit smaller in magnitude than the instrumental variable approach. Overall,
the results of Schwartz et al. (2018a) continue to support a relationship between short-term PM2 5
exposure and mortality.
Recent epidemiologic studies that employed causal modeling methods to examine the association
between short-term PM2 5 exposure and mortality reported consistent positive associations within large
cohorts across multiple cities in the U.S. Furthermore, the use of causal modeling methods within these
studies aims to reduce the uncertainties related to potential confounders that may bias reported
associations. Overall, these recent studies further support the conclusions of the 2019 PM ISA with
respect to short-term PM2 5 exposure and mortality.
3.2.1.4 Summary of Recent Evidence in the Context of the 2019 Integrated
Science Assessment for Particulate Matter Causality Determination
for Short-Term PM2.5 Exposure and Mortality
The few multicity epidemiologic studies conducted since the literature cutoff date of the 2019 PM
ISA, provide additional support to the evidence base that contributed to the conclusion of a causal
relationship between short-term PM2 5 exposure and mortality. Recent U.S. and Canadian studies in
combination with previously evaluated multicity studies provide evidence of consistent positive
associations with all cause and nonaccidental mortality, primarily within the first few days after exposure
(i.e., lag 0 and 1 day), across studies conducted in different geographic locations and in populations with
different demographic characteristics. Additionally, these positive associations persist across studies that
used different statistical models, exposure assessment approaches, and methods for confounder control.
Overall, recent studies continue to support a relationship between short-term PM2 5 exposure and
September 2021
3-69
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
mortality at lower mean 24-hour average concentrations, generally below 12 (ig/m3, as detailed in the
2019 PM ISA.
The limited assessment of cause-specific mortality in recent studies provides similar results to
previously evaluated studies demonstrating a consistent relationship with cardiovascular mortality and
more variability in the magnitude and precision of associations with respiratory mortality. Consistent with
studies evaluated in the 2019 PM ISA, recent studies indicate that associations between short-term PM2 5
exposure and mortality are relatively unchanged in copollutant models but may be larger in magnitude in
the presence of some co-occurring pollutants (i.e., oxidant gases). In addition, factors that have been
shown to vary between cities and regions of the U.S., such as housing characteristics, have been shown to
explain some of the city-to-city and regional variability observed in PM2 5-mortality associations in
multi-city epidemiologic studies. The continued assessment of the C-R relationship between short-term
PM2 5 exposure and mortality further supports a linear, no-threshold relationship, with less confidence in
the shape at concentrations below 5 (ig/m3. Additionally, recent studies that employed causal modeling
methods provide additional support for a relationship between short-term PM2 5 exposure and mortality.
3.2.2 Long-Term PM2.5 Exposure
The following sections represent a summary of the evidence and the corresponding causality
determination for long-term PM2 5 exposure and mortality presented within the 2019 PM ISA
(Section 3.2.2.1) along with an evaluation of recent epidemiologic studies that fall within the scope of the
Supplement (i.e., studies conducted in the U.S. and Canada) and were published since the literature cutoff
date of the 2019 PM ISA (Section 3.2.2.2V In addition, with the expansion of in epidemiologic studies
using causal modeling methods, recent studies that employed such methods are also evaluated
(Section 3.2.2.3). which can further inform the relationship between long-term PM2 5 exposure and
mortality. Lastly, a summary of the results of recent studies evaluated within the section is presented in
the context of the scientific conclusions detailed in the 2019 PM ISA (Section 3.2.2.4V The evaluation of
recent studies on long-term PM2 5 exposure and mortality presented in this Supplement adds to the
collective body of evidence reviewed in the process of reconsidering the PM NAAQS.
3.2.2.1 Summary and Causality Determination from 2019 Integrated
Science Assessment for Particulate Matter
Cohort studies evaluated in the 2019 PM ISA provided consistent evidence of positive
associations between long-term PM2 5 exposures and total (nonaccidental) mortality from studies
conducted mainly in North America and Europe. Many analyses further evaluated the association between
long-term PM2 5 exposures and the risk of mortality based on the original American Cancer Society
(ACS) study (Pope et al.. 1995). added new details about deaths due to cardiovascular disease (including
September 2021
3-70
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
IHD) and respiratory disease (including COPD), and extended the follow-up period of the ACS to
22 years (1982-2004). Adding to this evidence, U.S. and Canadian cohort studies demonstrated
consistent, positive associations between long-term PM2 5 exposure and mortality across various spatial
extents, exposure assessment metrics, and statistical techniques, and locations, where mean annual
average concentrations are <12 (ig/m3 (2019 PM ISA, Section 11.2.2.2). Additionally, the evidence from
these studies reduced uncertainties related to potential copollutant confounding (2019 PM ISA,
Section 11.2.3) and continued to provide strong support for a linear, no-threshold C-R relationship (2019
PM ISA, Section 11.2.4). The body of evidence for total mortality was supported by generally consistent,
positive associations with cardiovascular and respiratory mortality. There was coherence of effects across
the scientific disciplines (i.e., animal toxicological, controlled human exposure, and epidemiologic
studies) and biological plausibility for PM2 5-related cardiovascular (2019 PM ISA, Chapter 6),
respiratory (2019 PM ISA, Chapter 5), and metabolic (2019 PM ISA, Chapter 7) disease, which supports
the PM2 5-mortality relationship. This section describes the evaluation of evidence included in the 2019
PM ISA for total (nonaccidental) mortality, with respect to the causality determination for long-term
exposures to PM2 5 using the framework described in Table II of the Preamble to the ISAs (U.S. EPA.
2015). The key evidence, as it relates to the causal framework, is summarized in Table 3-5.
The strongest evidence supporting the conclusion of a causal relationship between long-term
PM2 5 exposure and total mortality in the 2009 PM ISA was derived from analyses of the ACS and
Harvard Six Cities (HSC) cohorts. Extended analyses and reanalysis of these cohorts included in the 2019
PM ISA continued to support this relationship, demonstrating consistent positive associations for total
(nonaccidental mortality) and across different cause-specific mortality outcomes. A series of analyses of
the Medicare cohort of U.S. individuals provided additional support, culminating with the largest cohort
study of nearly 61 million U.S. Medicare enrollees that reported positive associations with increases in
PM2 5 concentrations and stronger associations in areas where the mean annual PM2 5 concentrations were
<12 (ig/m3 (Di et al.. 2017b). Another series of studies conducted in Canada provided results consistent
with those of the Medicare cohort (i.e., positive associations between long-term PM2 5 exposure and total
mortality in areas where mean annual PM2 5 concentrations are <12 (.ig/ni'). One difference between these
studies was that the Canadian cohorts include all adults (age 25+ years) and the Medicare cohort only
included adults age 65+ years, demonstrating that these effects are not specific to one lifestage, but affect
all adults. Also, an additional line of evidence was available that includes results from a number of
cohorts that recruited subjects based on their place of employment, including female nurses, female
teachers, male health professionals, and male truck drivers, which show consistent, positive associations
between long-term PM2 5 exposure and total mortality.
September 2021
3-71
DRAFT: Do Not Cite or Quote
-------
Table 3-5 Summary of evidence for a causal relationship between long-term
PM2.5 exposure and total mortality from the 2019 Integrated Science
Assessment for Particulate Matter.
Rationale for
Causality
Determination3
Key Evidence13
Key References and
Sections in the 2019
PM ISAb
PM2.5
Concentrations
Associated
with Effects
(Hg/m3)c
Consistent
epidemiologic
evidence from
multiple, high-quality
studies at relevant
PM2.5 concentrations
Positive associations between long-term PM2.5 Section 11.2.2.1 Mean across
exposure and mortality in the multiple analyses studies:
of the ACS and HSC cohorts, with effect 11.4-23.6
estimates similar in magnitude, even after
adjustment for common potential confounders.
Positive associations between long-term PM2.5 Section 11.2.2.2 Mean across
exposure and mortality in the multiple analyses studies:
of the Medicare cohort, with effect estimates 8.12-12.0
similar in magnitude, even after adjustment for
common potential confounders.
Positive associations between long-term PM2.5
exposure and mortality in the multiple analyses
of Canadian cohorts, with effect estimates
similar in magnitude, even after adjustment for
common potential confounders.
Section 11.2.2.2 Mean across
studies: 8.7-9.1
Positive associations between long-term PM2.5 Section 11.2.2.2 Mean across
exposure and mortality in the multiple North studies:
American occupational cohorts, even after 12.7-17.0
adjustment for common potential confounders.
Positive associations with cardiovascular, Section 6.3.10.1 Mean across
respiratory, and lung cancer mortality. studies:
4.1-17.9
Section 5.2.10
Mean across
studies:
4.1-17.9
Section 10.2.5.1 Mean across
studies:
6.1-33.7
Epidemiologic
evidence from
copollutant models
provides some
support for an
independent PM2.5
association
Positive associations observed between Section 11.2.3;
long-term PM2.5 exposure and total mortality Figure 11-20;
remain relatively unchanged after adjustment Figure 11-21
for O3, NO2, and PM10-2.5.
When reported, correlations with copollutants
were highly variable (low to high).
September 2021
3-72
DRAFT: Do Not Cite or Quote
-------
Table 3-5 (Continued): Summary of evidence for a causal relationship between
long-term PM2.5 exposure and total mortality.
Rationale for
Causality
Determination3
Key Evidence13
Key References and
Sections in the 2019
PM ISAb
PM2.5
Concentrations
Associated
with Effects
(|jg/m3)c
Consistent positive
epidemiologic
evidence for
associations between
PM2.5 exposure and
total mortality across
exposure
measurement metrics
Positive associations consistently observed
across studies that used fixed-site
(i.e., monitors), model (e.g., CMAQ, dispersion
models), and satellite-based (e.g., AOD
observations from satellites) methods, including
hybrid methods that combine two or more of
these methods.
Section 11.2.2.5;
Jerrett et al. (2016)
Epidemiologic No evidence for deviation from linearity in Section 11.2.2.3
evidence supports a several U.S. and Canadian cohorts,
linear, no-threshold
C-R relationship
Biological plausibility
from studies of
cardiovascular and
respiratory morbidity
and lung cancer
incidence and
mortality
Cardiovascular morbidity studies provide Section 6.3 Mean across
expanded body of evidence for associations Miller et al (2007) studies:
between long-term PM25 exposure and CHD, nr\tR\ 10.7-13.4
stroke, and atherosclerosis, providing biological —' e a ' ' '
plausibility for a relationship between long-term
PM2.5 exposure and cardiovascular mortality.
Respiratory morbidity studies provide some Section 5.2.5
evidence for an association between long-term
PM2.5 exposure and development of COPD,
providing limited biological plausibility for a
relationship between long-term PM2.5 exposure
and respiratory mortality.
Consistent epidemiologic evidence for Section 10.2.5.1 Mean across
associations between PM2.5 exposure and lung Figure 10-3 U.S. and
cancer incidence and mortality in cohort studies Canadian
conducted in the U.S., Canada, Europe, and studies:
Asia. 6.3-23.6
ACS = American Cancer Society; AOD = aerosol optical depth; CHD = coronary heart disease; CMAQ = Community Multiscale Air
Quality; COPD = chronic obstructive pulmonary disease; C-R = concentration-response; HSC = Harvard Six Cities;
|jg/m3 = microgram per cubic meter; N02 = nitrogen dioxide; 03 = ozone; PM25 = particulate matter with a nominal mean
aerodynamic diameter less than or equal to 2.5 |jm; PM10-25 = particulate matter with a nominal mean aerodynamic diameter
greater than 2.5 |jm and less than or equal to 10 |jm.
aBased on aspects considered in judgments of causality and weight of evidence in causal framework in Table I and Table II of the
Preamble.
bDescribes the key evidence and references contributing most heavily to the causality determination and, where applicable, to
uncertainties or inconsistencies. References to earlier sections indicate where the full body of evidence is described in the 2019
PM ISA.
°Describes the PM25 concentrations with which the evidence is substantiated.
September 2021 3-73 DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Evidence included in the 2019 PM ISA helped to reduce uncertainties related to potential
copollutant confounding of the relationship between long-term PM2 5 exposure and mortality. Multiple
studies evaluated ozone (2019 PM ISA, Figure 11-20) and NO2 (2019 PM ISA, Figure 11-21) in
copollutant models and observed similar hazard ratios for PM2 5 regardless of whether ozone or NO2 were
included in the model. This supports an independent effect of long-term PM25 exposure on mortality.
Evidence for other potential copollutants (e.g., SO2, CO) was limited.
Studies evaluated in the 2019 PM ISA used a variety of both fixed-site (i.e., monitors), model
(e.g., CMAQ, dispersion models), and satellite-based (e.g., aerosol optical depth [AOD] measurements
from satellites) methods, including hybrid methods that combine two or more fixed-site, model, and/or
satellite-based techniques to measure, estimate, or predict PM2 5 concentrations for use in assigning
long-term PM2 5 exposure in epidemiologic studies. Overall, the exposure assessment technique had little
influence on study results, with consistently positive associations of similar magnitude observed across
studies using a variety of exposure assessment techniques. Notably, Jerrett et al. (2016) applied fixed-site
measurements and satellite-based observations of AOD to a common data set, the ACS cohort, and
calculated effect estimates for circulatory and IHD mortality associated with PM2 5 using both methods.
They observed consistently positive associations between long-term PM2 5 exposure and mortality,
regardless of the exposure assessment technique used to assign exposure. Additionally, Jerrett et al.
(2016) combined multiple exposure assessment techniques into an ensemble model, weighted by model
fit, and continued to observe similar positive associations with mortality. These results support an
independent effect of long-term PM2 5 exposure on mortality that is not overtly influenced by or a residual
of the exposure assessment technique used in the study.
The number of studies that examined the shape of the C-R function for long-term PM2 5 exposure
and mortality substantially increased between the 2009 PM ISA and the 2019 PM ISA. These studies used
a number of different statistical techniques to evaluate the shape of the C-R function, including natural
cubic splines, restricted cubic splines, penalized splines, thin-plate splines, and cutpoint analyses (2019
PM ISA, Table 11-7), and generally observed linear, no-threshold relationships down to 4-8 (ig/m3. Few
studies have conducted extensive analyses exploring alternatives to linearity when examining the shape of
the PM2 5-mortality C-R relationship. Among these studies, there was some emerging evidence for a
supralinear C-R function, with steeper slopes observed at lower PM2 5 concentrations. Although few, such
supralinear C-R functions were most commonly observed for cardiovascular mortality compared to total
(nonaccidental) or respiratory mortality.
The 2009 PM ISA concluded that there is not sufficient evidence to differentiate the components
or sources more closely related to health outcomes when compared with PM2 5 mass, although the
evidence for long-term exposure and mortality was limited. Several studies included in the 2019 PM ISA
examined the relationship between long-term exposure to PM components and mortality (2019 PM ISA,
Figure 11-24). Collectively, these studies continued to demonstrate that no individual PM2 5 component or
source was a better predictor of mortality than PM2 5 mass.
September 2021
3-74
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Overall, epidemiologic studies examined in the 2019 PM ISA built upon and further reaffirmed
the conclusions of the 2009 PM ISA for total mortality. The evidence, particularly from the assessment of
PM2 5-related cardiovascular and metabolic diseases, with more limited evidence from respiratory
morbidity, provided biological plausibility for mortality due to long-term PM2 5 exposures. In conclusion,
the consistent positive associations observed across cohort studies conducted in various locations across
North America were further supported by the results from copollutant analyses indicating robust
associations independent of O3 and NO2. Collectively, this body of evidence was sufficient to conclude
that a causal relationship exists between long-term PM2.5 exposure and total mortality.
3.2.2.2 Recent U.S. and Canadian Cohort Studies
In addition to evaluating epidemiologic studies that examined the relationship between long-term
PM2 5 exposure and mortality, the 2019 PM ISA characterized whether evidence supported biologically
plausible mechanisms by which long-term PM2 5 exposure could lead to mortality. This evaluation
consisted of an assessment of animal toxicological, controlled human exposure, and epidemiologic studies
of morbidity effects that are the largest contributors to total (nonaccidental) mortality, specifically,
cardiovascular and respiratory morbidity and metabolic disease (2019 PM ISA, Section 6.2.1,
Section 5.2.1, and Section 7.2.1, respectively). Plausible mechanisms were identified by which inhalation
exposure to PM2 5 could progress from initial events to endpoints relevant to the cardiovascular system
and to population outcomes such as IHD, stroke and atherosclerosis (2019 PM ISA, Section 6.2.1).
Similarly, available evidence was characterized by which inhalation exposure to PM2 5 could progress
from initial events to endpoints relevant to the respiratory system and to population outcomes such as
exacerbation of COPD (2019 PM ISA, Section 5.2.1). In addition, there was evidence for plausible
mechanisms by which inhalation exposure to PM25 could progress from initial events (e.g., pulmonary
inflammation, autonomic nervous system activation) to intermediate endpoints (e.g., insulin resistance,
increased blood glucose and lipids) and result in population outcomes such as metabolic disease and
diabetes. Collectively, the progression demonstrated in the available evidence for cardiovascular and
respiratory morbidity and metabolic disease supported potential biological pathways by which long-term
PM2 5 exposures could result in mortality (2019 PM ISA, Section 11.2).
Recent cohort studies conducted in the U.S. and Canada build upon the strong evidence base
evaluated in the 2019 PM ISA, as well as previous assessments, that provided the scientific rationale
supporting a causal relationship between long-term PM2 5 exposure and mortality (Section 3.2.2.1). In
addition to examining the relationship between long-term PM2 5 exposure and all-cause or nonaccidental
mortality (Section 3.2.2.2.1) and cause-specific mortality (Section 3.2.2.2.2). some studies also further
examined issues relevant to expanding the overall understanding of the effect of long-term PM2 5
exposure on mortality. Specifically, recent studies have assessed the effect of long-term PM2 5 exposure
on mortality in populations with underlying health conditions (Section 3.2.2.2.3). examined the role of
long-term PM2 5 exposure on life expectancy (Section 3.2.2.2.4). examined potential copollutant
September 2021
3-75
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
confounding (Section 3.2.2.2.5). explored new and innovative methods for assessing confounding
(Section 3.2.2.2.6). and assessed the shape of the concentration-response (C-R) relationship
(Section 3.2.2.2.1). The following sections evaluate recent cohort studies that inform each of the
aforementioned topics within the context of the evidence base evaluated and summarized in the 2019 PM
ISA.
3.2.2.2.1 All Cause and Total (Nonaccidental) Mortality
Recent North American cohort studies that examined the relationship between long-term PM2 5
exposure and mortality support and expand upon the cohort studies evaluated in the 2019 PM ISA that
spanned diverse geographical areas and study populations. These recent studies build upon the studies
evaluated in the 2019 PM ISA that addressed key uncertainties identified in the 2009 PM ISA
(e.g., PM2 5-mortality associations at low concentrations, shape of the concentration-response [C-R]
relationship).
Consistent with the North American cohort studies evaluated in the 2019 PM ISA (Figure 3-19).
recent studies representing cohorts in the U.S. and Canada report a similar pattern of associations between
long-term PM2 5 exposure and all-cause or nonaccidental mortality in locations with generally low mean
annual PM2 5 concentrations (Figure 3-20). Whereas the use of hybrid exposure models, which includes
some combination of monitoring, modeled, and satellite data, represented an advancement in methods in
the 2019 PM ISA, recent studies primarily relied on these exposure assessment methods demonstrating
the growth in their application and utility. Study-specific information for each of the recent studies
evaluated in this section including information on the cohort, exposure assessment methodology, and
PM2 5 concentrations over the study duration are detailed in Table A-7.
September 2021
3-76
DRAFT: Do Not Cite or Quote
-------
Reference
tPopeet al. (2014)
tLepeule etal. (2012)
t Thurston etal. (2015)
Zeger et al. (2008)
Zeeer et al. (2008)
Zeeer et al. (2008)
Eftlm et al. (2008)
tDi et al. (2017)
tDi et al. (2017)
tDi et al. (2017)
tKioumourtzoslou et al. (2016)
fPun etal. (2017)
tShi et al. (2015)
tShi et al. (2015)
tShi et al. (2015)
tShi et al. (2015)
I Wang et al. (2017)
fWang etal. (2017)
Lipfert et al. (2006)
Goss et al. (2004)
tCrouse et al. (2012)
tCrouse et al. (2012)
tClien et al. (2016)
tCrouse et al. (2015)
tWeichenthal etal. (2014)
fWeichenthal etal. (2014)
tPinault et al. (2016)
tLipsett et al. (2011)
tOstroetal. (2010)
tOstroetal. (2010)
tOstroetal. (2015)
tPuett et al. (2009)
tHartet al. (2015)
tHart et al. (2015)
tPuett etal. (2011)
tHartet al. (2011)
tKloog et al. (2013)
tGarcia et al. (2015)
tGarcia et al. (2015)
tGarcia et al. (2015)
tWang etal. (2016)
Enstrom (2005)
Enstrom (2005)
Enstrom (2005)
Cohort
ACS
Harvard Six Cities
NIH-AARP
MCAPS
MCAPS
MCAPS
ACS-Medicare
Medicare
Medicare
Medicare
Medicare
Medicare
Medicare
Medicare
Medicare
Medicare
Medicare
Medicare
Veterans Cohort
U.S. Cystic Fibrosis
CanCHEC
CanCHEC
EFFECT
CanCHEC
Ag Health
As Health
CCHS
CA Teachers
CA Teachers
CA Teachers
CA Teachers
Nurses Health
Nurses Health
Nurses Health
Health Prof
TrIPS
MA cohort
CA cohort
CA cohort
CA cohort
NJ Cohort
CA Cancer Prev
CA Cancer Prev
CA Cancer Prev
Notes
Eastern
Western
Central
exp<12 ^gm3
nearest monitor
mutual adj
exp <10 |ig, m3; mutual adj
no mutual adj
exp <10 ^ig m3. no mutual adj
exp<12 ng m3
satellite data
monitor data
more precise exp
within 30 km
within 8 km
nearest monitor
spatio-temp. model
foil model
CVD+Resp
krieins
IDW
closest monitor
Follow-up
19S2-2004
1974-2009
2000-2009
2000-2005
2000-2005
2000-2005
2000-2002
2000-2012
2000-2012
2000-2012
2000-2010
2000-2008
2003-2008
2003-2008
2003-200S
2003-2008
2000-2013
2000-2013
1997-2001
1999-2000
1991-2001
1991-2001
1999-2011
1991-2006
1993-2009
1993-2009
2000-2011
2000-2005
2002-2007
2002-2007
2001-2007
1992-2002
2000-2006
2000-2006
1989-2003
1985-2000
2000-2008
2006
2006
2006
2004-2009
1973-1982
1983-2002
1973-2002
Mean (IQR)
12.6
11.4-23.6
10.2-13.6
14.0 (3.0)
13.1(8.1)
10.7 (2.4)
13.6
11.5
11.5
11.5
12
12.5
8.12(3.78)
8.12(3.78)
8.12(3.78)
8.12(3.78)
10.7 (3.8)
10.7 (3.8)
14.34
13.7
8.9
11.2
10.7
8.9
9.44
9.44
6.3
15.6 (8.0)
17.5(6.1)
17(6.1)
17.9(9.6)
13.9(3.6)
12.7
12
17.8(4.3) -
14.1 (4)
9.9(1.6)
13.06
12.94
12.68
11.3
23.4
23.4
23.4
~
~
OS
1.0 1.2 1.4 16
Hazard Ratio (95% Confidence Interval)
Source; Figure 11-18, 2019 PM ISA.
ACS = American Cancer Society; adj = adjustment; Ag Health = Agricultural Health Study; Cancer Prev = Cancer Prevention;
CanCHEC = Canadian Census Health and Environment Cohort; CCHS = Canadian Community Health Survey;
CVD = cardiovascular; EFFECT = Enhanced Feedback For Effective Cardiac Treatment; exp = exposure; Health Prof = health
professionals; IDW = inverse distance weighting; IQR = interquartile range; km = kilometer; pg/m3 = microgram per cubic meter;
MCAPS = Medicare Cohort Air Pollution Study; NIH-AARP = National Institutes of Health—American Association of Retired
Persons; PM =- particulate matter; PM25 = particulate matter with a nominal mean aerodynamic diameter less than or equal to
2.5 pm; Resp = respiratory; TrIPS = Trucking Industry Particle Study.
Note: tStudies published since the 2009 PM ISA. Associations are presented per 5 jjg/m3 increase in pollutant concentration.
Circles represent point estimates; horizontal lines represent 95% confidence intervals for PM25. Black text and circles represent
evidence included in the 2009 PM ISA; red text and circles represent recent evidence not considered in previous ISAs or AQCDs.
Study results from Pope et al. (20141 are representative of the results from the American Cancer Society Cohort.
Figure 3-19 Associations between long-term PM2.5 exposure and total
(nonaccidental) mortality in recent North American cohorts.
September 2021
3-77
DRAFT; Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
Study
Cohort
Follow-up
Mean
Exposure
*Eum etal. (2018)
Medicare
2000 - 2012
11.65
1-year avg - base model
•
1-year avg - residual model
•
*Pope et al. (2019)
NHIS - Full Cohort
19S6 - 2015
10.7
17-year avg
#-
NHIS - Sub-Cohort
1986-2015
10.7
17-year avg
-~
*Lefler etal. (2019)
NHIS - Sub-Cohort
19S7-2015
10.67
28-year avg
•-
10.67
17-year avg
10.67
5-year avg
-•
10.67
2-year avg
~
*Wuetal. (2020)
Medicare
2000-2016
9.8
1-year avg - Cox
•
~Wuetal. (2020)
Medicare
2000 - 2016
9.8
1-year avg - Poisson
•
*Effiot etal. (2020)
NHS
19SS-200S
13.7
2-year avg
*Wang etal. (2020)
Medicare
2000-200S
10.55
1-year avg
•
*Pappin et al. (2019)
CanCHEC
1991-2016
6.68 - 7.95
3-year avg, 1-year lag
•
*Zhang etal. (2021)
Ontario Health Study
2009 - 2017
7.8
5-year avg, 1-year lag
*Pinault et al. (2017)
CanCHEC
1991 -2011
7.4
3-year avg, 1-year lag
~
*Crouse et al. (2020)
CanCHEC
2001-2011
7.98
8-year avg, 1-year lag; 1 km
~
7.27
S-year avg, 1-year lag; 5 km
~-
6.9
8-year avg, 1-year lag; 10 km
~
7.43
3-year avg, 1-year lag; 1km
~
6.79
3-year avg, 1-year lag; 5 km
~-
6.44
3-year avg, 1-year lag; 10 km
7.21
1-year avg, 1-year lag; 1km
~
6.59
1-year avg, 1-year lag; 5 km
~
6_24
1-year avg, 1-year lag; 10 km
~
*Christidis etal. (2019)
mCCHS
2000 - 2012
5.9
3-year avg, 1-year lag
Nonaccidental
1.00 1.20 1.40 1.60
Hazard Ratio (95% Confidence Interval)
avg = average; CanCHEC = Canadian Census Health and Environment Cohort; mCCHS = Canadian Community Health
Survey—mortality cohort; km = kilometer; |jg/m3 = microgram per cubic meter; NHIS = National Health Interview Survey;
PM = particulate matter; PM25 = particulate matter with a nominal mean aerodynamic diameter less than or equal to 2.5 |jm.
Note: 'U.S. and Canadian studies published since the literature cutoff date (~January 2018) for the 2019 PM ISA. Associations are
presented per 5 |jg/m3 increase in pollutant concentration. Circles represent point estimates; horizontal lines represent 95%
confidence intervals for PM25. Blue text and circles represent U.S. and Canadian studies published since the literature cutoff date of
the 2019 PM ISA.
Figure 3-20 Associations between long-term PM2.5 exposure and all cause and
total (nonaccidental) mortality in cohort studies in the U.S. and
Canada published since the 2019 Integrated Science Assessment
for Particulate Matter.
In the continuous evolution of studies aimed at assessing the relationship between long-term
PM2 5 exposure and mortality, recent research efforts focus on identifying increasingly larger and more
diverse cohorts, including the National Health Interview Survey (NHIS) cohort and the Medicare cohort.
Pope et al. (2019) formed the NHIS cohort by linking participants from the NHIS from 1986-2014 with
mortality data through 2015 resulting in approximately 1.5 million participants. The authors also formed a
subcohort comprised of over 650,000 participants that had individual-level data on smoking status and
body mass index (BMI). In analyses of both the full cohort and subcohort, Pope et al. (2019) uses a
combination of monitored and modeled PM2 5 data (see Table A-7) to estimate population-weighted
17-year average PM2 5 concentrations at the census tract centroid of each participant. In fully adjusted
Cox regression models, the authors report associations with all-cause mortality similar in magnitude for
both the full cohort (HR= 1.06 [95% CI: 1.05, 1.08]) and subcohort (HR= 1.06 [95% CI: 1.04, 1.07]). In
a sensitivity analysis, back-casted PM25 concentrations were imputed from 1988-1998, allowing for the
September 2021
3-78
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
use of a 28-year average PM2 5 exposure window. In both the full cohort and subcohort, associations were
attenuated, but remained positive when using the 28-year average exposure window, in comparison to the
main analyses using the 17-year average exposure window (quantitative results not presented).
Lefleretal. (2019) expands upon the analyses initially conducted by Pope et al. (2019) by
focusing on the NHIS subcohort with a primary focus being to examine potential sensitivity of
PM2 5-mortality associations to spatially and temporally decomposed PM2 5. However, it is important to
note that the spatial and temporal decomposition approaches employed by Lefler et al. (2019) attempt to
examine the impact of distance to source and the sensitivity of exposure window on associations,
respectively, and are not analogous to the decomposition methods discussed in Section 3.1.2.2.5. where
the focus is on examining whether there is confounding due to unmeasured variables. To assess the
temporal aspect, the authors employed two approaches to examine the sensitivity of associations to
different exposure windows. In the first approach, they recreated cohorts for each year starting in 1992
that were then assigned either 2- or 5-year average PM2 5 concentrations with an overall effect across the
entire study duration estimated through a fixed effects meta-analysis. In the second approach, the authors
applied the same exposure windows used in Pope et al. (2019) (i.e., 17-year or 28-year average PM2 5
concentrations) as a comparison. Lefler et al. (2019) reported associations consistent in magnitude,
regardless of the exposure windows used (HRs: 17-year: 1.06; 28-year: 1.04; 2-year: 1.05; 5-year: 1.05).
In the spatial decomposition approach, the authors developed exposure indicators indicative of
PM2 5 emanating from different sources and defined them as local (<1 km); neighborhood (1-10 km);
mid-range (10-100 km); and regional (>100 km). In developing each exposure indicator, Lefler et al.
(2019) subtracted the minimum PM2 5 concentration identified within the defined circular buffers around
each census tract for each year from 2000-2015. While there was some variability across each of the
exposure metrics based on an IQR increase in PM2 5 concentrations, overall the regional metric was closer
in magnitude to the primary PM2 5 exposure indicator (Figure 3-21).
September 2021
3-79
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
PM2.5 Primary Est.
Spatially decomposed, alorie
PM2.5 Local
PM2.5 Neighborhood
PM2.5 Mid-range
PM2.5 Regional
Spatially decomposed, together
PM2.5 Local
PM2.5 Neighborhood
PM2.5 Mid-range
PM2.5 Regional
Source: Copyright permission pending, Lefler et al. (20191
Figure 3-21 Hazard ratios for spatially-decomposed analyses for an
interquartile range increase in PM2.5 concentrations for all cause
mortality.
Although the majority of cohort studies focus primarily on adults, generally over the age of 20, a
number of recent studies focus on only individuals 65 years of age and older that are Medicare
beneficiaries. Wang et al. (2020) used a hybrid spatiotemporal exposure model (see Table A-7) to
estimate exposures at each Medicare participant's ZIP code centroid to examine nonaccidental mortality
from 2000-2008. Using an exposure that represented the 12-month average PM2 5 concentration prior to
death, the authors reported a HR of 1.03 (95% CI: 1.02, 1.03) in a Cox proportional hazards model that
adjusted for state-level socioeconomic status (SES) to account for urbanicity and annual mean gross
adjusted income.
Recent cohort studies published since the 2019 PM ISA, focus primarily on examining the
relationship between long-term PM2 5 exposure and mortality in demographically diverse populations that
are generally representative of the entire population in both the U.S. and Canada. Elliott et al. (2020)
differs from those studies by focusing on an occupation-based cohort of female registered nurses (i.e., the
Nurses" Health Study [NHS]). Elliott et al. (2020) represents an updated and extended analysis of Hart et
al. (2015a). which was evaluated in the 2019 PM ISA. Although the focus of Elliott et al. (2020) is on
examining the interaction between long-term PM2 5 exposure and physical activity, and the risk of
cardiovascular disease and mortality, the authors used a 24-month exposure window and more years of
data (i.e., 1988-2008) in the overall analysis of the association between long-term PM2 5 exposure and
mortality, which differs from Hart et al. (2015a). which used a 12-month exposure window for the years
1.00 1.02 1.04 1.06 1.08 1.10 1.12
HR per IQR
September 2021
3-80
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
2000-2008. In the fully adjusted Cox proportional hazards model the authors reported a HR of 1.07 (95%
CI: 1.00, 1.15), which is consistent with previous studies of the NHS cohort (Figure 3-19).
While the studies discussed above add to the total body of evidence supporting a relationship
between long-term PM2 5 exposure and mortality, key questions that often arise in the assessment of the
evidence are (1) Do associations persist at low concentrations? and (2) Is there a point below which there
is less confidence in that relationship? This led to the Health Effects Institute (HEI) initiating two recent
research efforts with a main focus on examining the relationship between long-term PM2 5 exposure and
mortality at low concentrations. One of these studies conducted in the U.S., referred to as the HEI
Medicare study, focused on using a cohort of Medicare beneficiaries 65 years of age or older (Dominici et
al.. 2019). while another study conducted in Canada, referred to as the Mortality-Air Pollution
Associations in Low Exposure Environments (MAPLE) study, relied on respondents from multiple years
of the long-form Canadian Census Health & Environment Cohorts (CanCHEC) and/or participants from
multiple years of the Canadian Community Health Survey (CCHS) (Brauer et al.. 2019). A third study
was conducted in Europe, using data from the European Study of Cohorts for Air Pollution Effects
(ESCAPE) but is beyond the scope of this Supplement. Both of these research efforts conducted extensive
analyses to further inform the PM2 5-mortality relationship in a series of studies, with a focus on
examining associations at low PM2 5 concentrations, which are often considered as below the level of the
current annual PM NAAQS of 12.0 (ig/m3.
As detailed in the 2019 PM ISA (Section 11.2.2.1), the initial publication from the HEI Medicare
study applied a hybrid exposure model at a refined spatial resolution (i.e., 1 km2 grid cells) to assign
PM2 5 exposures at the ZIP code level to all Medicare beneficiaries age 65 and older in the continental
U.S. between 2000 and 2012 to examine long-term PM2 5 exposure and all-cause mortality (Di et al..
2017b). Di et al. (2017b) reported a HR of 1.041 (95% CI: 1.039, 1.042) for the relationship between
PM2 5 and all-cause mortality (Figure 3-20) with associations remaining relatively unchanged in
copollutant models with ozone estimated from the nearest monitor. The majority of recent studies that fall
under the HEI Medicare study primarily focus on causal modeling methods and are evaluated in
Section 3.1.2.3. However, a recent study by Wu et al. (2020a) builds upon the original study by Di et al.
(2017b) of the Medicare cohort, by including additional years of data up to 2016. Within this study, the
authors focused on examining associations with all-cause mortality using both traditional and causal
modeling methods (Section 3.1.2.3). and the sensitivity of associations to different confounder adjustment
(Section 3.1.2.2.5). In the main analysis, the authors reported a HR of 1.03 using both a Poisson and Cox
model that adjusted for calendar year and meteorological variables including season, maximum daily
temperature, and relative humidity. Although the regression models used by Wu et al. (2020a) include
additional covariates not included in the models used in Di et al. (2017b). similar results were reported by
both studies.
Recent studies that fall under the umbrella of the MAPLE study instituted various advancements
in exposure assessment. Similar to the HEI Medicare study (Dominici et al.. 2019). MAPLE studies rely
September 2021
3-81
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
on a hybrid exposure model that is a combination of monitored, modeled, and satellite data. Within the
MAPLE studies, as detailed in Table A-7. an exposure model was used that predicts PM2 5 concentrations
at a 1 km2 grid cell through a combination of satellite-derived PM2 5 concentration estimates, and model
predictions in GEOS-Chem that are calibrated using monitor data. The 1 km2 resolution PM2 5
concentrations are then aggregated to the postal code of residence. In the process of assigning exposures
to cohort participants, the majority of epidemiologic studies examining long-term PM2 5 exposure and
mortality tend to use exposure windows that consist of average PM2 5 concentrations for a single year or
all years that PM2 5 data is available, which does not account for potential temporal changes in PM2 5
concentrations ITJ.S. EPA (2019). Table A-71. In addition, the exposure assigned often includes the year
of death which could result in exposure misalignment if the death occurred early in the year. In an attempt
to address both issues, MAPLE studies use a primary exposure representing a 3-year average with a
1-year lag prior to death to ensure that the exposure window for each individual within the cohort occurs
prior to death (Pinault et al.. 2017). Lastly, building upon Crouse et al. (2015). which accounted for
residential mobility during the course of the study, MAPLE studies, instead of assigning exposure based
on residential address at baseline or excluding individuals from the cohort when residential location
information was missing, developed a new method where postal codes were imputed if missing to
maintain the size of the cohort.
While the MAPLE studies discussed below employ the 3-year average, 1-year lag approach to
assigning exposure, it remained unclear how the combination of temporal and spatial scale of exposure
assessment impacted the association with nonaccidental mortality. Crouse et al. (2020). within the 2001
census cycle of CanCHEC, examined whether HRs varied depending on the combination of years of
PM2 5 data used to assign exposure (i.e., 1-year avg, 3-year avg, or 8-year avg) and the spatial resolution
of the exposure model (i.e., 1, 5, or 10 km). As depicted in Figure 3-20. across each temporal and spatial
combination, associations are consistently positive, with the magnitude of the association being smallest
when using the 1-year average. Additionally, the authors showed that the magnitude of the association
declined as the spatial scale increased regardless of the temporal scale assigned. Although associations are
larger in magnitude when using the 8-year average, they are relatively similar to the 3-year average
(e.g., 1 km spatial scale: 8-year avg HR= 1.11 [95% CI: 1.10, 1.13]; 3-year avg HR= 1.10 [95% CI:
1.08, 1.11]). It is worth noting that the difference in results based on the exposure window used by Crouse
et al. (2020) are not consistent with the results of Lefler et al. (2019). discussed above, that reported
associations similar in magnitude when using shorter and longer duration exposure windows. However,
because different cohorts and exposure assessment methods are used in each study it is not clear why
these differences are observed. Overall, the results of Crouse et al. (2020) provide additional support for
the use of the 3-year average, 1-year lag approach to assigning exposure predicted at the 1 km2 spatial
scale within the MAPLE studies.
Of the MAPLE studies, Pinault et al. (2017) was the first to institute the series of exposure
assessment advancements discussed above. Pinault et al. (2017) expands upon the initial analyses of the
CanCHEC cohort conducted by Crouse et al. (2012) and Crouse et al. (2015). which focused on the 1991
September 2021
3-82
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
CanCHEC [see U.S. EPA (2019). Section 11.2.2.1; Figure 3-191. In addition to using the 2001
CanCHEC, which allowed for an updated analysis using a larger and more recent cohort, Pinault et al.
(2017) was able to address some of the limitations of the original CanCHEC studies. Specifically, Pinault
et al. (2017). used a hybrid exposure model that predicted concentrations at the 1 km2 spatial scale versus
the 10 km2 spatial scale as was done in Crouse et al. (2012) and Crouse et al. (2015). potentially allowing
for a better representation of spatial gradients in PM2 5 concentrations. In addition, Pinault et al. (2017)
used a 3-year average, 1-year lag PM2 5 exposure instead of assigning exposure based on the average of
all years of available PM2 5 data as was done for each previous CanCHEC study, and employed the new
approach to capture missing data on residential location. In a Cox proportional hazards model that was
stratified by age, sex, airshed, population center size, and included all available individual and ecological
covariates the authors reported a hazard ratio (HR) with nonaccidental mortality of 1.08 (95% CI: 1.07,
1.10).
Most of the studies examining the CanCHEC cohort, such as Pinault et al. (2017). focused on an
individual census year, but Pappin et al. (2019) examined all three census cycles (i.e., 1991, 1996, and
2001) individually and in a pooled analysis. Although, the main focus of Pappin et al. (2019) was to
examine the C-R relationship across the different census cycles (Section 3.2.2.2.7). in examining the
relationship between long-term PM2 5 exposure and mortality the authors expanded upon the regression
model used in Pinault et al. (2017) and Crouse et al. (2016). This resulted in a regression model that
included a variable to account for population size of a participant's community (Pinault et al.. 2017). a
covariate to account for airshed of residence (Crouse et al.. 2016). and a variable of neighborhood
marginalization based on the Canadian Marginalization Index (CAN-Marg). In the pooled analysis, the
authors reported aHR= 1.03 (95% CI: 1.02, 1.03) in the fully adjusted model.
In addition to the MAPLE studies focusing on the larger CanCHEC cohort, other studies either
focused solely on the CCHS-mortality (mCCHS) cohort (Christidis et al.. 2019) or relied on data from it
due to its extensive individual-level data rPinault et al. (2018) in Section 3.1.2.2.2 and Erickson et al.
(2019) in Section 3.1.2.2.41. Christidis et al. (2019) represents an extended analysis of Pinault et al.
(2016). detailed in the 2019 PM ISA (Section 11.2.2.1), which increased the year of analysis by 1 year to
2012. While both Christidis et al. (2019) and Pinault et al. (2016) relied on the same cohort, there were
fundamental differences between the two studies, including different contextual variables included in
statisical models and different criteria around the inclusion of immigrants in the cohort, that resulted in a
difference in the magnitude of the association in both studies even though both relied on a 3-year average,
1-year lag exposure [i.e., HR= 1.12 (95% CI: 1.09, 1.16) in Pinault et al. (2016) and HR = 1.05 (95% CI:
1.02, 1.09) in Christidis et al. (2019)1. Overall, the primary driver of the difference in the magnitude of
the association in both studies can be attributed to Christidis et al. (2019) including immigrants that have
resided in Canada for 10 or more years, who have substantially lower HRs of mortality compared to the
non-immigrant population. This differs from Pinault et al. (2016) where only immigrants that resided in
Canada for 20 or more years were included in the analysis. However, this difference due to years of
residence in Canada may be specific to the mCCHS cohort. For example, Erickson et al. (2020) in an
September 2021
3-83
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
analysis focusing on immigrants within the 2001 CanCHEC, reported larger risk estimates for immigrants
that resided in Canada for <10 years and 11-20 years (HRs of 1.09 and 1.11, respectively), compared to
established immigrants (>30 years residence) and non-immigrants (HR of 1.05 and 1.04, respectively).
An additional study conducted in Canada, the Ontario Health Study, that was not part of MAPLE,
but used the same exposure model and a similar exposure assignment approach, provides continued
support for a relationship between long-term PM2 5 exposure and mortality (Zhang et al.. 2021). Whereas
the CanCHEC cohort lacked information on some individual-level risk factors, the Ontario Health Study
collected individual-level data on socio-demographics, medical history, lifestyle factors, and health care
utilization from 2009-2017. In fully adjusted Cox models that included all individual-level and
contextual variables, Zhang et al. (2021) used a 5-year average exposure with a 1-year lag and reported a
HR of 1.20 (95% CI: 1.09, 1.32) for nonaccidental mortality. In a sensitivity analysis comparing the
5-year average exposure metric to alternative exposure windows of 4- and 3-year averages, the authors
reported an increase in the magnitude of the association as the number of years increased, which is
consistent with Crouse et al. (2020) discussed above.
3.2.2.2.2 Cause-Specific Mortality
Studies that examine the association between long-term PM2 5 exposure and cause-specific
mortality outcomes, including cardiovascular and respiratory mortality, can provide additional support for
PM2 5-related cardiovascular and respiratory effects, specifically whether there is evidence of an overall
continuum of effects. Some of the studies evaluated in Section 3.1.2.2.1. in addition to examining
all-cause or nonaccidental mortality, conducted analyses of cardiovascular and respiratory mortality
which builds upon the evidence detailed in the 2019 PM ISA (respiratory mortality, Section 5.2.10 and
Section 11.2.2.3; cardiovascular mortality, Section 6.2.10 and Section 11.2.2.2). As detailed in Figure
3-22. recent cohort studies provide evidence of consistent positive associations in analyses of both
cardiovascular and respiratory mortality. In addition to examining all cardiovascular and respiratory
mortality outcomes, some studies also examined individual cardiovascular and respiratory mortality
outcomes, as well as mortality in individuals with preexisting cardiovascular-related disease. The
following sections provide an overview of the evidence presented in recent cohort studies with respect to
cause-specific mortality.
September 2021
3-84
DRAFT: Do Not Cite or Quote
-------
12)
Study
lb&lW¥
tLipsettetal. (2011)
tHartet al. (2011)
tThurstonetal. (2015)
M
f Gafcia et si (2015)
*Zhans etal. (2021
•Crousfe et al. n02|
*Pinault et al. (2018)
fPinault et al. (2016)
tLipsett et al. (2011)
tHartet al. (2011)
t Thurston etal. (2015)
tTumeretal. (2016)
UtefMrn,
fCrouse et al. (2015)
•asiii
*Pinault et al. 12017)
tPinault et al. (2016)
Cohort
Ha^y^rsJ. Six^Cities
CA_7?i:h?i'i-
NIH-AARP
tWeichenthal etal. (2014)
tCrouse et al.Q015)
Chen et al. (2020)
- ¦-
\i32L
Ag Health
"'TJSaib5""15,
Medicare
CanCHEC
°°'"aM:srady
CanCHEC
CCHS
Follow-up
2006
1982-2004
3»:3H
1999-2011
MiM
1993 - 2009
Mm
1991 -20:
m
1991
"ffl6
102-13.6
3-4
All C ardiovascular
: ,V
m
J ::
u
7.4
2-33
10.2-13.6
0.5
driver
6rE
Nearest monitor
^loreggcise exposure
Monitor within 30 km
Monitor within 8 km
Excluding long-haul drivers
LUR-BME
-Njarocurce
Regional
AllRespiratoiy
1.20 1.40 1.60 1.80
Hazard Ratio (95% Confidence Internal)
Source: Update of Figure 11-19, 2019 PM ISA.
ACS = American Cancer Society; Ag Health = Agricultural Health Study; CanCHEC = Canadian Census Health and Environment
Cohort; CCHS = Canadian Community Health Survey; CVD = cardiovascular disease; EFFECT = Enhanced Feedback For Effective
Cardiac Treatment; IDW = inverse distance weighting; km = kilometer; LUR-BME = land use regression—Bayesian maximum
entropy; mCCHS = Canadian Community Health Survey—mortality cohort; |jg/m3 = microgram per cubic meter; NHIS = National
Health Interview Survey; NIH-AARP = National Institutes of Health—American Association of Retired Persons; ONPHEC = Ontario
Population Health and Environment Cohort; PM = particulate matter; PM25 = particulate matter with a nominal mean aerodynamic
diameter less than or equal to 2.5 |jm; TrIPS = Trucking Industry Particle Study; WHI = Women's Health Initiative.
Note: tStudies published since the 2009 PM ISA. *U.S. and Canadian studies published since the literature cutoff date (~January
2018) for the 2019 PM ISA. Associations are presented per 5 |jg/m3 increase in pollutant concentration. Circles represent point
estimates, closed published before 2009 PM ISA, and open published after 2019 PM ISA; horizontal lines represent 95% confidence
intervals for PM25. Black text and circles represent evidence included in the 2009 PM ISA; red text and circles represent recent
evidence not considered in previous ISAs or AQCDs; and blue text and circles represent recent U.S. and Canadian studies
published since the literature cutoff date of the 2019 PM ISA. Results from Crouse et al. (20201 are for 3-year average, 1-year lag
PM25 concentrations at 1 km resolution.
Figure 3-22 Associations between long-term PM2.5 exposure and all
cardiovascular disease and all respiratory disease mortality in
recent North American cohorts.
Cardiovascular Mortality
1 Studies investigating cardiovascular mortality provided some of the strongest evidence for a
2 cardiovascular effect related to long-term PM2 5 exposure in the 2009 PM ISA, which was further
3 supported by studies evaluated in the 2019 PM ISA (Figure 3-22). Generally, across the cohort studies
4 evaluated in the 2019 PM ISA, most of the PM2 5 effect estimates relating long-term PM2 5 exposure and
5 cardiovascular mortality remained relatively unchanged or increased in magnitude in copollutant models
6 adjusted for ozone, NO2, PM10-2.5, or SO2. The results of recent cohort studies provide additional evidence
7 for associations with cardiovascular mortality outcomes across the distribution of PM2 5 concentrations,
8 the potential implications of comorbidity on the PM2 5-cardiovascular mortality relationship, and
9 associations with individual cardiovascular mortality outcomes.
September 2021
3-85
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
In a study of the NIH-AARP cohort, which consists of participants ranging from 50-71 years of
age, Haves et al. (2020) examined not only the overall relationship between long-term PM2 5 exposure and
cardiovascular-related mortality, but whether associations changed over different ranges of PM2 5
concentrations. In the main analysis, the authors applied a spatiotemporal model that predicted PM2 5
concentrations at the census tract and reported a HR of 1.06 (95% CI: 1.03, 1.08) for all cardiovascular
mortality. This result was consistent with a sensitivity analysis that assigned exposure based on distance
to monitor from participant residence when focusing on monitors 0-25 km away (HR = 1.07 [95% CI:
1.03, 1.10]), with more variability and less precision for greater distances (i.e., 25-50 km and
50-100 km) due to the smaller sample size and the increased exposure error that can occur as distance to
monitor increases.
The results of Haves et al. (2020) are consistent with another study focusing exclusively on
cardiovascular mortality conducted by Chen et al. (2020) using the Ontario Population Health and
Environment Cohort (ONPHEC). Within ONPHEC, Chen et al. (2020) examined associations between
long-term PM2 5 exposure and cardiovascular mortality in individuals 35-85 years of age using an
exposure model similar to the one used in the MAPLE studies (Table A-7). In analyses using 1-year
average PM2 5 exposures with a 1-year lag prior to death, the authors conducted single pollutant analyses
using a traditional Cox proportional hazards model where PM2 5 is fit as a linear term and then a Cox
proportional hazards where PM2 5 could be nonlinearly associated with cardiovascular mortality using the
Shape Constrained Health Impact Function (SCHIF) detailed in Nasari et al. (2016). The authors reported
that the nonlinear model is a better predictor of cardiovascular mortality through an assessment of model
fit based on Akaike information criterion (AIC). However, relatively similar HRs were reported for both
the linear (HR= 1.11 [95% CI: 1.10, 1.12]) and nonlinear (HR = 1.10 [95% CI: 1.09, 1.12]) models,
which are consistent with the results presented in other studies of cardiovascular mortality (Figure 3-22).
In addition to conducting single-pollutant analyses, the authors introduced a new approach to assess
whether the PM2 5-cardiovascular mortality association varies depending on the proportion of PM2 5
attributed to selected components, i.e., sulphate, nitrate, ammonium, black carbon, organic matter,
mineral dust, and sea salt. In a Cox proportional hazards model that adjusted for the proportion of each of
the seven selected components Chen et al. (2020) reported that cardiovascular mortality associations
increased on average by 27% when compared to single-pollutant results across each of the five regions of
Ontario. These results provide some evidence that variability in the proportion of individual components
that comprise PM2 5, could explain regional variability in mortality risk estimates.
While there is evidence of consistent positive associations between long-term PM2 5 exposure and
cardiovascular mortality (Figure 3-22). it is plausible that comorbidities may increase the overall risk of
cardiovascular mortality. As a result, Pinault et al. (2018) examined both the CanCHEC and mCCHS
cohorts to assess whether the combination of cardiovascular disease and diabetes together yielded larger
PM2 5-mortality risk estimates than cardiovascular disease alone. Within CanCHEC, the authors identified
all participants where in addition to having cardiovascular disease as the primary cause of death, there
was also mention of diabetes on the death certificate, whereas for the mCCHS, which consisted of
September 2021
3-86
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
individual-level data, participants self-reported diabetes status by noting insulin or medication use to
manage diabetes at baseline. In both analyses, the authors applied the same base hybrid exposure model
used in the MAPLE studies, which ultimately resulted in a 3-year average PM2 5 concentration assigned to
each participant, lagged 1-year prior to death (Table A-7). In both cohorts, there was an almost doubling
of the HR for mortality for the combination of cardiovascular disease and diabetes compared to
cardiovascular disease alone (CanCHEC: CVD, HR= 1.12 [95% CI: 1.10, 1.14], CVD+diabetes,
HR= 1.23 [95% CI: 1.18, 1.28]; mCCHS: CVD, HR= 1.14 [95% CI: 1.08, 1.21], CVD+managed
diabetes, HR= 1.23 [95% CI: 1.04, 1.46]).
In addition to the studies that examined all cardiovascular mortality outcomes, a number of
studies also examined specific cardiovascular mortality outcomes (Figure 3-23). As described in
Section 6.2.10 of the 2019 PM ISA, there were generally positive associations across studies for ischemic
heart disease (IHD) mortality with a more limited assessment of other outcomes (e.g., cerebrovascular,
heart failure, hypertensive disorders). Recent studies within the Medicare cohort (Wang et al.. 2020). the
CanCHEC cohort (Crouse et al.. 2020; Cakmak et al.. 2018; Pinault et al.. 2017). and NIH-AARP cohort
(Haves et al.. 2020) report positive associations with IHD, supporting the results of studies evaluated in
the 2019 PM ISA. However, Cakmak et al. (2018) conducted a slightly different analysis of the
CanCHEC cohort than Pinault et al. (2017) by using the 1991 CanCHEC and a 7-year average exposure
instead of the 3-year average, 1-year lag exposure of the studies within MAPLE. Although Crouse et al.
(2020) also examined the 2001 CanCHEC, as noted previously, the study focused on examining different
temporal (1-, 3-, and 8-year average) and spatial (1, 5, and 10 km) scales of exposure assignment. Across
each temporal scale the authors reported consistent positive associations regardless of spatial scale, but
overall, the magnitude of the PM2 5-IHD association was smaller when using 1-year average PM2 5
concentrations. Of the studies that included an assessment of IHD mortality, only Cakmak et al. (2018)
examined copollutant models and reported that IHD associations, although attenuated, remained positive
in models with ozone (PM2 5: HR= 1.12 [95% CI: 1.10, 1.14]; PM25 + ozone: HR = 1.06 [95% CI: 1.04,
1.09]).
September 2021
3-87
DRAFT: Do Not Cite or Quote
-------
Study
fOstroeta
%3£°AW8)
tPopeet al. (2014)
IteiW
*Wane etal.,(T~ ~~
t Crouse et-1
*Crouse et
*Ptnault et al. (2( .
Chen et al. (2005
lUtV
tPopeet al. (2014)
tCrouse et al. (2012)
psettetaLGQll)
Cohort
&iu
CA
Follow-up
Monitor within 30 km
Monitor within S km
Nearest monitor
HF. cardiac arrest, related
Extended
ij-vearave; ,
Extended
Sj-yearave; ;
Hypertension
Circulatory
CBYD
1.00 1.20 1.40 1.60 1.80
Hazard Ratio (95% Confidence Interval)
Source: Update of Figure 6-19, 2019 PM ISA
ACS = American Cancer Society; AHSMOG = Adventist Health Study and Smog; AMI = acute myocardial infarction; AQCD = Air
Quality Criteria Document; CA = California; CanCHEC = Canadian Census Health and Environment Cohort;
CBVD = cerebrovascular disease; CCHS = Canadian Community Health Survey; CHD = coronary heart disease; CI = confidence
interval; CVD = cardiovascular disease; EFFECT = Enhanced Feedback for Effective Cardiac Treatment; HF = heart failure;
IDW = inverse-distance weighting; IHD = ischemic heart disease; km = kilometer; NHIS = National Health Interview Survey;
NIH-AARP = National Institutes of Health—American Association of Retired Persons; PM25 = particulate matter with a nominal
mean aerodynamic diameter less than or equal to 2.5 |jm; WHI = Women's Health Initiative.
Note: tStudies published since the 2009 PM ISA. *U.S. and Canadian studies published since the literature cutoff date (~January
2018) for the 2019 PM ISA. Associations are presented per 5-|jg/m3 increase in pollutant concentration. Circles represent point
estimates; horizontal lines represent 95% confidence intervals for PM25. Black text and circles represent evidence included in the
2009 PM ISA; red text and circles represent recent evidence not considered in previous ISAs or AQCDs, with open red circles
representing studies published since the 2019 PM ISA; and blue text and circles representing U.S. and Canadian studies published
since the literature cutoff date of the 2019 PM ISA. Study results from Pope et al. (2014) are representative of the results from the
American Cancer Society cohort.
Figure 3-23 Associations between long-term PM2.5 exposure and
cause-specific cardiovascular mortality in recent North American
cohorts.
1 Few studies to date have examined associations between long-term PM2 5 exposure and
2 cerebrovascular disease or stroke mortality, but studies by Wang et al. (2020). Pinault et al. (2017).
3 Crouse et al. (2020). and Haves et al. (2020) along with a study within the NHIS cohort by Pope et al.
4 (2019) add to the growing body of evidence indicating a positive relationship. Although Pope et al.
5 (2014) in a study of the ACS cohort, evaluated in the 2019 PM ISA indicated a positive association with
6 congestive heart failure (CHF) mortality, in a recent study of the Medicare cohort Wang et al. (2020)
7 reported a null association. Lastly, a few studies using the CanCHEC cohort (Crouse et al.. 2020; Pinault
8 et al.. 2017) examined the combination of cardiovascular mortality with either metabolic-related or
9 diabetes mortality, and found that associations were similar in magnitude to cardiovascular mortality
September 2021
3-88
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
alone [e.g., within Pinault et al. (2017) all cardiovascular: HR= 1.12 (95% CI: 1.09, 1.14); all
cardiovascular + diabetes: HR= 1.13 (95%CI: 1.10, 1.15)].
Respiratory Mortality
Evidence from studies investigating respiratory-related mortality provided limited and
inconsistent evidence for a respiratory effect related to long-term PM2 5 exposure in the 2009 PM ISA
(U.S. EPA. 2009). Studies evaluated in the 2019 PM ISA (Section 5.2.10; Figure 5-34) primarily focused
on all respiratory mortality with a more limited assessment of COPD and respiratory infection. Across
studies there was evidence of generally consistent, positive associations. Recent cohort studies provide an
additional assessment of the relationship between long-term PM2 5 exposure and COPD and respiratory
infection mortality.
Studies conducted within the NHIS (Pope et al.. 2019) and Medicare (Wang et al.. 2020) cohorts
provided limited evidence of an association with chronic lower respiratory and COPD mortality,
respectively (Figure 3-24). However, in a study of the CanCHEC cohort Pinault et al. (2017) reported a
HR for COPD mortality of 1.11 (95% CI: 1.05, 1.18), which is similar in magnitude to the association
reported in Pinault et al. (2016) within the CCHS cohort as detailed in the 2019 PM ISA. In another study
of the CanCHEC cohort, which as noted earlier used a different exposure assessment approach compared
to Pinault et al. (2017). Cakmak et al. (2018) reported evidence of a positive association between
long-term PM2 5 exposure and COPD mortality in analyses of never movers. In addition, Cakmak et al.
(2018) alone examined potential copollutant confounding of the PM2 5-COPD mortality relationship, but
in analyses focused on movers and never movers, not the entire cohort. In models with O3, in both movers
and never movers the authors reported larger HRs, albeit with wide confidence intervals, with ozone in
the model, which was substantially larger for movers (movers: HR = 1.08 [95% CI: 1.03, 1.12]; never
movers: HR= 1.05 [95% CI: 0.88, 1.26]).
September 2021
3-89
DRAFT: Do Not Cite or Quote
-------
Study
Cohort
Follow-up
Mean
Notes
tLepeule et al. 2012
Harvard Six Cities
1974 - 2009
11.4-23.6
fHart et aL 2010
TrIPS
1985 -2000
14.1
Whole cohort
Excluding long-haul drivers
tTuraer et aL 2016
ACS
1982 -2004
12.6
LUR-BME
12
Near-Source
0.5
Regional
*Pope et al. (2019)
NHIS
1999 -2015
10.7
Chronic lower respiratory
*Wang et aL (2020)
Medicare
2000 - 2008
10.55
tCrouse et al. 2015
CanCHEC
1991 -2006
8.9
*Pinault etaL (2017)
CanCHEC
1998 -2010
7.4
*Cakmak et aL (2018)
CanCHEC
1991 -2011
3.8-7.4
Movers
3.8 - 7.4
Never movers
+Pinault et al. 2016
CCHS
1998 -2011
6.3
tGan et aL 2013
Metro Vancouver
1999 - 2002
4.1
tTuraer et al- 2016
ACS
1982 -2004
12.6
LUR-BME
12
Near-Source
*Pope et aL (2019)
•Wang et aL (2020)
*Pinault etaL (2017)
NHIS
Medicare
CanCHEC
1999 -2015
2000 - 2008
1998 -2010
0.5
10.7
10.55
7.4
Regional
Influenza pneumonia
Pneumonia
Pneumonia
Respiratory Infection
1.00 1.20 1.40
Hazard Ratio (95% Confidence Interval)
Source: Modification of Figure 5-34, 2019 PM ISA.
ACS = American Cancer Society; AQCD = Air Quality Criteria Document; CanCHEC = Canadian Census Health and Environment
Cohort; CCHS = Canadian Community Health Survey; CI = confidence interval; LUR-BME = Land Use Regression—Bayesian
Maximum Entropy; pg/m3 = micrograms per cubic meter; NHIS = National Health Interview Survey; PM2.5 = particulate matter with a
nominal mean aerodynamic diameter less than or equal to 2.5 pm; TrIPS = Trucking Industry Particle Study.
Note: tStudies published since the 2009 PM ISA. *U.S. and Canadian studies published since the literature cutoff date (~January
2018) for the 2019 PM ISA. Associations are presented per 5-pg/m3 increase in pollutant concentration. Circles represent point
estimates; horizontal lines represent 95% confidence intervals for PM25. Red text and circles represent recent evidence not
considered in previous ISAs or AQCDs, with blue text and circles representing U.S. and Canadian studies published since the
literature cutoff date of the 2019 PM ISA. Study results from Pope et al. (2014) are representative of the results from the American
Cancer Society cohort.
Figure 3-24 Associations between long-term PM2.5 exposure and
cause-specific respiratory mortality in recent North American
cohorts.
1 The examination of respiratory infection mortality is more limited, with recent cohort studies
2 examining either pneumonia alone (Wang et al.. 2020; Pinault et al.. 2017) or the combination of
3 influenza and pneumonia (Pope et al.. 2019). Across the studies, which employed different approaches to
4 assign PM2 5 exposures including a 1-year average in the Medicare cohort (Wang et al.. 2020). a 3-year
5 average with a 1-year lag in the CanCHEC cohort (Pinault et al.. 2017). and a 17-year average in the
6 NHIS cohort (Pope et al.. 2019). each reported positive associations with the magnitude of the association
7 increasing as the length of the exposure window increased.
September 2021
3-90
DRAFT; Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
Other Mortality Outcomes
While cardiovascular- and respiratory-related mortality comprise the majority of mortality
outcomes examined, as noted in the 2019 PM ISA, additional mortality outcomes are also associated with
long-term PM2 5 exposure, including cardiopulmonary, diabetes, and lung cancer (2019 PM ISA,
Section 10.2.5 and Section 11.2.2). Recent cohort studies published since the 2019 PM ISA also provide
evidence of positive associations with cardiopulmonary (Lefler et al.. 2019; Pope et al.. 2019).
diabetes/cardiometabolic disease (Crousc et al.. 2020; Erickson et al.. 2020; Lim et al.. 2018; Pinault et
al.. 2017). and lung cancer (Crousc et al.. 2020; Erickson et al.. 2020; Erickson et al.. 2019; Pope et al..
2019; Cakmak et al.. 2018; Pinault et al.. 2017) mortality.
3.2.2.2.3 Long-Term PM2.5 Exposure and Mortality in Populations with
Preexisting Conditions
In addition to recent studies examining cause-specific mortality, a few studies focused on
examining the overall risk of mortality in individuals with preexisting cardiovascular conditions,
specifically heart failure (HF) (Ward-Caviness et al.. 2020) and previous myocardial infarction (MI)
(Malik et al.. 2019). To date, relatively few studies have been conducted with the sole focus being on
examining associations between long-term PM2 5 exposure and mortality within a cohort of individuals
with a preexisting cardiovascular condition. Instead, studies have traditionally relied on the examination
of effect modification, through stratified analyses, of the PM2 5-mortality association by specific
cardiovascular conditions.
Ward-Caviness et al. (2020) examined associations between annual average PM2 5 exposure at the
time of initial HF diagnosis with all-cause mortality in a hospital-based cohort within North Carolina
developed from electronic health records of individuals diagnosed with heart failure. The authors assigned
annual average PM2 5 exposures to each participant based on their residential address at the time of HF
diagnosis. Exposures were assigned based on the nearest PM2 5 monitor and using the hybrid exposure
model used in the Medicare studies discussed previously that estimated concentrations at 1 km2 (Table
A-7). Over 10,000 of the 35,000 patients within the cohort died less than 1 year after diagnosis, which
does not follow the traditional pattern of HF mortality. As a result, the authors excluded the year after
diagnosis as a time at risk in the model. In a Cox proportional hazards model that controlled for
individual-level covariates, distance to monitor, and neighborhood level socioeconomic variables, Ward-
Caviness et al. (2020) reported a HR for all-cause mortality of 1.84 (95% CI: 1.61, 2.01). In sensitivity
analyses, associations were similar in magnitude to the main analysis when restricting it to participants
<30 km from a PM2 5 monitor and larger in magnitude when applying the hybrid exposure model.
Whereas Ward-Caviness et al. (2020) examined the risk of mortality attributed to long-term PM2 5
exposure in the years following HF diagnosis over the entire study duration, Malik et al. (2019) focused
on examining the 5-year survival for all-cause mortality following an MI event. A total of 5,650 patients
September 2021
3-91
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
with clinically diagnosed MI as defined by having biomarker evidence of myocardial necrosis and
additional clinical evidence of MI, including prolonged ischemic signs/symptoms or electrocardiographic
criteria of ST-segment changes, were enrolled in the Translational Research Investigating Underlying
Disparities in Acute Myocardial Infarction (TRIUMPH) and Prospective Registry Evaluating Myocardial
Infraction: Events and Recovery (PREMIER) studies. Using U.S. EPA's downscaler Community
Multi-Scale Air Quality Model (CMAQ), PM2 5 concentrations were estimated at the census tract centroid
of each participants residence. Exposures assigned to each participant consisted of the 12-month average
PM2 5 concentration prior to MI, which allowed for the examination of the association between PM2 5
exposure and survival after an MI. In a Cox regression model that controlled for numerous
individual-level covariates as well as ozone, Malik et al. (2019) reported a HR of 1.34 (95% CI: 1.17,
1.54) for 5-year all-cause mortality.
Overall, the studies conducted by Ward-Caviness et al. (2020) and Malik et al. (2019) indicate
that preexisting cardiovascular conditions substantially increase the risk of all-cause mortality. This is
further reflected when comparing the magnitude of associations between these studies and the numerous
cohort studies spanning diverse populations summarized in Section 3.1.2.2.1 and Figure 3-19.
3.2.2.2.4 Studies of Life Expectancy
The 2019 PM ISA characterized a recent series of studies evaluating the relationship between
long-term exposure to PM2 5 and mortality by examining the temporal trends in PM2 5 concentrations and
changes in life expectancy. These studies generally observed that decreases in PM2 5 concentrations are
associated with increases in life expectancy. A few recent studies add to this evidence base. Bennett et al.
(2019) reported that PM2 5 concentrations in excess of the lowest observed concentration (2.8 |ig/m3) were
associated with a lower national life expectancy by an estimated 0.15 years for women and 0.13 years for
men (Figure 3-25). Using a different approach, Ward-Caviness et al. (2020) compared participants
residing in areas with PM2 5 concentrations >12 |ig/m3 to participants living in areas with PM2 5
concentrations <12 |ig/m3 and estimated that the years of life lost due to living in areas with higher PM2 5
concentrations was 0.84 years (95% CI, 0.73-0.95) over a 5-year period.
September 2021
3-92
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Female
Male
Life expectancy loss (years)
Source: Copyright permission pending from Bennett et al. (2019).
Figure 3-25 Life expectancy losses by county for females and males as
estimated by Bennett et al. (2019).
3.2.2.2.5 Potential Copollutant Confounding of the PM2.5-Mortality
Relationship
At the completion of the 2009 PM ISA there was limited assessment of potential confounding of
the relationship between long-term PM2 5 exposure and mortality by co-occurring pollutants. Studies
evaluated in the 2019 PM ISA (Section 11.2.3) examined the potential for copollutant confounding by
evaluating copollutant models that included O3 (Figure 11-20), NO2, PM : SO2, and benzene
(Figure 11-21). These studies addressed a previously identified data gap by informing the extent to which
effects associated with exposure to PM2 5 are independent of coexposure to correlated copollutants in
long-term analyses. Overall, PM2 5 effects to remained relatively unchanged in copollutant models
adjusted for NO2, PMick2.j, SO2, or benzene. Recent North American cohort studies conducted additional
analyses that further inform whether the relationship between long-term PMaf exposure and mortality is
confounded by gaseous pollutants or other particle size fractions (i.e., PM ; 2 >).
In an analysis of the NHIS subcohort, Lefler et al. (2019) reported that the PM25-mortality
association was relatively unchanged in copollutant analyses with SO2, NO2, O3, CO, and PM10-2 5. In
addition to Lefler et al. (2019). within the Medicare cohort, Wang et al. (2020) also conducted an analysis
of potential confounding of the PM2 5-mortality association with a focus on ozone and traffic related
pollution. The ozone analysis was conducted in a subset of the cohort that resided in ZIP codes within
6 miles of an ozone AQS monitor. To control for traffic-related pollution, instead of including NO2 in a
copollutant model, Wang et al. (2020) regressed 12-month PM), on NO2 and used the residuals as the
September 2021
3-93
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
exposure metric to estimate the PM2 5 association that is unrelated to traffic. The authors reported
associations similar in magnitude in both a single (HR = 1.03 [95% CI: 1.02, 1.03]) and copollutant
model with ozone (HR= 1.03 [95% CI: 1.02, 1.03]). Positive associations were also reported for the
non-traffic PM2 5 analyses, and although the association was smaller in magnitude to the total PM2 5
analysis, this result provides evidence that the PM2 5-mortality association persists independent of traffic
related-pollutants (HR= 1.01 [95% CI: 1.00, 1.01]).
A few studies within the MAPLE study also conducted analyses to assess potential copollutant
confounding by gaseous pollutants. Within CanCHEC, Crouse et al. (2020) assessed potential copollutant
confounding by ozone, NO2, and oxidant gases in sensitivity analyses using both 3-year and 8-year
average PM2 5 concentrations at 1 km2 resolution. Associations between PM2 5 and mortality were found
to be slightly attenuated but remained positive in copollutant models with each of the gaseous pollutants
when using either 3-year or 8-year average PM2 5 exposures.
While there was no evidence of copollutant confounding in the analysis of CanCHEC by Crouse
et al. (2020). there was some evidence in the pooled analysis of CanCHEC by Pappin et al. (2019). In the
single-pollutant analysis, the authors reported a HR of 1.03 (95% CI: 1.02, 1.03) in the fully adjusted
model, which is consistent with the copollutant model including NO2 (HR = 1.02 [95% CI: 1.01, 1.03]).
However, there was no evidence of a PM2 5-mortality association in copollutant models with ozone
(HR = 0.99 [95% CI: 0.98, 1.00]) and oxidants (HR = 0.98 [95% CI: 0.97, 0.98]). The ozone and oxidants
results of Pappin et al. (2019) are consistent with those of Christidis et al. (2019) within the mCCHS
cohort. However, Christidis et al. (2019) reported evidence of potential confounding by NO2. In a
single-pollutant analysis, the authors reported a HR of 1.05 (95% CI: 1.02, 1.09). In copollutant models,
PM2 5-mortality associations are null in models with ozone (HR= 1.00 [95% CI: 1.00, 1.01]) and oxidants
(HR= 1.00 [95% CI: 0.99, 1.01]), and attenuated in a model withN02 (HR= 1.01 [95% CI: 1.00, 1.02]).
Zhang et al. (2021) also conducted copollutant analyses when examining associations between
long-term PM2 5 exposure and mortality in the Ontario Health Study. The authors reported that the
PM2 5-mortality association in a single-pollutant model (HR= 1.20 [95% CI: 1.09, 1.32]) is relatively
unchanged in a model with NO2 (HR = 1.18 [95% CI: 1.08, 1.30]), which is consistent with Lefler et al.
(2019) in the NHIS cohort and Pappin et al. (2019) within CanCHEC. However, it is inconsistent with the
results from the mCCHS cohort where Christidis et al. (2019) reported evidence of attenuation of the
PM2 5-mortality association.
While there is some evidence from recent studies that PM2 5-mortality associations remain
relatively unchanged in copollutant models, there are some differences in results across studies,
particularly for copollutant analyses including ozone. This difference in results is most evident between
MAPLE studies and other studies that previously examined copollutant confounding by ozone as
summarized in Figure 11-20 of the 2019 PM ISA. This difference between studies could be attributed to
the MAPLE studies using different spatial resolutions in estimating air pollutant concentrations, 1 km2 for
PM2 5 and 21 km2 for ozone, potentially resulting in some degree of exposure error.
September 2021
3-94
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
3.2.2.2.6
Novel Methods to Address Potential Confounding
As discussed in Section 11.2.2.4 of the 2019 PM ISA, some recent studies used statistical
techniques to reduce uncertainties related to potential confounding that can further inform the relationship
between long-term PM2 5 exposure and mortality. An initial study conducted by Janes et al. (2007) and
then followed up by Greven et al. (2011) attempted to assess whether there is evidence of unmeasured
confounding in the relationship between long-term PM2 5 exposure and mortality using data from the
Medicare cohort from 2000-2006. In both studies the authors decomposed the association between PM2 5
and mortality into two components: (1) the association between the "national" trend in the monthly PM2 5
concentrations averaged over the previous 12 months and the national average trend in monthly mortality
rates (purely temporal association); and (2) the association between the "local" trend in the deviation in
the community-specific trend from the national average trend of monthly averages of PM2 5 and the
deviation of the community-specific trends from the national average trend of mortality rates (residual
spatiotemporal association). The authors concluded that differences in effect estimates at these two
spatiotemporal scales raise concerns about confounding bias in these analyses, with the association for the
national trend more likely to be confounded than the association for the local trend. However, in the
process of decomposing the data, it eliminated all spatial variation in air pollution and mortality. Thus,
while the results of the papers themselves provide evidence for an association between exposure to PM2 5
and mortality, it is not possible to directly compare the results of these studies to the results of other
cohort studies investigating the relationship between long-term exposure to PM2 5 and mortality, which
make use of spatial variability in air pollution and mortality data.
Similarly, Pun et al. (2017) completed a sensitivity analyses as part of their Medicare cohort
study for the years 2000-2008 in which they decomposed PM2 5 into "temporal" and "spatiotemporal"
variation, analogous to what was done in Greven etal. (2011). The purpose of this sensitivity analysis
was to determine the presence or absence of bias due to unmeasured confounding. Pun et al. (2017)
observed positive associations for the "temporal" variation model and approximately null associations for
the "spatiotemporal" variation model for all causes of death except for COPD mortality. The difference in
the results of these two models for most causes of death suggests the presence of unmeasured
confounding, though the authors do not indicate anything about the direction or magnitude of this bias. It
is important to note that the "temporal" and "spatiotemporal" coefficients are uninterpretable when
examined individually and can only be used in comparison with one another to evaluate the potential for
unmeasured confounding bias.
As a result of the studies noted above suggesting the presence of unmeasured confounders,
including long-term time trends, Eum et al. (2018) focused specifically on whether temporal confounding
exists in the PM2 5-mortality relationship using the Medicare cohort from 2000-2012. In a base model that
did not account for temporal confounding the authors used the exact methods of Greven etal. (2011) to
decompose PM2 5 into "temporal" and "spatiotemporal" components, with and without the inclusion of
behavioral covariates from the Behavioral Risk Factor Surveillance System (BRFSS), to identify whether
September 2021
3-95
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
there was evidence of unmeasured confounding. Consistent with the previous studies larger associations
were reported for the "temporal" compared to the "spatiotemporal" component, with results being
positive for both components. As a result of the magnitude of associations between these two components
being different, Eum et al. (2018) attempted to assess whether temporal confounding of the
PM2 5-mortality relationship could explain the observed difference in results. The authors developed a
base model which used data for the entire 13 year period and shorter periods ranging from 3 to 12 years
(e.g., 2001-2012, 2009-2012, etc.) as well as three additional models: (1) residual-base model: calculated
the residuals of a linear regression of PM25 on time in 4-year intervals (i.e., 2000-2004, 2005-2008, and
2009-2012), where the residuals were then used as the exposure in the base model; (2) spline-based
model: added a penalized spline with two knots per year to the base model; and (3) decomposition-based
model: added the temporal component of decomposed PM2 5 to the base model based on the approach
described in Greven etal. (2011). In analyses of the base model, the authors observed that as the years of
data included in the analysis increased so did the magnitude of the mortality risk ratio (MRR) (Figure
3-26). Contrary to the base model, as depicted in Figure 3-26. there was a steady decline in the MRR for
both the spline-based and decomposition-based models while MRRs remained relatively stable using the
residual-based model. These results provide some evidence indicating that long-term temporal trends in
PM2 5 concentrations may be one source of unmeasured confounding to consider in long-term exposure
studies lasting many years, and that a residual-based approach could potentially account for those trends.
The results of Eum et al. (2018) are consistent with a confounder analysis of temporal trends
conducted by Wu et al. (2020a). originally discussed in Section 3.1.2.2.1. In the main analysis, the
authors controlled for temporal trends by adjusting for calendar year and also included meteorological
variables to account for season (i.e., summer and winter), maximum daily temperature, and relative
humidity. In Cox and Poisson models that did not adjust for calendar year, HRs increased in analyses of
both the entire cohort (Cox, main analysis, HR = 1.03 [95% CI: 1.03, 1.04]; Cox, minus calendar year,
HR = 1.08 [95% CI: 1.08, 1.09]) and those limited to individuals living in locations where PM2 5
concentrations were <12 (ig/m3 for the entire study duration (main analysis, HR =1.17 [95% CI: 1.16,
1.18]; minus calendar year, HR = 1.25 [95% CI: 1.24, 1.27]), indicating that not accounting for temporal
trends may overestimate associations. In addition, in analyses that excluded meteorological variables,
HRs were relatively unchanged compared to the main analysis (Cox, HR = 1.03 [95% CI: 1.02, 1.03]),
indicating they are not a source of residual confounding in long-term exposure studies.
September 2021
3-96
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Length of study period (years)
Source: Copyright permission pending, Eum et al. (20181.
Figure 3-26 Mortality risk ratios for a 10 |jg/m3 increase in PM2.5 by length of
study for the base model and temporal adjusted models in Eum et
al. (2018).
Although there is extensive evidence of a relationship between long-term PM2.5 exposure and
mortality provided by numerous cohort studies discussed throughout this section, potential residual
confounding remains a concern as reflected in the studies discussed above. Of these studies, only Eum et
al. (2018) and Wu et al. (2020a) attempted to address a specific potential source of confounding
(i.e., long-term trends). Erickson et al. (2019) also examined the potential implications of unmeasured
confounders on the PM2 5-mortality relationship by using an approach that indirectly adjusts for missing
covariates via partitioned regression, based on an approach developed by Shin et al. (2014). This
approach controls for unmeasured potential confounders that are not available in the primary data set for
an individual cohort by using data from an ancillary matching data set. Shin et al. (2014) suggested that in
applying indirect adjustment it is important for the primary data set and ancillary data set to have a similar
distribution of the primary exposure (i.e., PM2.5) among subjects across demographic characteristics, and
to conduct a "gold-standard" evaluation to assess the magnitude of bias correction by excluding and
indirectly adjusting for specific variables in both data sets. Erickson et al. (2019) applied both of these
suggestions by using three cohorts from the MAPLE studies. Specifically, the 2001 CanCHEC, which
represented the main data set, and the CCHS as the ancillary matching data set were used for validation
analyses, while the assessment of the direction and magnitude of bias of this approach was conducted
September 2021
3-97
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
using the mCCHS cohort as the ancillary data set that contained information, missing in CanCHEC, on
cigarettes/day, alcohol use, fruit and vegetable intake, and leisure exercise.
A multi-step process was used to conduct this analysis that consisted of the following: (1) assess
the representativeness of the primary data set (i.e., CanCHEC) to the ancillary data set (CCHS) to
compare absolute and proportional differences in the distribution of PM2 5 concentrations within the two
populations; (2) conduct an internal validation to assess the degree of bias in HRs when applying the
indirect adjustment; and (3) conduct an external validation to assess bias of using CCHS as the ancillary
data set to indirectly adjust CanCHEC. In (1), Erickson et al. (2019) applied sampling weights to the
CCHS to ensure the distribution of PM2 5 concentrations was the same across demographic and
socioeconomic characteristics at baseline. For (2), the authors used a "gold-standard" that consisted of
removing and indirectly adjusting for variables available within the CanCHEC cohort, i.e., education and
income. The results of this adjustment were then compared to a "True Model" that adjusted for education,
income, and all other individual-level covariates. To conduct the internal validation, a "Partial Model"
was developed that excluded education and income. The coefficient and variance terms for education and
income were derived from the "True Model" with PM2 5 excluded. The coefficient and variance terms for
education and income were used to indirectly adjust for each covariate in the "Partial Model" resulting in
the "Internal (validation) Model". A comparison of the "True Model" and "Internal (validation) Model"
allows for an assessment of the indirect adjustment approach. In (3), the authors used a similar approach
to the internal validation for conducting the external validation using variables available in both
CanCHEC and CCHS (i.e., education and income) instead of using only the CanCHEC data. Lastly, after
going through each of the aforementioned steps to validate the indirect adjustment approach, Erickson et
al. (2019) used the data from mCCHS on cigarettes/day, alcohol use, fruit and vegetable intake, and
leisure exercise to indirectly adjust for the covariates within CanCHEC. For all analyses both
time-varying and static PM2 5 concentrations were evaluated, but for comparison with the exposure
assignment approach used within the MAPLE studies, the results below primarily focus on the
time-varying exposure.
In the internal and external validations, which focused on the educational and income variables,
in analyses of non-accidental, cardiovascular, IHD, and lung cancer mortality, the authors reported an
overall reduction (i.e., bias) in the HRs compared to the "True Model" when indirectly adjusting for the
covariates. There was a 3-5.6% bias across mortality outcomes for the "Partial Model," which did not
control for either covariate, which was reduced to 1.7-2.9 and 1.3-2.3% in the internal and external
validation analyses, respectively. The results of the validation analyses were further confirmed in the
main analysis using the mCCHS cohort as the ancillary data set, where there was an approximately 1.5%
increase in the HR when indirectly adjusting for missing covariates in the CanCHEC cohort.
Additionally, the results using the ancillary data set within the CanCHEC cohort for nonaccidental,
cardiovascular, and IHD mortality were consistent in magnitude to those reported in mCCHS.
September 2021
3-98
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
The methods detailed by Erickson et al. (2019) provide an approach to address the issue of
unmeasured confounders, but this method is limited by the availability of an ancillary data set. In
addition, the direction and magnitude of the bias is dependent upon the direction and magnitude of the
correlation between air pollution and the missing covariates within the population of interest. However,
the direction of the bias can vary when using this confounder adjustment approach, with an
underestimation of the HR observed when using a time-varying exposure and an overestimation observed
when using a static exposure when compared to the "True Model." Overall, the results of Erickson et al.
(2019) indicate that the lack of data for some covariates leads to an underestimation, not an
overestimation, of the PM2 5-mortality association.
In conclusion, recent studies that further evaluate the potential implications of unmeasured
confounders on the association between long-term PM2 5 exposure and mortality indicate that bias can
occur in either direction. However, across the studies evaluated, the control for unmeasured confounders
as detailed in Eum et al. (2018). Wu et al. (2020a). and Erickson et al. (2019) do not result in the
elimination of the association, but instead provide additional confirmation that an association between
long-term PM2 5 exposure and mortality exists when accounting for additional confounders.
3.2.2.2.7 Examination of the Concentration-Response (C-R) Relationship
between Long-Term PM2.5 Exposure and Mortality
An important consideration in characterizing the association between long-term PM2 5 exposure
and mortality is whether the concentration-response (C-R) relationship is linear across the full
concentration range that is encountered, or if there are concentration ranges where there are departures
from linearity. The 2009 PM ISA characterized the results of an analysis by Schwartz et al. (2008) that
demonstrated that the shape of the C-R curve was generally linear. A substantially larger number of
studies was evaluated in the 2019 PM ISA, which provided strong evidence for a linear, no-threshold
concentration-response relationship for long-term PM2 5 exposure and total (nonaccidental) mortality
rU.S. EPA (2019); Section 11.2.4], Evidence presented in the 2019 PM ISA demonstrated less certainty
in the shape of the C-R curve at mean annual PM2 5 concentrations generally below 8 (ig/m3, though some
studies characterized the C-R relationship with certainty down to 4 (ig/m3.
A number of recent studies conducted analyses that further inform the shape of the C-R
relationship for the association between long-term PM2 5 exposure and mortality and are summarized in
Table 3-6. Generally, the results of these analyses continue to support a linear, no-threshold relationship
for total (nonaccidental) mortality, though there is some evidence for a sublinear (Zhang et al.. 2021;
Pope et al.. 2019) or supralinear (Christidis et al.. 2019; Pappin et al.. 2019; Pinault et al.. 2017)
relationship at lower ambient PM2 5 concentrations. Many of the recent studies that conducted C-R
analyses include concentration ranges that extend below the level of the current annual PM2 5 NAAQS.
September 2021
3-99
DRAFT: Do Not Cite or Quote
-------
Table 3-6 Summary of studies examining the concentration-response (C-R)
relationship or conducted threshold analyses for long term PM2.5
exposure and mortality.
Study Location—Cohort
Table/Figure from Reference
Christidis etal. (2019)
mCCHS
(Figure 2)
Exposure; PM2.5 Mean;
Range in |jg/m3
PM2.5 estimates at 1 km2 over 3-yr
average (3-yr/1-km model) with
single-yr lag assigned to postal code
of residence
7.37 (0.37-20.0)
Statistical Analysis
Summary
C-R: Shape Constrained Health
Impact Function (SCHIF) fits a class
of flexible, but monotonically
nondecreasing functions to select
best fitting model
Supralinear at lower concentrations
(<5 pg/m3)
Elliott et al. (2020)
Nurses Health Study
(Table 2)
24-mo average ambient PM2.5
exposures were estimated at
residential addresses using a
spatiotemporal prediction model
13.7; (NR)
Exposure categories (quintiles)
Monotonic (linear) relationship
demonstrated by HRs remaining
generally consistent and statistically
significant. P for trend = 0.07
Pappin et al. (2019)
CanCHEC
(Figure 2)
PM2.5 estimates at 1 km over 3-yr
average (3-yr/1-km model) with
single-yr lag assigned to postal code
of residence
6.68-7.95 (0.37-20.0)
C-R analysis, 3 step approach:
(1) restricted cubic splines (RCS)
with a large number of knots;
(2) smooth potential erratic
predictions from the large number of
knots using monotonically increasing
smoothing splines (MISS); and (3) fit
the shape constrained health impact
function (SCHIF) to the MISS
predictions
Supralinear at lower concentrations
(<5 pg/m3)
Pinault et al. (2017)
CanCHEC
(Figure 2; Table S4)
PM2.5 estimates at 1 km over 3-yr
average (3-yr/1-km model) with
single-yr lag assigned to postal code
of residence
7.37 (0.37-20.0)
C-R: Shape Constrained Health
Impact Function (SCHIF) fits a class
of flexible, but monotonically
nondecreasing functions to select
best fitting model (counterfactual is
0 pg/m3); cutpoint analyses: 0-5,
5-10, >10 pg/m3
Supralinear at lower concentrations
(<5 pg/m3); HRs remained positive
and statistically significant in the 2
lowest cutpoint categories with
highest HRs for the 0-5 pg/m3
category, consistent with the
supralinear C-R function
September 2021
3-100
DRAFT: Do Not Cite or Quote
-------
Table 3-6 (Continued): Summary of studies examining the concentration-
response (C-R) relationship or conducted threshold
analyses for long term PM2.5 exposure and mortality.
Study Location—Cohort
Table/Figure from Reference
Pope etal. (2019)
NHIS Cohort
(Figure 4)
Wang et al. (2020)
Medicare Cohort
(Figure 1; Table S4)
Ward-Caviness et al. (2020)
HF Patient Cohort
(Table 2; Figure 3)
Wu et al. (2020a)
Medicare Cohort
(Figure 3)
Exposure; PM2.5 Mean;
Range in |jg/m3
Population-weighted annual PM2.5
concentrations averaged for
census-tract centroids.
10.7; (2.5-19.2)
Daily PM2.5 was estimated on a 6-km
grid using a spatiotemporal model
10.3 (NR)
An ensemble-based prediction model
was used to estimate daily PM2.5
concentrations for a 1-km2 grid
network across the contiguous U.S
9.8 (NR)
Statistical Analysis
Summary
C-R: Integrated model that fit a class
of flexible, but monotonically
nondecreasing functions to select
best fitting model.
Generally linear, though some
evidence of a shallower slope at
lower concentrations (<8 |jg/m3)
C-R: RCS model with 3 knots;
Threshold: PM2.5 <8, <10, <12 |jg/m3
C-R: Linear across distribution of
exposure concentrations with no
evidence of a threshold; HRs
remained positive and statistically
significant when only participants with
exposure concentrations below 8, 10,
or 12 |jg/m3 were included
HRs remained positive and
statistically significant when only
participants with exposure
concentrations below 12 |jg/m3 were
included; linear C-R curve, with
greatest certainty between 9 and
13 |jg/m3
Threshold model (PM2.5 <12 |jg/m3)
HRs remained positive and
statistically significant when only
participants with exposure
concentrations below 12 |jg/m3 were
included
Nearest monitor (Threshold model) or Threshold model (PM2 5<12 |jg/m3);
Harvard's 1 km * 1 km modeled C-R: limited to PM2.5 concentration
PM2.5 surface (C-R figure); within inner 95% of distribution
10.3; (8-14) (8-14 |jg/m3)
September 2021
3-101
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Table 3-6 (Continued): Summary of studies examining the concentration-
response (C-R) relationship or conducted threshold
analyses for long term PM2.5 exposure and mortality.
Study Location—Cohort
Table/Figure from Reference
Zhang et al. (2021)
Ontario Health Study
(Figure 2; Tables S3, S8)
Exposure; PM2.5 Mean;
Range in |jg/m3
PM2.5 estimates at 1 km2 over 3-yr
average (5-yr/1-km model) with
single-yr lag assigned to postal code
of residence
7.8 (NR)
Statistical Analysis
Summary
C-R: Shape Constrained Health
Impact Function (SCHIF) fits a class
of flexible, but monotonically
nondecreasing functions to select
best fitting model; threshold model
(PM2.5 <10 and <8.8 |jg/m3);
Categorical exposure (quartiles)
Sublinear relationship with shallower
slope at lower concentrations and
steeper slope at mid-range
concentrations; HRs remained
positive and statistically significant
when only participants with exposure
concentrations below 10 |jg/m3 were
included. Results were positive but
attenuated and no longer statistically
significant below 8.8 |jg/m3;
categorical exposure results
demonstrate sublinear relationship,
similar to C-R function with strongest
association for concentrations
>8.5 |jg/m3
CanCHEC = Canadian Census Health and Environment Cohort; HF = heart failure; HR = hazard ratio; km = kilometer;
mCCHS = mortality Canadian Community Health Survey; NHIS = National Health Interview Survey; NR = not reported;
SCHIF = Shape Constrained Health Impact Function.
Wang et al. (2020) and Ward-Caviness et al. (2020) observed linear, no-threshold
concentration-response relationships for total (nonaccidental) mortality, with confidence in the
relationship down to a concentration of 5 and 9 (ig/m3, respectively. Using exposure categories
(i.e., quintiles of exposure) to estimate the shape of the concentration-response relationship, Elliott et al.
(2020) report evidence that supports a monotonic (linear) function.
Studies that relied on data from Canadian cohorts evaluated the shape of the
concentration-response relationship using a Shape Constrained Health Impact Function (SCHIF) approach
to model the function (Zhang et al.. 2021; Christidis et al.. 2019; Pappin et al.. 2019; Pinault et al.. 2017).
The SCHIF approach, developed by Nasari et al. (2016). fits a class of flexible, but monotonically
nondecreasing functions to select the best fitting model of the concentration-response relationship. Most
of the studies that used the SCHIF approach (Christidis et al.. 2019; Pappin et al.. 2019; Pinault et al..
2017) identified a supralinear concentration-response relationship at relatively low PM2 5 concentrations
(<5 |_ig/m3) (for example, see Figure 3-27). In contrast, Zhang et al. (2021) applied the SCHIF approach to
their analysis of the Ontario Health Study and identified a sublinear concentration-response relationship,
with a more shallow slope observed for PM2 5 concentrations <8 (ig/m3. Analyses of exposure categories
(i.e., quartiles) by Zhang et al. (2021) provides additional support for a sublinear concentration-response
September 2021
3-102
DRAFT: Do Not Cite or Quote
-------
1 relationship. A similar sublinear relationship was reported by Pope et al. (2019) for a U.S. cohort (Figure
2 3-28).
Pooled SCHIF
in
0 5 10 15
PM? 5 -
Source: Copyright permission pending from Pappin et al. (2019).
Note: Uncertainty bounds are displayed as gray shaded area.
Figure 3-27 Shape Constrained Health Impact Function predictions by PM2.5
concentration for the pooled CanCHEC cohort.
September 2021
3-103
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
1 2 -
a
*
«
ro
(E
X
I 1.1 -
8
<
1 T W V
5 10 15 20
Source: Copyright permission pending from Pope et al. (20191.
Note: Shaded area represents the 95% uncertainty bounds.
Figure 3-28 Estimated concentration-response associations between PM2.5
and all-cause mortality with a flexible modeling approach within
the NHIS cohort.
In addition to statistical analyses of the concentration-response relationship, several studies
conducted threshold analyses to estimate associations between all-cause mortality and PM2.5
concentrations below a certain concentration. Pinault et al. (2017) reported that HRs remained positive
and statistically significant when examining cut-point categories of 0-5 and 5-10 (ig/m3, with the highest
HRs for the 0-5 (ig/m3 category, consistent with the supralinear concentration-response function
estimated by the SCHIF analysis. Wang et al. (2020) observed that HRs remained positive and
statistically significant when restricting analyses to participants with exposure concentrations below 8, 10,
or 12 (.ig/rn3. Similarly, Ward-Caviness et al. (2020) and Wu et al. (2020a) reported that HRs remained
positive and statistically significant when restricting analyses to participants with exposure concentrations
below 12 (.ig/rn3. Zhang et al. (2021) noted that HRs remained positive and statistically significant when
only participants with exposure concentrations below 10 (ig/m3 were included, though the results were
positive, but attenuated, and no longer statistically significant when restricting to exposure concentrations
below 8.8 (ig/m3.
In addition to examining the C-R relationship between long-term PM2 5 exposure and all-cause
mortality, a limited number of studies evaluated the C-R relationship with cause-specific mortality. Wang
et al. (2020) reported a linear, no-threshold C-R relationship for both cardiovascular and respiratory
mortality. When the authors adjusted for ozone, the C-R relationship remained the same for
September 2021
3-104
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
cardiovascular mortality, but the C-R relationship for respiratory mortality became supralinear below
10 (ig/m3. In threshold analyses, Wang et al. (2020) observed that HRs remained positive and statistically
significant when restricting analyses to participants with exposure concentrations below 8, 10, or
12 (ig/m3. In analyses stratified by PM2 5 concentration, Haves et al. (2020) reported evidence of positive
associations with cardiovascular mortality that increased in magnitude and decreased in precision as the
range of PM25 concentrations examined increased from 8-12, to 12-20, and finally to over 20 (ig/m3.
When evaluating deaths due to diabetes, Lim et al. (2018) observed a linear C-R relationship, with the
greatest confidence between 10 and 15 (ig/m3. Overall, recent studies that evaluated the C-R relationship
for long-term PM2 5 exposure and cause-specific mortality are consistent with those that examined
all-cause mortality.
Consistent with the conclusions of the 2019 PM ISA, recent studies provide evidence that
continues to support a linear, no-threshold C-R relationship for long-term PM2 5 exposure and all-cause or
cause-specific mortality across the range of exposure concentrations observed in North American cohort
studies, with some studies characterizing the C-R relationship with certainty down to 4 (ig/m3. The
evidence remains consistent in supporting a no-threshold relationship, and in supporting a linear
relationship for PM2 5 concentrations >8 (ig/m3. However, some uncertainties remain about the shape of
the C-R curve at relatively low PM2 5 concentrations (<8 (.ig/ni3). with some recent studies providing
evidence for either a sublinear, linear, or supralinear relationship at these lower concentrations.
3.2.2.3 Recent Studies Examining the PIVh.s-Mortality Relationship through
Accountability Analyses and Causal Modeling Methods
Within the 2019 PM ISA, a few studies were evaluated that conducted analyses that further
informed the relationship between long-term PM2 5 exposure and mortality through the use of causal
modeling methods (2019 PM ISA, Section 11.2.2.4). These initial studies provided additional support for
a causal relationship between long-term PM2 5 exposure and mortality. Since the literature cutoff date of
the 2019 PM ISA, additional epidemiologic studies have been identified consisting of accountability
analyses as well as studies that implemented causal modeling methods that have the ability to reduce
uncertainties related to confounding bias in the examination of the relationship between long-term PM2 5
exposure and mortality (Table A-8). Study-specific details of the methods implemented in these recent
studies can be found in Table 3-7.
September 2021
3-105
DRAFT: Do Not Cite or Quote
-------
Table 3-7 Description of methods from epidemiologic studies using accountability analyses or causal
modeling methods to examine long-term exposure to PM2.5 and mortality.
Study/Location/Popu la-
tion (Cohort)/Years
Statistical Method
Results
Wei et al. (2020)
Massachusetts
Medicare
2000-2012
Generalized Propensity Score (GPS): estimates the conditional probability of an individual being
exposed to the observed concentration level, accounting for all measured potential confounders
In the design stage, an ordinary least squares (OLS) model regressed predicted PM2.5 concentrations
against a linear combination of covariates, including the other pollutants (Table A-8). In the analysis
stage, an ordinary least squares regression was used to fit a linear probability model relating
mortality with the predicted PM2.5 concentration and the estimated GPS.
35.4 (95% CI: 33.4, 37.6)
excess deaths per 10 million
person-days for each 1 |jg/m3
increase in annual PM2.5
concentrations
September 2021
3-106
DRAFT: Do Not Cite or Quote
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied causal modeling
statistical methods in examining long-term PM2.5 exposure and mortality.
Study/Location/Popu la-
tion (Cohort)/Years
Statistical Method
Results
Wei et al. (2021b)
Massachusetts
Medicare
2000-2012
Three GPS approaches were used:
Linear Probability Model: In the design stage, GPS was constructed by fitting a linear regression of
predicted PM2.5 concentration against a column vector of covariates including copollutants (Table
A-8) In the analysis stage, a linear probability model was fitted with the outcome of death, against the
predicted PM2.5 concentration and the GPS.
Weighted Least Squares: In the design stage, the person-days that had the same sex, race, age,
Medicaid eligibility, ZIP code of residence, and date were aggregated as a single record and
assigned the numbers of person-days for that record as a weight. The GPS was constructed by fitting
a weighted linear regression of predicted PM2.5 concentrations against all the covariates from this
aggregated data set, with continuous covariates modeled with cubic polynomials. The average
outcome for each aggregated person-day group was calculated and assigned to the person-day in
the aggregated data set. A weighted linear regression was fitted for the averaged outcome against
the predicted PM2.5 concentration and the estimated GPS.
Linear Probability Model and
Weighted Least Squares:
1053 (95% CI: 984, 1,122)
annual early deaths for a
1 |jg/m3 increase in annual
PM2.5 concentrations
Moon RF: 1,058 (95% CI:
988, 1,127) annual early
deaths for a 1 |jg/m3 increase
in annual PM2.5
concentrations
Moon RF: In the design stage, the number of person-days aggregated for each record was used as
the frequency weight and sampled 62,000 person-days without replacement. With this sample, trees
were built for PM2.5 to make predictions of the exposure for each person-day, which was repeated
100 times. The final predicted PM2.5 concentration for each person-day was obtained by averaging
the predictions of the 100 trees. The GPS was constructed by using the averaged predictions of the
100 trees as the predicted PM2.5 concentrations and covariates for each person-day in the
aggregated data set. In the analysis stage, weighted regression of the averaged outcome was fitted
against the predicted PM2.5 concentration and the estimated GPS to obtain the effect estimate.
September 2021
3-107
DRAFT: Do Not Cite or Quote
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied causal modeling
statistical methods in examining long-term PM2.5 exposure and mortality.
Study/Location/Popu la-
tion (Cohort)/Years
Statistical Method
Results
Wu et al. (2020a)
U.S.
Medicare
2000-2016
Three GPS approaches were used that required GPS estimation as the first step. The conditional
density of predicted annual average PM2.5 concentration on the 14 ZIP code or county-level
covariates, with dummy variables for region and calendar year, were modeled using gradient
boosting machine with normal residuals.
GPS matching: a matched pseudo-population was constructed. Once the covariate balance was
achieved, a univariate Poisson regression model was fit to regress the death counts, with an offset of
person-time term, on PM2.5 exposure, stratifying by the individual-level covariates and the same
follow-up year.
GPS weighting: a weighted pseudo-population was constructed. Once the covariate balance of the
weighted pseudo-population was achieved, a weighted univariate Poisson regression was fitted,
regressing the death count, with an offset term of person-time, on PM2.5 exposure incorporating the
assigned weights and stratifying by the individual-level covariates and the same follow-up year.
For a 10 |jg/m3 decrease in
annual PM2.5 concentrations
GPS matching: HR = 1.068
(95% CI: 1.054, 1.083)
GPS weighting: HR = 1.076
(95% CI: 1.065, 1.088)
GPS adjustment: HR = 1
(95% CI: 1.061, 1.082)
072
GPS adjustment: the conditional expectation of the death counts given the exposure and estimated
GPS was modeled as a stratified Poisson regression with flexible formulation of bivariate variables,
with the corresponding offset term of person-time. A univariate linear regression regressed the
counterfactual mean hazard rates for each PM2.5 concentration, stratified by the individual-level
covariates and the same follow-up year.
September 2021
3-108
DRAFT: Do Not Cite or Quote
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied causal modeling
statistical methods in examining long-term PM2.5 exposure and mortality.
Study/Location/Popu la-
tion (Cohort)/Years
Statistical Method
Results
Wu etal. (2019)
New England (Vermont,
New Hampshire,
Connecticut,
Massachusetts, Rhode
Island, and Maine)
Medicare
2000-2012
GPS: Regression calibration-GPS (RC-GPS) approach. The RC step of this approach adjusts for
measurement error in a continuous exposure. The adjustment relies on two assumptions:
transportability and nondifferential measurement error. Transportability assumes that the relationship
between continuous exposure (X), the error-prone continuous exposure (W), and the error-free
covariates associated with the measurement error (D) would be the same in the validation study
where X is observed and the main study in which it is not. The nondifferential measurement error
assumption is equivalent to the surrogacy assumption and means the conditional distribution of
outcome given X, W, D depends only on X and D. The relationship between true exposures X and
error-prone exposures W, conditional on other covariates D, was modeled using a regression model,
specified by mean and variance. More specifically in the RC stage, Wu et al. (2019) obtained the
PM2.5 exposure at each grid cell in New England and then fit the regression model to include 14
meteorological variables as predictors.
In the second step, the GPS is implemented. The authors define the GPS as the conditional
probability of receiving each category of the exposure given other pre-exposed covariates (C) (Table
A-8). There are two assumptions for the GPS implementation: overlap/positivity and weak
unconfoundedness. The overlap/positivity assumption guarantees that for all possible values of C,
the average treatment effect can be estimated for each category of the exposure without relying on
extrapolation (Wu et al.. 2019). Weak unconfoundedness assumes that the assignment mechanism
is weakly unconfounded (Wu et al.. 2019). Three GPS approaches were considered:
Subclassification GPS:
IRR = 1.03 (95% CI: 1.01,
1.05)
IPTWGPS: IRR = 1.02 (95%
CI: 1.01, 1.04)
Matching GPS: IRR = 1.03
(95% CI: 1.01, 1.05)
Subclassification GPS: individuals were classified into groups based on the GPS elements, with each
group containing the observations having similar values of the corresponding estimated GPS
elements.
Inverse Probability of Treatment Weighting (IPTW) GPS: weights each individual by the inverse of
their GPS.
GPS matching: involves matching individuals who receive one category of exposure to individuals
who received another category of exposure based on the estimated GPS.
For each of the three GPS approaches, Wu et al. (2019) used multinomial logistic regression with 16
area-level covariates as confounders to construct the GPS. After constructing GPS for each
approach, a stratified log-linear outcome model with a person-time offset estimated the incidence rate
ratio (IRR) of the effect of long-term PM2.5 exposure on all-cause mortality. The stratification variables
were the individual-level covariates (age, as 5-yr age categories; race; sex; and Medicare eligibility).
September 2021
3-109
DRAFT: Do Not Cite or Quote
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied causal modeling
statistical methods in examining long-term PM2.5 exposure and mortality.
Study/Location/Popu la-
tion (Cohort)/Years
Statistical Method
Results
Schwartz et al. (2018b)
Northeastern and
Mid-Atlantic States
(Maine, New Hampshire,
Vermont, Massachusetts,
Rhode Island,
Connecticut, New York,
New Jersey, Delaware,
Pennsylvania, Maryland,
Washington, DC, Virginia,
and West Virginia)
Inverse Probability Weights (IPW): A GPS was fitted with a linear regression of continuous predicted
PM2.5 concentrations against the measured covariates. The IPW is constructed as the marginal
probability density of exposure as the numerator and the probability density of each observation
receiving the predicted PM2.5 concentration in a year given the covariates in that year as the
denominator. Separate logistic regression models were fitted to estimate the risk of dying at that age
given the annual average PM2.5 concentration at each subject's residential ZIP code and age-specific
IPW weights to allow the influence of potential confounders to change with age.
The estimated mean age at
death for a population with an
annual average PM2.5
concentration of 12 |jg/m3
was 0.89 (95% CI: 0.88, 0.91)
years
Medicare
2000-2013
Awad et al. (2019)
U.S.
Medicare
2000-2012
IPW: Constructed from a GPS model in which the weights were the inverse conditional probabilities
of the continuous exposure given the covariates (Table A-8). The weights were then stabilized using
the marginal probability of exposure as the numerator and the denominator was the conditional
density function of change in exposure where the vector of covariates evaluated at observed
covariate values for each participant. The Cox proportional model estimated the effect of change in
annual PM2.5 concentrations on the risk of all-cause mortality, stratified by ZIP code before moving,
with the IPW to account for confounding after moving and follow-up time.
For a 10 |jg/m3 increase in
annual PM2.5 concentrations:
HR = 1.21 (95% CI = 1.20,
1.22) among White
individuals
HR = 1.12 (95% CI = 1.08,
1.15) among Black individuals
Hiqbee et al. (2020)
U.S.
National Center for Health
Statistics
1986-2015
IPW: Constructed by taking the inverse of the conditional probability of the exposure to a given value For a 10 |jg/m3 increase in
from a continuous scale of PM2.5 concentrations and stabilized by multiplying the weights by the
marginal probability of the PM2.5 concentration. The Cox proportional hazard models estimated the
hazard ratios associated with a 10 |jg/m3 increase in PM2.5 concentration. Each of the covariates
(Table A-8) were included as confounders while constructing the IPW for weighting the estimated
model.
annual PM2.5 concentrations:
All-cause mortality
HR = 1.12 (95% CI: 1.08,
1.15)
Cardiopulmonary mortality
HR = 1.23 (95% CI: 1.17,
1.29)
September 2021
3-110
DRAFT: Do Not Cite or Quote
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied causal modeling
statistical methods in examining long-term PM2.5 exposure and mortality.
Study/Location/Popu la-
tion (Cohort)/Years
Statistical Method
Results
Wei et al. (2021a)
U.S.
Medicare
2000-2016
IPW: Proposed a decile binning, which divided PM2.5 concentrations by deciles and predicted the
inverse probability of being assigned to the observed group for each observation, adjusting for
copollutants (ozone and NO2), personal characteristics, meteorological, socioeconomic, behavioral,
and medical access variables, and long-term time trend (Table A-8). The IPW was constructed in two
stages—a design stage and an analysis stage. In the design stage, a randomized pseudo-population
was constructed by weighting the observed population by the inverse probability of the exposure
given all the measured confounders. In the analysis stage, the treatment effect was estimated among
the constructed pseudo-population. The IPW for each PM2.5 decile was stabilized by using the
probability of any observed exposure being within the decile as the numerator. The denominator was
the inverse logistic link function of a gradient boosting machine model with logistic loss function for
predicting the probability of the observed binned exposure given the set of confounders, weighted by
the number of person-years aggregated in the stratum. A log linear regression model was then fitted
to estimate the number of deaths and PM2.5 decile category, weighted by the stabilized inverse
probabilities.
2nd decile group (mean
annual PM2.5 concentration of
6.60 pg/m3): RR = 1.02 (95%
CI: 1.02, 1.03)
10th decile group (mean
annual PM2.5 concentration of
15.47 pg/m3): RR = 1.21
(95% CI: 1.20, 1.21)
Corriqan et al. (2018)
U.S.
National Center for Health
Statistics
Difference-in-Difference (DID): A linear regression model was used to estimate the association
between the change in cardiovascular mortality rate and the change in PM2.5 across U.S. counties,
adjusted for potential confounders.
1.10 (95% CI: 0.37, 1.82)
fewer cardiovascular deaths
per year per 100,000 people
for each 1 pg/m3 decrease in
annual PM2.5 concentrations
2000-2010
Sanders et al. (2020)
U.S.
Medicare
2000-2013
DID: Examined if policy actions reduced PM2.5 concentrations and mortality in treatment counties
relative to control counties. DID flexibly captures the effects of treatment overtime by using separate
dummy variables for each year.
With the event study design in the DID models, nearest neighbor matching based on propensity
score was used as an estimate of probability of attainment status. The independent variables used to
estimate the propensity score were the mortality counts and the population for each year from
2000-2005 and were sampled without replacement.
An instrumental variable analysis was applied by dividing the mortality coefficient by the PM2.5
coefficient to obtain the estimate of the effect of PM2.5 on mortality, which is equivalent to a Wald
instrument variable estimator, where the standard errors of the estimates were calculated.
After the regulatory changes
in 2005, PM2.5 concentrations
decreased 1.59 pg/m3 (95%
CI: 1.39, 1.80) and mortality
rates among those 65 years
and older also decreased by
0.93% (95% CI: 0.10, 1.77).
September 2021
3-111
DRAFT: Do Not Cite or Quote
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied causal modeling
statistical methods in examining long-term PM2.5 exposure and mortality.
Study/Location/Popu la-
tion (Cohort)/Years
Statistical Method
Results
Henneman et al. (2019)
U.S.
Medicare
DID: Estimated changes in each morality associated with changes in annual PM2.5 concentrations.
The differences between 2012 and 2005 mortality rates (per 10,000 beneficiaries) were estimated
from a linear model that included the difference in PM2.5 concentrations from 2012 and 2005,
controlled for changes in census and meteorological variables.
Reduction in all-cause
mortality of-0.38 (95% CI:
-2.76, 2.01) per 10,000
person-days for each 1 |jg/m3
decrease in annual PM2.5
concentrations during the
time period
2005-2012
Fan and Wang (2020)
U.S.
Medicare
1999-2013
DID: power plant retirement
decreased both monthly
PM2.5 concentrations by
2.1 |jg/m3, and the monthly
age-adjusted mortality by
approximately 15 people per
100,000 people (or 3.6%) in
treated counties, relative to
control counties
DID: Treatment counties based on the location of the power plant and wind direction. For each
treatment county, covariate matching was used, with matching based on county characteristics to
select controls. After constructing a set of treatment and control counties, the effect of annual PM2.5
on monthly age-adjusted mortality was estimated using an instrument variable (IV) approach to
compare the before and after the retirement of coal plants between the treated and control counties.
The instrumental variables were weather variables and time-varying socioeconomic variables. DID
approach was also used to estimate the impact of coal plant retirements on PM2.5 and mortality
among populations older than 65.
IV: a 1-|jg/nr reduction in
annual PM2.5 concentrations
led to 7.17 fewer deaths per
100,000 people per month, or
a 1.7% lower monthly
mortality rate among people
older than 65 yr of age
September 2021
3-112
DRAFT: Do Not Cite or Quote
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied causal modeling
statistical methods in examining long-term PM2.5 exposure and mortality.
Study/Location/Popu la-
tion (Cohort)/Years
Statistical Method
Results
Yitshak-Sade et al.
(2019b)
Maine, New Hampshire,
Vermont, Massachusetts,
Rhode Island,
Connecticut, New York,
New Jersey, Delaware,
Pennsylvania, Maryland,
Washington, DC, Virginia,
and West Virginia
DID: Used a Poisson survival analysis using the Anderson-Gill formulation with time-varying
covariates. The data was randomly split into subsets due to computational limitations. The effect
estimates were then pooled using a fixed-effect meta-analysis.
4.04% (95% CI: 3.49, 4.59)
increase in mortality rates for
an IQR (3 |jg/m3) increase in
annual PM2.5 concentrations
Medicare
2000-2013
Schwartz et al. (2021)
U.S.
Medicare
2000-2016
DID: Applied the standard approach for continuous predictors. The mortality rate in a ZIP code given Probability of dying in each
the demographic group (age, sex, race, Medicaid coverage) was associated with annual PM2.5
concentrations, given the time-invariant or slowly changing confounders in a ZIP code, and the
time-varying confounders that are common across ZIP codes. The time-invariant confounders were
controlled by fitting individual intercepts for each ZIP code, while the time-varying confounders were
removed by fitting non-linear time trend using a natural spine function of year with three degrees of
freedom. An additive model was used to estimate the additive effect of PM2.5 on the probability of
dying.
year increased by 3.85 * 10 4
(95% CI 1.95 x 10"4,
5.76 x 10"4) for each 1 |jg/m3
increase in annual PM2.5
concentrations
September 2021
3-113
DRAFT: Do Not Cite or Quote
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied causal modeling
statistical methods in examining long-term PM2.5 exposure and mortality.
Study/Location/Popu la-
tion (Cohort)/Years
Statistical Method
Results
Peterson et al. (2020)
U.S.
National Center for Health
Statistics
1990-2010
Accountability: To determine the portion of the temporal change in cardiovascular mortality
attributable to the temporal change in the annual PM2.5 concentrations, linear models were fitted,
adjusted for time variant covariates (age-standardized annual COPD mortality rates) and time
invariant covariates (median household income, percent of non-White population, and population).
The authors first estimated the overall national temporal trend in annual cardiovascular mortality,
which accounted for the temporal changes in the cardiovascular mortality adjusted for the time
variant and time invariant covariates, but not adjusted by PM2.5. Secondly, the national temporal trend
in the annual PM2.5 concentrations was estimated, while adjusting for the time variant and time
invariant covariates. Then, the association between PM2.5 concentrations and cardiovascular
mortality was estimated, assuming that this association was consistent nationally after adjusting for
the covariates set and for other time-varying changes in cardiovascular morality unrelated to PM2.5
concentrations. The PIVh s-related cardiovascular mortality association was calculated as a product of
the risk to cardiovascular mortality for each unit change in PM2.5 concentration. A county-level
random intercept term was included in the models to account for variation due to repeated measures
from the same county and difference in baseline cardiovascular morality rates.
3.88 (95% CI: 3.56, 4.21)
fewer deaths per 100,000
persons for each 1-|jg/m3
reduction in annual PM2.5
concentrations
DID:
IRR :
difference-in-difference; GPS = generalized propensity score; HR :
incidence rate ratio; IV = instrument variable.
hazard ratio; IPTW = inverse probability treatment weighting; IPW = inverse probability weighting;
September 2021
3-114
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
Wei et al. (2020). Wei et al. (2021b). and Wu et al. (2020a) implemented different generalized
propensity score (GPS) modeling approaches to assess associations between long-term PM2 5
concentrations and mortality among Medicare beneficiaries. GPS statistical approaches were developed as
an extension of propensity score methods for continuous exposures and represent the relative likelihood
of being exposed to the observed pollutant concentration and all measured confounders (Wei et al.. 2020).
The GPS approach estimates the conditional probability of an individual being exposed to the observed
pollutant concentration, accounting for all measured potential confounders (Table A-8).
Wei et al. (2020) evaluated the association of long-term PM2 5 concentrations and all-cause
mortality among Medicare beneficiaries residing in Massachusetts during 2000-2012 using a GPS
approach. The authors define long-term exposure as a 1-year moving average of the PM25 concentration.
Wei et al. (2020) reported that each 1 (ig/m3 increase in long-term PM2 5 concentrations was associated
with 35.4 (95% CI: 33.4, 37.6) excess deaths per 10 million person-days. When the analysis was
restricted to different low-level concentrations, the number of excess deaths associated with a 1 |ig/m3
increase in annual PM25 concentrations increased to 35.5 per 10 million person-days (95% CI: 33.4, 37.7)
when restricting to PM2 5 concentrations <14 |ig/m3 and to 60.7 per 10 million person-days (95% CI: 47.9,
73.9) when restricting to PM2 5 concentrations <7 |ig/m3.
Whereas Wei et al. (2020) focused on a single GPS approach, Wei et al. (2021b) presented three
GPS-based approaches in examining long-term PM2 5 concentrations and mortality including a linear
probability model, weighted least squares, and m-out-of-n random forests (moonRF), for assessing the
effect of long-term PM2 5 exposures on mortality rates among Medicare beneficiaries residing in
Massachusetts between 2000 and 2012. The moonRF method is based on the random forest method,
which is a non-parametric machine learning approach of classification for possible non-linear
relationships and interactions through building individual decision trees through resampling. To reduce
the computational burden of the linear probability model GPS approach, weighted least squares and
moonRF GPS approaches were proposed as alternatives. Wei et al. (2021b) showed that the linear
probability model and the weighted least squares model produced nearly identical results. The annual
number of early deaths associated with 1 |ig/m3 increase in annual PM2 5 concentrations was 1,053 (95%
CI: 984, 1,122) using linear probability model and weighted least squares approaches and was 1,058 (95%
CI: 988, 1,127) using moonRF. When restricting the analysis to annual PM2 5 concentrations below
<12 |ig/m3. 1,203 (95% CI: 1,126, 1,280) annual early deaths were associated with a 1 |ig/m3 increase in
annual PM2 5 concentrations when using both the linear probability model and weighted least squares
approaches with a slightly higher number of early deaths with the moonRF approach (1,214 [95% CI:
1,137, 1,292]).
Similar to Wei et al. (2021b). Wu et al. (2020a) also assessed the association between long-term
PM2 5 concentrations and mortality among Medicare beneficiaries between 2000 and 2016 using multiple
GPS approaches including: (1) matching by GPS; (2) weighting by GPS; and (3) adjustment by GPS,
meaning the GPS was included as a covariate in the health outcome model. Wu et al. (2020a) reported
September 2021
3-115
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
that the three GPS approaches yielded similar results with a 10 |ig/m3 decrease in annual PM2 5
concentrations corresponding to a reduction in mortality rate ranging from 6 to 7% (matching by GPS:
HR= 1.07 [95% CI: 1.05, 1.08]; weighting by GPS: HR= 1.08 [95% CI: 1.07, 1.09]; and adjustment by
GPS: HR= 1.07 (95% CI: 1.06, 1.08]). The estimated hazard ratios were even larger when restricting the
cohort of Medicare enrollees to only those that lived in locations where annual PM2 5 concentrations were
lower than 12 |ig/m3: matching by GPS: HR= 1.26 (95% CI: 1.23, 1.29); weighting by GPS: HR= 1.27
(95% CI: 1.24, 1.30); and adjustment by GPS: HR= 1.23 (95% CI: 1.18, 1.28).
Whereas Wei et al. (2020). Wei et al. (2021b). and Wu et al. (2020a) used different GPS
approaches to examine the association between long-term exposure to PM2 5 and mortality, Wu et al.
(2019) developed a new two-stage GPS approach consisting of a regression calibration-GPS (RC-GPS)
based adjustment for continuous error-prone exposure combined with GPS to adjust for potential
confounding. The new proposed method provides a correction for measurement error in the exposure for
both the design and analysis stages with GPS, allows the GPS implementation to be paired with any
generalized linear model, and shows how standardized bias can be used to assess fit in the context of GPS
analysis for categorical exposures. The RC approach was applied in conjunction with three GPS
approaches: subclassification, inverse probability of treatment weighting (IPTW), and matching. After
constructing GPS for each approach, a stratified log-linear outcome model with a person-time offset
estimated the incidence rate ratio (IRR) of the effect of long-term exposure to PM2 5 on all-cause
mortality. When applying the RC-GPS based adjustment, the authors reported IRRs that were consistent
across the three GPS approaches: RC-GPS subclassification approach: IRR was 1.03 (95% CI: 1.01,
1.05); 1.02 (95% CI: 1.01, 1.04) for the IPTW GPS approach, and 1.03 (95% CI: 1.01, 1.05) for the GPS
matching approach, when comparing moderate levels of PM2 5 concentrations (810 |ig/m3) to
low concentrations (PM2 5<8 |ig/m3). the IRRs were 1.04 (95% CI: 1.00, 1.07) for subclassification, 1.03
(95% CI: 1.01, 1.06) for IPTW, and 1.04 (95% CI: 1.02, 1.06) for matching.
In addition to GPS, other causal modeling methods have been employed in epidemiologic studies
including inverse probability weighting (IPW). IPW generates weights by taking the inverse of the
conditional probability of the exposure of a pollutant concentration. The weight is then stabilized by
multiplying the weights by the marginal probability of the level of exposure, meaning that by applying
weights the exposure is no longer associated with the confounders (Higbcc et al.. 2020; Awad et al..
2019).
Schwartz et al. (2018b) and Awad et al. (2019) both applied IPW methods to estimate the effect
of long-term exposure to PM2 5 on mortality among Medicare beneficiaries, while controlling for similar
confounders (Table A-8). Schwartz et al. (2018b) estimated the marginal effect of annual PM2 5
concentrations on the distribution of life expectancy among Medicare beneficiaries residing in the
Northeastern and Mid-Atlantic region of the U.S. between 2000-2013 by applying an IPW survival
model. The estimated mean age at death for a population with an annual average PM2 5 concentration of
September 2021
3-116
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
12 |ig/m3 was 0.89 (95% CI: 0.88, 0.91) years less than estimated for a counterfactual PM2 5
concentrations of 7.5 |ig/m3. Schwartz et al. (2018b) estimated that 23.5% of the Medicare population
would die before 76 years of age if they were exposed to an annual PM2 5 concentration of 12 |ig/m3
compared with 20.1% if the Medicare population was exposed to an annual average PM2 5 concentration
of 7.5 |ig/m3. Furthermore, the authors estimated that 40.8% of Medicare recipients would live past
85 years of age if exposed to an annual PM2 5 concentration of 12-|ig/m3 PM2 5 compared with 44.5% at
7.5-(ig/m3.
A wad et al. (2019) estimated the effect of a change in annual PM2 5 concentrations due to moving
on the risk of mortality among Medicare beneficiaries from 2000 to 2012 in the U.S. from Cox
proportional hazards using an IPW approach to control for potential confounding (Table A-8). The Cox
proportional model estimated the effect of change in PM2 5 concentrations on the risk of all-cause
mortality, stratified by ZIP code before moving, with the IPW to account for confounding after moving
and follow-up time. Awad et al. (2019) estimated a HR of 1.21 (95% CI: 1.20, 1.22) among White
individuals and 1.12 (95% CI: 1.08, 1.15) among Black individuals for a 10 |ig/m3 increase in annual
PM2 5 concentrations for all-cause mortality. When restricting the analysis to movers with PM2 5
concentration <12 |ig/m3. the hazard ratio was 1.25 (95% CI: 1.24, 1.27) for White individuals and 1.08
(95% CI: 1.01, 1.14) for Black individuals for all-cause mortality.
Wei et al. (2021a) used an IPW approach to emulate a dose-response between PM2 5 and all-cause
mortality from Medicare beneficiaries between 2000-2016. This newer IPW approach used a decile
binning, which divided PM2 5 concentrations into ten equally sized deciles and predicted the inverse
probability of being assigned to the observed group for each observation, adjusting for copollutants
(i.e., ozone and N02) and other covariates (Table A-8). The lowest decile group is treated as the reference
with effects estimated for the other decile groups compared to the reference. The relative risk of all-cause
mortality associated with long-term exposure to PM2 5 ranged from 1.02 (95% CI: 1.02, 1.03) at
6.60 |ig/m3 (2nd decile group) to 1.21 (95% CI: 1.20, 1.21) at 15.47 |ig/m3 (10th decile group). Assuming
that the IPW models were correctly specified and the counterfactual framework is valid, the
dose-response curves demonstrated that in general, higher concentrations of PM2 5 are associated with a
greater risk of all-cause mortality (Wei et al.. 202 la). While the previous studies all used the IPW
approach in analyses of Medicare beneficiaries, Higbee et al. (2020) examined the association between
long-term exposure to PM2 5 and all-cause and cardiopulmonary mortality from the National Health
Interview Survey from 1986 to 2015. Within this study the authors applied a series of Cox proportional
hazards models, adjusted using IPW. The hazard ratio for all-cause mortality was 1.12 (95% CI: 1.08,
1.15) per 10 |ig/m3 increase in PM2 5 concentration and for cardiopulmonary mortality, it was 1.23 (95%
CI: 1.17, 1.29) per 10 |ig/m3 increase in PM2 5 concentration.
In addition to GPS and IPW, additional epidemiologic studies used a difference-in-difference
(DID) approach to control for unmeasured confounders. In this method, the mean exposure is calculated
for exposed and non-exposed groups between two time periods, such as pre- and post-intervention
September 2021
3-117
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
(Schwartz et al.. 2021; Yitshak-Sade et al.. 201%). The predictors of the outcome, such as socioeconomic
status, education, and smoking status, are the same in each group in both time periods, therefore,
theoretically the difference between outcomes in the two time periods in the exposed group cannot be
confounded by those predictors (Schwartz et al.. 2021).
To account for changes in PM2 5 concentrations due to policy or the implementation of an
intervention, Corrigan et al. (2018). Sanders et al. (2020). Henneman et al. (2019). and Fan and Wang
(2020) used DID methods to assess whether there was evidence of changes in associations with mortality
due to changes in annual PM2 5 concentrations. Corrigan et al. (2018) examined whether there was a
change in the cardiovascular mortality rate before (2000-2004) and after (2005-2010) implementation of
the first annual PM2 5 NAAQS in 2005 based on mortality data from the National Center for Health
Statistics (Table A-8). The authors reported 1.10 (95% CI: 0.37, 1.82) fewer cardiovascular deaths per
year per 100,000 people for each 1 |ig/m3 reduction in annual PM2 5 concentrations. When comparing
whether counties achieved NAAQS compliance (attainment) or not (non-attainment), there were 1.96
(95% CI: 0.77, 3.15) fewer cardiovascular deaths for each 1 |ig/m3 reduction in annual PM2 5
concentrations between the two periods for attainment counties, whereas for non-attainment counties,
there were 0.59 (95% CI: -0.54, 1.71) fewer cardiovascular deaths between the two periods. Sanders et
al. (2020) conducted a similar study as Corrigan et al. (2018) by attempting to isolate the relationship
between regulation of and reductions in PM2 5 concentrations and associated mortality from Medicare
beneficiaries between 2000-2013 (Table A-8). After the release of the first annual PM2 5 NAAQS
implementation in 2005, annual PM2 5 concentrations decreased by 1.59 |ig/m3 (95% CI: 1.39, 1.80)
which corresponded to a reduction in mortality rates among individuals 65 years and older of 0.93% (95%
CI: 0.10, 1.77).
While Corrigan et al. (2018) and Sanders et al. (2020) used DID to examine the relationship
between changes in annual PM2 5 concentrations and mortality due to the PM2 5 NAAQS, Henneman et al.
(2019) and Fan and Wang (2020). applied DID methods to explore the changes in PM2 5 concentrations
following retirement of coal-fueled power plants and mortality. Henneman et al. (2019) conducted an
accountability analysis of emissions reductions from coal-fueled power plants in the U.S. between
2005-2012 and whether there were corresponding reductions in all-cause mortality using Medicare data
(Table A-8). The authors reported a reduction in all-cause mortality of-0.38 (95% CI: -2.76, 2.01)
deaths per 10,000 person-days for each 1 |ig/m3 reduction in annual PM2 5 concentrations during the time
period. Fan and Wang (2020) also used Medicare data from 1999-2013 to estimate the relationship
between mortality and low PM2 5 concentrations in response to the retirement of five large coal plants by
applying both instrumental variable and DID approaches (Table A-8). The instrumental variable
represents variations in the exposure (e.g., PM2 5) that are randomized with respect to both measured and
unmeasured confounders and can therefore provide an estimate of the effect of the exposure (Schwartz et
al.. 2018a). The authors reported that a 1 -|ig/m3 reduction in annual PM2 5 concentrations corresponded to
7.17 fewer deaths per 100,000 people per month, or a 1.7% lower monthly mortality rate among people
older than 65 years of age when applying an instrument variable approach. The power plant retirement
September 2021
3-118
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
decreased both monthly PM2 5 concentrations by 2.1 |ig/m3. and the monthly age-adjusted mortality by
approximately 15 people per 100,000 people (or 3.6%) in treated counties, relative to control counties,
based on the DID approach.
Yitshak-Sade et al. (201%) and Schwartz et al. (2021) also used DID approaches to evaluate the
changes in long-term exposure to PM25 overtime on mortality rates among Medicare beneficiaries.
Yitshak-Sade et al. (201%) applied a DID approach to the Medicare population of the Northeastern and
mid-Atlantic states to incorporate individual covariates and assess the effect of annual PM2 5
concentrations on all-cause mortality rates from 2000 to 2013 (Table A-S). For an IQR (3 |ig/m3) increase
in annual PM25 concentrations, the authors reported a 4.04% (95% CI: 3.49, 4.59) increase in mortality
rates. In a sensitivity analysis, the mortality effect was modified by eligibility to Medicaid insurance and
race, with larger associations among people who are eligible to receive Medicaid services (5.99% [95%
CI: 4.38, 7.62] and among Black individuals (10.10% [95% CI: 8.56, 11.67], Whereas Yitshak-Sade et al.
(2019b) applied DID to a portion of Medicare beneficiaries.Schwartz et al. (2021) applied two DID
approaches to assess whether changes in PM2 5 are associated with changes in mortality rates nationally
among Medicare participants between 2000 and 2016 (Table A-8). The authors reported that the
probability of dying in each year increased by 3.85 x 10 4 (95% CI: 1.95 x 10 4. 5.76 x 10 4) for each
1 |ig/m3 increase in PM2 5 in that year. When the analysis was restricted to those beneficiaries residing in
locations where PM2 5 concentrations are below 12 |ig/m3 during the follow-up period, the probability of
dying in each year increased to 4.26 x 10 4 (95% CI 1.43 x 10 4. 7.09 x 10 4) for 1 (ig/m3 increase in
annual PM2 5 concentrations.
Lastly, Peterson et al. (2020) conducted an accountability analysis to examine which portion of
the observed declining trend in cardiovascular mortality from the National Center for Health Statistics
between 1990 and 2010 was associated with changes in ambient PM2 5. To determine the portion of the
temporal change in cardiovascular mortality attributable to the temporal change in the ambient
concentration of PM2 5, Peterson et al. (2020) fit linear models, and adjusted for time-variant covariates
(age-standardized annual COPD mortality rates) and time-invariant covariates (median household
income, percent of non-White population, and population). On average, each 1 -|ig/m3 reduction in annual
PM2 5 concentrations was associated with 3.88 (95% CI: 3.56, 4.21) fewer deaths per 100,000 persons.
When combined with the annual decline in PM2 5 concentrations, the PM2 5-trend accounted for 0.52 (95%
CI: 0.48, 0.57) fewer deaths per 100,000 persons each year, for a total change of 10.44 (95% CI: 9.56,
11.32) fewer deaths per 100,000 persons over the entire study period.
Overall, recent epidemiologic studies employed a variety of causal modeling methods such as
GPS, IPW, and DID and reported consistent results among large study populations across the U.S. These
causal modeling methods in combination with accountability analyses further informs the relationship
between long term PM2 5 exposure and total mortality and supports the conclusions of the 2019 PM ISA.
September 2021
3-119
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
3.2.2.4 Summary of Recent Evidence in the Context of the 2019 Integrated
Science Assessment for Particulate Matter Causality Determination
for Long-Term PM2.s Exposure and Mortality
Recent epidemiologic studies published since the 2019 PM ISA support and extend the evidence
base that contributed to the conclusion of a causal relationship between long-term PM2 5 exposure and
mortality. Numerous cohort studies conducted in the U.S. and Canada in locations with mean annual
concentrations that, in many instances, are lower than those reported in studies evaluated in the 2019 PM
ISA, add to the large evidence base indicating consistent, positive associations between long-term PM2 5
exposure and mortality. These positive associations were reported across studies using different: cohorts,
exposure windows, approaches for confounder adjustment, and exposure assessment methods that used
different sources of data and were conducted at different spatial resolutions. In addition, studies
examining cause-specific mortality further expand upon the studies that reported consistent, positive
associations with cardiovascular- and respiratory-related mortality as well as other mortality outcomes.
Recent studies also provide some of the initial evidence demonstrating that individuals with preexisting
health conditions (i.e., heart failure and diabetes) are at increased risk of mortality overall, which, until
recently, was primarily examined in studies using stratified analyses, rather than a cohort of individuals
with an underlying health condition.
Additional support for a relationship between long-term PM2 5 exposure and mortality stems from
studies that examined the influence of potential confounding bias. While there is some evidence of
potential confounding of the PM2 5-mortality association by copollutants within the MAPLE studies, this
is not consistent with other recent studies evaluated in the U.S. and Canada that reported associations
consistent in magnitude in both single and copollutant models or with the studies evaluated in the 2019
PM ISA. In addition to copollutants, a few studies examined whether additional potential confounders,
such as temporal trends and meteorological variables, could explain the PM2 5-mortality association.
Analyses examining these additional covariates, which have previously been hypothesized to be
confounders of the association, further confirm that the relationship between long-term PM2 5 exposure
and mortality is unlikely to be biased by these factors.
Consistent with the conclusions of the 2019 PM ISA, recent studies provide evidence that
continues to support a generally linear, no-threshold C-R relationship for long-term PM2 5 exposure and
all-cause or cause-specific mortality. Recent studies extend the evidence base by using novel statistical
techniques for estimating the C-R relationship, by examining the C-R relationship among populations
with relatively low PM2 5 concentrations, and by evaluating cause-specific mortality in addition to
all-cause mortality. A limited number of recent studies add to and support the evidence from the 2019 PM
ISA by examining the temporal trends in PM2 5 concentrations and changes in life expectancy. These
studies generally observed that higher exposures to PM2 5 concentrations are associated with decreases in
life expectancy.
September 2021
3-120
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Lastly, a number of recent studies employed causal modeling methods or conducted
accountability analyses in the process of examining the relationship between long-term PM2 5 exposure
and mortality. These studies which used different statistical approaches in a causal modeling framework
in combination with accountability analyses that examined the impact of policies or interventions on
PM2 5 concentrations and mortality, collectively provide additional support for the consistent positive
associations between long-term PM2 5 exposure and mortality reported in cohort studies spanning diverse
geographic locations and populations.
3.3 Key Scientific Topics that Further Inform the Health
Effects of PM2.5
This section evaluates recent studies that address specific scientific topics that further inform the
relationship between PM2 5 exposure and health and are relevant to consider in the process of
reconsidering the PM NAAQS. The topics covered within this section include recent experimental studies
conducted at near-ambient concentrations, which can further inform the biological plausibility of health
effects at the ambient concentrations reported in epidemiologic studies (Section 3.3.1); the role of PM2 5
exposure on COVID-19 infection and death (Section 3.3.2); and an evaluation of studies that examine
PM2 5 exposure and health risk disparities among racial and ethnic groups and socioeconomic status (SES)
(Section 3.3.3V
3.3.1 Recent Experimental Studies at Near-Ambient Concentrations
Few controlled human exposure studies have investigated the effects of exposure to near ambient
concentrations of PM2 5, i.e., at or below the 24-hour NAAQS PM2 5 standard of 35 ug/m3. As discussed in
Sections 6.1.10,6.1.11, and 6.1.13 of the 2019 PM ISA, one such study was conducted in Copenhagen
(Hemmingsen et al.. 2015b; Hemmingsen et al.. 2015a). In this study, older (55-83 years) overweight
study participants (n = 60) were exposed for 5 hours to urban street air or filtered urban street air while at
rest. The average PM2 5 concentration was 24 ug/m3. The study was a randomized, repeated measures,
single blinded cross-over study. One arm of the study reported decreased vasomotor function immediately
(within 1 hour) after exposure. The decrease in nitroglycerin-mediated vasodilation was statistically
significant, while the decrease in reactive hyperemia-induced vasodilation was not. These two responses
represent endothelium-independent and endothelium-dependent mechanisms, respectively. No changes in
blood biomarkers of oxidative stress or inflammation were observed. Similarly, blood pressure, blood
lipids, and metabolic biomarkers were unaffected by exposure. However, decreases in heart rate
variability (HRV) were observed in participants exposed to nonfiltered street air, with a statistically
significant reduction in the high frequency domain and a statistically significant increase in the low
frequency domain. In the other arm of the study, no oxidative stress or DNA damage was found in
peripheral blood monocytes of participants exposed to unfiltered street air.
September 2021
3-121
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
A recent study conducted at near ambient PM2 5 concentrations by Wvatt et al. (2020a') adds to
this limited evidence base of controlled human exposure studies conducted at near ambient
concentrations. This study was a randomized double-blind crossover study in healthy young participants
(18-35 years, n = 21) who were subject to intermittent moderate exercise. Study participants were
exposed for 4 hours to clean air or to an average concentration of 37.8 ug/m3 PM2 5 CAPs in Chapel Hill,
NC. Ventilation rate was monitored, and workload adjusted so that participants achieved a ventilation rate
of 20 L/min/m2 body surface area. Thus, dose was approximately the same among participants. Changes
in lung function were observed in PM2 5 exposed participants, including a statistically significant decrease
in FEV1/FVC of 1.2% at 1-hour postexposure that returned to baseline by 20 hours postexposure.
Decreases in PEF (1.8%) and FEV1 (0.8%) were also observed 1-hour postexposure, but they did not
achieve statistical significance. Markers of the acute phase response—serum amyloid A and C reactive
protein—were increased at both 1 and 20 hours postexposure. Serum amyloid A was increased by 8.7% at
1-hour postexposure and by 34.6% at 20 hours postexposure. C reactive protein was increased by 9.1% at
1-hour postexposure and by 22.8% at 20 hours postexposure. These changes, except for C reactive protein
at 20 hours postexposure, were statistically significant. Other statistically significant changes observed
were increases in markers of vascular inflammation, sICAM (10.7%) and sVCAM (6.6%), at 1-hour
postexposure, decreases in percent blood neutrophils (5.7%) at 20 hours postexposure and decreases in
blood LDH levels of 6.9 and 11.2% at 1 and 20 hours postexposure, respectively. The percentage of white
blood cells other than neutrophils were unchanged by exposure, as were cytokines, red blood cells, and
measures of blood chemistry. Hematocrit levels were decreased, but this change failed to reach statistical
significance. No statistically significant changes were found in measures of HRV in the study population
overall, but sex-specific changes were noted for men and women, albeit they were in opposite directions.
P-wave duration, a measure of cardiac repolarization, was increased by 10.5% at 1-hour postexposure, a
statistically significant change. Another measure of cardiac repolarization, QRS complex, was altered by
PM2 5 exposure, but this change did not reach statistical significance. Some sex-related changes that
reached statistical significance were found in HRV measures, cardiac repolarization parameters, FEV1
and PEF, however the small sample size precludes any conclusions.
The higher ventilation rate and longer exposure duration of Wvatt et al. (2020a) compared to
most controlled human exposure studies of PM, is roughly equivalent to a 2-hour exposure of
75-150 (ig/m3 PM2 5. Thus, dosimetric considerations may explain the observed changes in lung function
and inflammation in this population of young healthy individuals exposed to near-ambient concentrations
of PM2 5. While Wvatt et al. (2020a) provides evidence of some effects at lower PM2 5 concentrations,
overall there is inconsistent evidence for changes in lung function (2019 PM ISA, Section 5.1.7.2 and
Section 5.1.2.3.3) and inflammation (2019 PM ISA, Section 6.1.11.2.1) in other controlled human
exposure studies evaluated in the 2019 PM ISA. Strengths of Wvatt et al. (2020a) include power
calculations made based on the primary endpoints, normalization of results, and good diurnal control
(i.e., all exposures occurred within a half hour of the same time—9:30 a.m.). However, a limitation of this
study is that no Bonferroni corrections were made to account for the multiple comparisons made in the
study; this limitation was acknowledged by the study authors.
September 2021
3-122
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
3.3.2
PM2.5 Exposure and COVID-19 Infection and Death
With the advent of the global Coronavirus Disease 2019 (COVID-19) pandemic, several recent
studies examined the role of ambient air pollution, specifically PM2 5, on COVID-19 infections and
deaths, including a few with study locations in the U.S. and Canada. The following sections evaluate
studies that examined the relationship between short-term (Section 3.3.2.1) and long-term
(Section 3.3.2.2) PM25 exposure and COVID-19, including infections, hospital admissions, replication
rate, and deaths. While there is no exact corollary within the 2019 PM ISA for these types of studies,
there is evidence presented in the 2019 PM ISA which evaluates the potential relationship between
short-term and long-term PM25 exposure and respiratory infection (2019 PM ISA, Section 5.1.5 and
Section 5.2.6). Briefly, studies outlined in the 2019 PM ISA reported some evidence of positive
associations between short-term PM2 5-exposure and emergency department (ED) visits and hospital
admissions for respiratory infections, but interpretation of these results was complicated by the variability
in the type of respiratory infection outcome examined (2019 PM ISA, Figure 5-7). Studies of long-term
PM2 5 exposure were limited in number and although some positive associations were reported there was
minimal overlap in respiratory infection outcomes examined across studies. As detailed in Section 5.1.1
and 5.1.5 of the 2019 PM ISA, exposure to PM25 has been shown to impair host defense, specifically
altering macrophage function, providing a biologically plausible pathway by which PM2 5 exposure could
lead to respiratory infection. Additionally, there is some evidence that exposure to PM2 5 can lead to
decreases in an individual's immune response and can subsequently facilitate replication of respiratory
viruses (Bourdrel et al.. 2021).
3.3.2.1 Short-Term PM2.5 Exposure
Since the onset of the COVID-19 pandemic, studies have evaluated the relationship between
short-term PM2 5 exposure and COVID-19 outcomes (Table A-9). Specifically, these studies examine
whether or not daily-changes in PM2 5 can influence COVID-19 outcomes. A recent study conducted in
Queens County, NY evaluated the relationship between short-term PM2 5-exposure and incident
COVID-19 infections and deaths between March 1, 2020 and April 20, 2020 (Adhikari and Yin. 2020).
This time frame corresponds to the timing of the first wave of the pandemic in Queens County, NY
(https://wwwl.nvc.gov/site/doh/covid/covid-19-data-trends.page). This study used negative binomial
regression to independently model PM2 5 collected from stationary monitors, as well as several other
meteorological factors to predict new COVID-19 cases and deaths, controlling for the lagged outcome
and a trend for day. Using a 21-day moving average of PM25 exposure, the authors identified a null
association between PM25 concentrations and increased risk of COVID-19 infection (Incidence Rate
Ratio [IRR]: 0.02 [95% CI: 0.01, 0.02] or death [IRR: 0.32 (95% CI: 0.10, 0.97)]).
September 2021
3-123
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
3.3.2.2 Long-Term PM2.5 Exposure
Several recent studies have evaluated long-term PM2 5-exposure and COVID-19 outcomes in
North America (Chakrabartv et al.. 2021; Mendv et al.. 2021; Liang et al.. 2020; Stieb et al.. 2020; Wu et
al.. 2020b). These studies evaluated whether chronic PM2 5 exposure is related to increased susceptibility
to COVID-19 outcomes. When considered together, these studies that examined the relationship between
long-term PM2 5 exposure and the global COVID-19 pandemic have several methodological limitations
(e.g., studies conducted during an ongoing pandemic) and as a result caution is warranted when
interpreting results (Table A-10).
Two large ecological studies (Liang et al.. 2020; Wu et al.. 2020b) evaluated the association
between long-term PM2 5 concentrations and county-level COVID-19 deaths in the U.S. The study by Wu
et al. (2020b) evaluated COVID-19 death rates (ratio of COVID-19 deaths to county-level population) in
3,089 U.S. Counties through June 18, 2020. At that point in the COVID-19 pandemic, over 40% of U.S.
counties had 0 cases. The Mortality Rate Ratio (MRR) was predicted using a negative binomial regression
using the 17-year average (2000-2016) of PM2 5 as the main exposure and controlling for 20 covariates
(19 count-level, 1 state-level). Overall, the authors reported an increased risk of COVID-19 mortality
(MMR: 1.69 [95% CI: 1.34, 2.19]). This result was validated by the use of over 80 sensitivity analyses.
These sensitivity analyses included alternative methods by which to estimate PM2 5 exposure, model
specifications, transformation of confounding variables, and repeating the analysis daily between April
18, 2020 and June 18, 2020 to evaluate temporal changes. Another ecologic study by Liang et al. (2020).
evaluated the mortality ratio (MR) (COVID-19 deaths per 1 million population) and the case fatality rate
(CFR) (ratio of COVID-19 deaths to COVID-19 cases) within 3,122 U.S. counties between January 22,
2020 to July 17, 2020. At that point in the pandemic there were still over 30% of U.S. counties reporting
0 cases (https://usafacts.org/visualizations/coronavirus-covid-19-spread-map/). This study also relied on
negative binomial regression but used 7-year average (2010-2016) PM2 5 as the main exposure of interest
while controlling for county-level characteristics and meteorology. The authors estimated the MR to be
1.40 (95% CI: 1.08, 1.83), and CFR was estimated to be 1.18 (95% CI: 0.92, 1.49). When including either
NO2 or O3 in the model, the results remained consistent (Figure 3-29).
September 2021
3-124
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Case-fatalityt
# !
to *g
<0
T, c
CU
W O
3 S
20
O
Q CTJ
II
C c
V> Qj
10
C %
oj Cf
o —
-10
Single-pollutant
Air Pollutants
Bi-pollutant ' Tri-pollutant
NOs # Ozone PMzs
B Mortality
c 40
.2
"ffl c
C <1>
o "
2 s
Q CO
I!
30
20
c %
oj O
o —
•£.£ 10
-10
I
i
Single-pollutant
Bi-pollutant
Tri-pollutant
Source: Liang et al. (2020), copyright permission pending.
Figure 3-29 Percent change in county-level COVID-19 case fatality rate and
mortality rate in single and multipollutant models (January 22,
2020-July 17, 2020).
In contrast, a Canadian ecologic study by Stieb et al. (2020) evaluated COVID-19 infections
within 111 Canadian Health Regions through May 18, 2020. These data were used to estimate the
incidence rate ratio (IRR) of COVID-19 infections and the 17-year average (2000-2016) of PM2.5. This
analysis used negative binomial regression and controlled for several health region-specific demographics
and descriptors as well as meteorology. The authors reported null associations between long-term PM2.5
exposure and COVID-19 infections (IRR: 1.40 [95% CI: 0.86, 2.29]).
A small U.S. study conducted by Mendv et al. (2021) examined the relationship between
COVID-19 hospitalizations, among individuals with a COVID-19 infection, and 10-year average
(2008-2017) PM2.5 exposure, measured at the ZIP code level. This cross-sectional study within the
University of Cincinnati Hospital System considered several different individual-level factors such as
self-reported age, race/ethnicity, and smoking status, along with other co-morbidities abstracted from
medical records. However, this study did not consider community-based factors, such as population
density. Mendv et al. (2021) observed that among individuals with a COVID-19 infection, those with
COPD or asthma had increased odds of hospitalization associated with long-term PM2.5 exposure (OR:
11.16 [95% CI: 1.00, 128.24]) compared to those without COPD or asthma (OR: 0.42 [95% CI: 0.12,
1.54]).
In a departure from examining solely COVID-19 infections, deaths, or hospitalizations, another
recent study evaluated the association between 6-year average PM2.5 concentrations (2012-2017) and the
COVID-19 reproduction ratio (R0) for each state (Chakrabartv et al.. 2021). The R0 refers to the
approximate number of individuals that an infected individual would infect in a completely susceptible
population. The R0 was calculated for the time between March 2, 2020 and April 30, 2020, which
coincided with nationwide stay-at-home orders. Using a susceptible-exposed-infected-recovered (SEIR)
September 2021
3-125
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
model, confirmed COVID-19 cases were recorded by state and R0 was predicted, while controlling for
over 40 different state-level variables. This study showed that long-term PM2 5 exposure was associated
with increases in the Ro (ARO: 1.25 [95% CI: 0.24, 2.24]).
3.3.2.3 Summary of Recent Epidemiologic Studies Examining PM2.5
Exposure and COVID-19 Infection and Death
The body of evidence that examined associations between long- and short-term PM2 5-exposure
and COVID-19 outcomes consists mostly of studies employing an ecological study design. While some of
these studies exploring the relationship between PM2 5 concentrations and COVID-19 infections and
deaths reported positive associations, there are a number of methodological issues that may influence
results. Specifically, all of these studies were conducted in the midst of an ongoing pandemic before
COVID-19 had reached many parts of the country and at that time the etiology of COVID-19 was still not
well understood. However, recent investigations have noted important differences in COVID-19 related
health outcomes based on occupation, race, socioeconomic status, and health insurance status, among
others.
While all the included studies account for population density, with the exception of Mendv et al.
(2021). they mostly do not consider other crucial factors which may strongly influence the spread of
COVID-19. In particular, factors such as social distancing, stay-at-home orders, use of N95 masks, as
well as other preventative measures are important for slowing the spread of SARS-CoV-2 (the virus
responsible for COVID-19 infection). Two critiques of the PM2 5 exposure and COVID-19 literature
conducted by Bourdrel et al. (2021) and Villeneuve and Goldberg (2020) highlight the importance of
control for race/ethnicity and other sociodemographic factors, which may strongly influence exposure or
susceptibility to SARS-CoV-2. Additionally, they indicate the potential for exposure misclassification and
the likelihood of underreporting of cases and deaths, particularly in the early stages of the pandemic.
Taken together, there is limited evidence at this point in the COVID-19 pandemic to determine if short- or
long-term exposure to air pollutants, such as PM2 5, influence the spread or susceptibility of SARS-CoV-2
in the population.
3.3.3 Populations and Lifestages at Potentially Increased Risk of a
PM-Related Health Effect
As discussed in the 2019 PM ISA in Chapter 12, the NAAQS are intended to protect public health
with an adequate margin of safety, which includes protection for the population as a whole and for those
groups potentially at increased risk for health effects from exposure to a criteria air pollutant [e.g., PM;
see Preamble to the ISA (U.S. EPA. 2015)1. While there is strong evidence for PM-related health effects
occurring in the exposed general population and in some specific populations or lifestages, it is also
important to evaluate and characterize the evidence to determine whether there are populations or
September 2021
3-126
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
lifestages potentially at increased risk of a PM-related health effect, with specific emphasis placed on
studies that compare responses to a reference population, where appropriate [see Preamble to the ISA
(U.S. EPA. 2015)1.
As discussed in the Preamble to the ISAs (U.S. EPA. 2015). the risk of health effects from
exposure to an ambient air pollutant, including PM, may be modified as a result of intrinsic
(e.g., preexisting disease, genetic factors) or extrinsic factors (e.g., sociodemographic or behavioral
factors), differences in internal dose (e.g., due to variability in ventilation rates or exercise behaviors), or
differences in exposure to air pollutant concentrations (e.g., more time spent in areas with higher ambient
concentrations). Taking into consideration each of these factors, Chapter 12 of the 2019 PM ISA
conducted a full evaluation and characterization of the evidence and conveyed the overall confidence
(i.e., adequate evidence, suggestive evidence, inadequate evidence, or evidence of no effect) as to whether
specific populations or lifestages are at increased risk of a PM-related health effect (see Section 2.2.5)
using a framework detailed in Table 12-1 of the 2019 PM ISA. As a result, this Supplement does not
include a full evaluation and characterization of all studies published since the literature cutoff date of the
2019 PM ISA that provided evidence as to whether specific populations and lifestages are at increased
risk of a PM-related health effect. Instead, given recent Agency guidance addressing environmental
justice [e.g., U.S. EPA (2021)1 and the expansion of studies examining the role of PM25 on populations
with environmental justice concerns, the focus in this section is on recent studies that examine disparities
in exposure or risk to PM by socioeconomic status (Section 3.3.3.1) and race and ethnicity
(Section 3.3.3.2).
3.3.3.1 Socioeconomic Status
The 2019 PM ISA noted that socioeconomic status (SES)—a composite measure that can include
metrics such as income, education, or occupation—plays a role in access to healthy environments and
access to healthcare in the U.S., therefore indicating that SES may underlie differential risk for
PM2 5-related health effects. Additionally, some evidence demonstrated that having low income or
residing in low-income areas results in stronger associations (i.e., larger in magnitude) between mortality
and long-term PM2 5 exposures, when compared with their higher income counterparts. When considering
educational attainment, as a proxy for SES, there was no clear pattern of differential risk when comparing
those with low educational attainment and those with higher educational attainment. Taken together, the
2019 PM ISA concluded that the combination of exposure disparities and health evidence was suggestive
that lower SES populations are at increased risk for PM2 5-related health effects compared to higher SES
populations. The following sections evaluate recent studies pertaining to both PM2 5 exposure among
different SES groups (Section 3.3.3.1.1) and PM2 5-related health risks among different SES groups
(Section 3.3.3.1.2).
September 2021
3-127
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
3.3.3.1.1
Exposure Disparity
Several recent studies within the U.S. and Canada evaluated the relationship between PM2 5
exposure and SES since the completion of the 2019 ISA (Lee and Park. 2020; Richmond-Bryant et al..
2020; Son et al.. 2020; Lee. 2019; Tanzer et al.. 2019; Weaver etal.. 2019; Rosofskv et al.. 2018; Han et
al.. 2017). These recent studies add to the initial conclusions drawn within the 2019 ISA, that there is a
disparate PM2 5 exposure among lower SES communities. When considered together, these additional
studies provide further evidence that lower SES communities are exposed to higher concentrations of
PM2 5 compared to higher SES communities (Figure 3-30—SES exposure, Table A-1 1).
Educational attainment is a common metric in which to assess the SES of a community. A study
by Lee (2019) in the state of California, observed that census block-groups with a high percent (>75th
percentile) of low educational attainment had a higher exposure to PM2 5 (9.7 (ig/m3) compared to those
with a low percent (<25th percentile) of low educational attainment (8.7 (.ig/ni3). Rosofskv et al. (2018)
conducted a study in Massachusetts that examined exposure differences by educational attainment for
both 2003 and 2010, but did not find as substantial of an exposure difference by educational attainment.
Specifically, census block groups with more individuals with less than high school education were
exposed to PM25 concentrations of 11.3 (ig/m3 in 2003 and 2010: 8.2 (ig/m3 in 2010; whereas those with a
Master's degree or higher were exposed to 11.2 (ig/m3 in 2003 and 8.0 (ig/m3 in 2010. Additionally, those
with a high school education, post-secondary education, or Bachelor's degree were exposed to slightly
lower PM2 5 concentrations ranging from 11.1-11.2 (ig/m3 in 2003 and 7.9-8.0 (ig/m3 in 2010 (Rosofskv
et al.. 2018).
Measures of poverty or household income are other indicators that can be used to assess SES in a
community. Median household income was also evaluated in assessing potential exposure disparities in
the study conducted by Rosofskv et al. (2018) noted above. Within this study it was observed that census
block groups with a median household income categories with ranges <$75,000 were exposed to slightly
higher PM2 5 concentrations of PM2 5 in 2003 11.2-11.4 (ig/m3 and 2010 8.0-8.2 (ig/m3 compared to
block groups with median household income >$75,000 (11.1 (ig/m3 in 2003 and 7.9 (ig/m3 in 2010).
Another study evaluating income differential in North Carolina, provided some evidence that those living
in census tracts with a median household income >$52,269 (75th percentile) (11.3 (.ig/ni3) are exposed to
slightly higher concentrations of PM2 5 compared to census tracts above the 75th percentile (11.2 (.ig/ni3)
for median household income (Son et al.. 2020). Instead of examining specific household income
cutpoints, Lee (2019) specifically evaluated the percent living in poverty (defined as those living two
times below the poverty line in the state of California). In 2016, those living in census block groups in
poverty (9.7 (.ig/ni3) were shown to be exposed to higher concentrations of PM2 5 compared to block
groups not in poverty (8.7 (.ig/ni3).
A study by Richmond-Bryant et al. (2020) using data from the National Emissions Inventory
(NEI), evaluated potential changes in PM2 5 burden due to the closure of 92 coal-fired electricity
generating units (EGUcf) facilities in the U.S. This study was an extension of the study presented in the
September 2021
3-128
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
2019 PM ISA by Mikati et al. (2018). The authors demonstrated that census tracts below the poverty line
were subject to a slightly greater decrease (8.8%) in EGUcf PM25 emissions compared to census tracts
not below the poverty line (8.5%). However, when evaluating proportional burden, or the ratio of burden
among each subgroup to the burden among the entire population, there was less than a 1% change for
those above vs. below the poverty line. This study, while not including all sources of ambient PM2 5,
generally indicates that census tracts below the poverty line experience a greater burden of exposure to
PM2.5.
Several recent studies evaluated composite neighborhood level SES characteristics and in
assessing potential disparities in PM2 5 exposure. A study conducted in Houston, TX observed higher
PM2 5 concentrations obtained from stationary monitors, in a single ZIP code with low SES characteristics
(lower median household income, low percentage of non-Hispanic (NH)-White populations, and high
percentages of NH-Black and Hispanic populations) (11.3 (.ig/ni3) compared to a ZIP code with high SES
characteristics (9.6 (ig/m3) (Han et al.. 2017). Similarly, a study by Lee and Park (2020) in the state of
California, compared vulnerable to less vulnerable communities based on the Social and Health
Vulnerability (SHV) metric within the Environmental Justice Screening Method. The SVH score is on a
scale between 1-5 and is a composite of: percent residents of color, percent of residents twice below the
national poverty line, home ownership, housing value, educational attainment, biological vulnerability
(percent of residents <5 and >60 and birth outcomes), and civic engagement (linguistic isolation and voter
turnout). Low vulnerability communities (SVH score 1-2) (6.8 (ig/m3) had lower PM2 5 exposure,
compared to communities with higher vulnerability (SVH score 4-5) (10.8 (.ig/ni3). Additionally, Tanzer
et al. (2019) reported that Environmental Justice (EJ) communities, as defined by the state of
Pennsylvania as census tracts with >20% of the population living below the poverty line and/or >30% of
the populations belonging to a minority group, had a slightly greater exposure to PM2 5 (10.6 (ig/m3)
compared to non-EJ communities (10.3 (.ig/ni3). A study by Weaver et al. (2019) evaluated certain
neighborhood clusters located within three counties (Durham, Orange, and Wake) in Central North
Carolina. Cluster 1 was defined as urban and having a high percent of Black residents and nonmanagerial
occupations. Cluster 2 was defined as urban and having a high percent of: poverty, Black residents, public
assistance, single-parent homes, unemployment, and nonmanagerial occupations. Cluster 3 was defined as
urban, high percent with Bachelor's degrees, and low percentages of: nonmanagerial occupations,
poverty, and unemployment. Cluster 4 was defined as urban, and a high percentage of: other race,
Bachelor's degrees, poverty, and owner-occupied housing. Cluster 5 was the only highly rural cluster and
had a high percent of owner occupied housing, and low percentages of: poverty, Black individuals, and
unemployment. Lastly, Cluster 6 was similar to Cluster 5 but was more urban. When compared to Cluster
3 (urban, high Bachelor's degrees) (12.8 (.ig/ni3). Cluster 2 (urban, high poverty, Black race) (13.2 (.ig/ni3)
and Cluster 1 (urban, high Black race) (12.9 (.ig/ni3) were exposed to slightly higher concentration of
PM2 5. Additionally, Clusters 4 and 5 either had the same level of exposure, or slightly lower levels of
exposure compared to Cluster 3 (11.9-12.8 (.ig/ni3).
September 2021
3-129
DRAFT: Do Not Cite or Quote
-------
Study
tMikati et al (2018)
jBell & Ebisu (2012)
tBravo et al (2016)
Population
US: American Community Survey, 2009-2013
US: Census (tracts with PM ComponentMonitors),
2000
US: Census (Eastern two-thirds), 2000
*Lcc ct al (2019)
*Son et al (2020)
*Wang et al 2020
*Rosofsky el al (2018)
US: American Community Survey (California),
2016
US: North Carolina State Center for Health
Statistics, 2002-2013
TJS: Medicare, 2000-2008
US: Census and American Community Survey
(Massachusetts), 2003-2010
(Bravo ct al (2016) US: Census (Easterntwo-thirds), 2000
*Lee et al (2019)
*Rosofsky et al (2018)
~Tanzer et al (2019)
*Lee et al (2020)
""Weaver et al (2019)
#Han et al (2020)
US: American Community Survey (California),
2016
US: Census and American Community Survey
(Massachusetts), 2003-2010
Reference Group
Not in poverty
Not in poverty
Not in poverty
All K '
Urban
Suburban
Rural
<25th percentile % in poverty
Statewide '
SF Bay Area
San Joaquin Valley
South Coast
-25th percentile
High
Medium
Low
>$75,000
2003
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
3.3.3.1.2
Health Risk Disparity
Since the literature cutoff date of the 2019 PM ISA, several additional studies evaluated health
disparities and short- and long-term PM2 5 exposure, stratified by SES. The 2009 PM ISA, summarized
evidence that indicated an increased risk between mortality and long-term exposure to PM2 5 among
groups with lower SES, which was further extended within the 2019 PM ISA. However, there was little
evidence of any SES differences among studies assessing health outcomes, such as cardiovascular
disease. Recent evidence is consistent with an increase in the risk of PM2 5-related health effects by SES
as detailed in the 2019 PM ISA. Overall, there was minimal evidence of differential health risks by SES
within studies assessing PM2 5-related all-cause or non-accidental mortality. However, stronger
associations in lower SES groups were often observed in studies assessing certain cause-specific mortality
outcomes or other health endpoints. The following sections describe recent literature pertaining to
short-term (Table A-12) and long-term (Table A-13) PM2 5 exposure and health risks among different
SES groups.
Short-Term PM2.5 Exposure
In a time-stratified case-crossover, Yitshak-Sade et al. (2019a) examined the intersection of
greenspace, cardiovascular mortality, and PM2 5 exposure. This study showed that among census block
groups with a low percentage of those without a high school diploma, there was a 1.42% (95% CI: -0.72,
3.62) increase in cardiovascular mortality in less green areas, and a 2.64% (95% CI: 0.46, 4.68) increase
in cardiovascular mortality in more green areas associated with 2-day average (lag 0-1) PM25 exposure.
However, in census block groups with a high percentage of those without a high school diploma, less
greenspace was associated with a 3.31% (95% CI: 1.26, 5.41) increase, in cardiovascular mortality, while
more greenspace was associated with a 2.64% (95% CI: 0.60, 4.72) increase in cardiovascular mortality.
These results indicate that the association between PM2 5 and cardiovascular mortality was attenuated by
greenspace only in census block groups with lower SES. Additionally, in a time-stratified case crossover
analysis in North Carolina, Son et al. (2020) reported null associations for total mortality (excluding
external causes) when stratified at the median household income (<$41,500), or by educational
attainment.
Long-Term PM2.5 Exposure
Recent studies explore whether SES modifies the relationship between exposure to PM2 5 and
premature total mortality. Similarly, a study by Wang et al. (2020) evaluated over 53 million Medicare
beneficiaries using a combination of hybrid machine learning and Cox proportional hazards to assess
long-term PM2 5 exposure and mortality at the ZIP code level, but also showed null associations for
PM2 5-related non-accidental mortality by income (high, medium, low) (Figure 3-31). However, a
September 2021
3-131
DRAFT: Do Not Cite or Quote
-------
Canadian study evaluating household income and PM2 5-related non-accidental mortality reported the
greatest magnitude of association in the lowest income category (<$25,000CAD Hazard Ratio 111R|: 1.61
[95% CI: 1.23, 2.12]) (Zhang et al.. 2021). The overall magnitude of the association was reduced with
each increasing income category.
Urbanicity
Non-accidental
1
1
1
Cardiovascular
IHD
CBV
CHF
Respiratory
COPD
Pneumonia
Cancer
1
1 t
1
Lung Cancer
1
1
1
1.0
—I—
1.2 1.0
Risk Ratio (95% Confidence Interval)
—— Noriurban I High-Income
Urban I Mid-Income
1.2
I Low-Income
Source: Wang et al. (20201. copyright permission pending.
SES = socioeconomic status.
Figure 3-31 Risk ratio of association between PM2.5 and mortality, stratified by
socioeconomic status.
Using a more comprehensive method of assessing SES, another Canadian study evaluated if a
combination of greenspace and social deprivation modified the relationship between long-term PMis
exposure and total non-accidental mortality (Crouse et al.. 2019). Community-level deprivation was
assessed using the Canadian Marginalization Index, which incorporates measures such as
September 2021
3-132
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
community-level material deprivation, residential instability, dependency, and ethnic concentration.
Crouse et al. (2019) found that the group with the lowest deprivation and lowest amount of greenspace
had a stronger association, in terms of magnitude, with non-accidental mortality compared to groups with
high deprivation and low greenspace. However, there was little differences in the association comparing
high to low deprivation in areas with high greenspace (Figure 3-32).
1.10
0.90
Non-accidental causes
Cardiometabolic causes
Cardiovascular causes
1 2
3 4
1 2
3 4
1 2
3 4
Source: Crouse et al. (20191. copyright permission pending.
1 = low greenness and low deprivation; 2 = low greenness and high deprivation; 3 = high greenness and low deprivation; 4 = high
greenness and high deprivation.
Figure 3-32 Hazard ratios for the association between PM2.5 and mortality, by
greenspace and community-level material deprivation.
A study by Bennett et al. (2019) evaluated life-expectancy changes and PM2 5 among differing
SES groups in the U.S., using data from the National Center for Health Statistics. In this study, PM2 5
concentrations that were greater than the lowest observed concentration of 2.8 (ig/m3 were associated with
lower life expectancy among counties with: lower income, higher percent poverty, and those with a low
percent of the population who graduated from high school. These differences were greater among females
September 2021
3-133
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
compared to males (Figure 3-33). In another study examining life expectancy, Jorgenson et al. (2020)
evaluated the relationship between PMa exposure, Black race, and income inequality. This study
observed that PM2 5 had a stronger effect on life expectancy at birth in states with high income inequality.
Black race further exacerbated the impact of PM25 exposure on life expectancy at birth.
Female Male
Q5 to Q1 difference^
Q5 to Q1 difference:
" -0.041 (-0.044, -0.038)
-0.018 (-0.020,-0.016)
: ¦*" '** .* \.
¦&0:
¦i
jif*'
PM2 5 exceedance in 2015 (ng/m3)
Per capita income * 1 Least 2 • 3 • 4 • 5 Most
(quintiles)
Female Male
"ST0-3
nj
9
i°,
-
to,
CD
£
0.0
Female Male
Q5 to Q1 difference:
Q5 to Q1 difference:
— 0.3-
0.046 ( 0.043, 0.048)
0.019 ( 0.017, 0.020)
«
(D
0 0.2-
X; /.•
>»
0
c
ro
0
1
a 0.1 -
®
a>
0.0-
^ *
jaF3
tV~
PM2 5 exceedance in 2015 (ng/m3)
Percentage of population whose family income « ^ Least 2 • 3 • 4 • 5 Most
is below the poverty threshold (quintiles)
Female Male
-j. 0.3
c
to,
$100,000CAD HR: 1.46 [95% CI: 0.94, 2.25]). Less differences were noted between
the lowest (<25,000CAD HR: 3.20 [95% CI: 1.40, 7.34]) and the highest (>$100,000CAD HR: 4.48 [95%
CI: 1.69, 11.83]) income categories and respiratory mortality. However, the study by Wang et al. (2020)
using data on Medicare beneficiaries, consistently reported null associations in any cause-specific
mortality by income (low, medium, high) (Figure 3-31).
September 2021
3-134
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
A cross-sectional study by Bevan et al. (2021) evaluated the association between PM2 5 exposure
and age-adjusted cardiovascular mortality, modified by the 2015 Social Deprivation Index (SDI). The
SDI is generated from a factor analysis using information collected from the American Community
Survey (ACS), including county-level descriptors of race/ethnicity, income, housing, and education. The
SDI ranges between 1 (least deprived) and 100 (most deprived). Bevan et al. (2021) demonstrated that
when examining the annual average PM2 5 and age-adjusted cardiovascular mortality, stratified by SDI,
there is increased risk of cardiovascular mortality in counties with higher social deprivation (SDI 1-25:
39.1 deaths/100,000 persons [95% CI: 32.1, 46.1]; SDI 26-50: 48.2 deaths/100,000 persons [95% CI:
38.5, 57.9]; SDI 51-75: 71.0 deaths/100,000 persons [95% CI: 57.6, 84.4]; SDI 76-100: 52.0
deaths/100,000 persons [95% CI: 30.9, 73.1]). In another study focusing on social deprivation, Wvatt et
al. (2020b') estimated the change in PM2 5 concentration between 1990 and 2010 and the associated annual
change in the age-adjusted cardiovascular mortality rate (CMR) in the U.S. A social deprivation variable
was created using 1990 U.S. Census measures including: percent of households below the poverty line,
median household income, percent of those > with at least a high school education, civilian
unemployment rate, percent female households with no spouse, percent vacant housing units, and percent
owner occupied housing units. Overall, in the earliest years, the counties with the highest social
deprivation benefited the least by the reduction in PM2 5. However, as time progressed through to 2010,
reductions in CMR were the largest in the most deprived counties (Figure 3-34).
September 2021
3-135
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
A) Annual change in PM2.5
(Mg/m3)
National
Deprivation Quintile
0.0
-0.1
a -0.2
-0.3
-0.4
B) Annual change in CMR per unit decrease in PM25
(deaths/100,000 person-years/pg/m3)
National Deprivation Quintile
•$> #
CV .OP
Year
C) CMR change attributed to PM2 5
(deaths/100,000 person-years)
National
Deprivation Quintile
Year
D) % CMR mediated via PM2.5
(%)
National
Deprivation Quintile
§
County deprivation
National
1 Lowest deprivation
|^— 2 Low deprivation
3 Mid deprivation
— 4 High deprivation
5 H ighest deprivation
Year
Year
Source: Wvatt et al. (2020b). copyright permission pending.
Note: Estimated national and socioeconomic deprivation (SED) quintile specific annual rates (95% confidence intervals). (A) depicts
the change in PM25 (a); (B) depicts the change in cardiovascular mortality rate (CMR) per 1 pg/m3 change in PM25 ((3); (C) depicts
PM25-related change in CMR (ap); and (D) depicts the PM25-related change in CMR as a portion of the overall CMR. A negative
sign for a indicates a decrease in ambient PM25 concentrations, while a positive sign indicates an increase, A positive sign for p
indicates an increase in CMR per unit increase in PM25, while a negative sign indicates a decrease. A negative sign for ap indicates
a net reduction in CMR related to PM25 change, while a positive sign indicates a net increase.
Figure 3-34 Estimated national and socioeconomic deprivation
quintile-specific mortality rates.
Schulz et al. (2018) evaluated cardiopulmonary mortality and PM2 5 in the Detroit Metropolitan
Area in 2013, stratified by area-level vulnerability. The census-tract level vulnerability index developed
by the authors incorporated percent of households below the poverty index, median home value
(reversed), percent of homes occupied by renters, percent of population >24 with less than a high school
education, linguistic isolation, percent people of color, and the percent of the population <5 and >60 years
of age. Hierarchical linear models, with a logit link independently evaluated PM2 5 and cardiopulmonary,
cardiovascular, and ischemic heart disease (IHD) mortality associated with PM2 5 exposure and increased
vulnerability. Both PM2 5 and vulnerability were independently associated with cardiopulmonary,
cardiovascular, and IHD mortality, and the associations remained, when both PM and vulnerability were
included. Additionally, the Canadian study on greenspace and social deprivation by Crouse et al. (2019)
estimated that greenspace protected against PM2 5 attributable cardiovascular and cardiometabolic
mortality, but demonstrated a stronger association in areas with low greenness and low deprivation
(Figure 3-32).
A prospective analysis by Bai et al. (2019) evaluated the association between PM2 5 and either
congestive heart failure (CHF) or acute myocardial infarction (AMI) using the Canadian Ontario
Population Health and Environment Cohort (ONPHEC). The authors observed the strongest association
September 2021
3-136
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
for both CHF (HR: 1.12 [95% CI: 1.10, 1.13]) and AMI (HR: 1.12 [95% CI: 1.09, 1.15]) among those in
the lowest income groups, compared to the uppermost income groups (CHF HR: 1.01 [95% CI: 1.00,
1.04], AMI HR: 1.03 [95% CI: 1.00, 1.06]). A study based on the same cohort also showed stronger
associations for atrial fibrillation (HR: 1.06 [1.04, 1.08]) and stroke (HR: 1.08 [1.04, 1.13]) among the
lowest income groups (Shin et al.. 2019). The central North Carolina study by Weaver et al. (2019)
evaluated the relationship between certain health outcomes, such as: coronary artery disease, myocardial
infarction (MI), hypertension, and diabetes, among cardiac catheterization patients. When compared to
Cluster 3 (urban, high percent with Bachelor's degrees, and low percentages of: nonmanagerial
occupations, poverty, and unemployment) (OR: 0.70 [95% CI: 0.86, 1.07], per 5 |_ig/m3). there was a
greater association between PM2 5 and hypertension in Cluster 1 (urban and having a high percent of
Black residents and nonmanagerial occupations) (OR = 2.70, [95% CI: 0.95, 7.59], per 5 |_ig/m3) and 2
(urban and having a high percent of: poverty, Black residents, public assistance, single-parent homes,
unemployment, and nonmanagerial occupations) (OR = 11.86, 95% [95% CI: 2.10, 67.21], per 5 (.ig/nr1).
While PM2 5 was associated with MI, the associations did not vary by cluster, and there were null
associations noted for diabetes. However, the strongest association observed among coronary artery
disease (CAD) outcomes was in Cluster 3 (OR= 1.15 [95% CI: 1.00, 1.31], per 5 (ig/m3) (Figure 3-35).
September 2021
3-137
DRAFT: Do Not Cite or Quote
-------
2
1
O
-p
o~-
LO
a:
O
2
1
Source: Weaver et al. (2019V copyright permission pending.
CAD = Coronary Artery Disease, Ml = Myocardial Infarction. Results resented for 1 |jg/m3 increase in PM25 concentrations.
Figure 3-35 Odds ratios for the association between PM2.5 and
cardiometabolic outcomes by neighborhood cluster.
3.3.3.2 Race/Ethnicity
1 The 2019 PM ISA provided evidence indicating that people of different racial and ethnic
2 backgrounds experience disparities in the risk of PiVh ^-related health effects. The 2009 PM ISA observed
3 little evidence for increased PM2 5-related risk by race and some evidence of increased risk by Hispanic
4 ethnicity. However, the 2019 PM ISA demonstrated evidence that there are consistent racial and ethnic
5 disparities in PMj 5 exposure across the U.S., particularly for Black/African Americans, compared to
6 non-Hispanic White individuals. Additionally, some studies provided evidence of increased PM2 5-related
A CAD
I ]
f 1
' I '
i
I
1
1
Hypertension
1_T1
~l 1 1 1 1 1 T
B
Ml
<
1 <
¦ 1:
: 1'
1
I
Diabetes
"i 1 1 1 1 1 r
1 2 3 4 5 6 Total 1 2 3 4 5 6 Total
Cluster
September 2021
3-138
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
mortality and other health effects from long-term exposure to PM2 5 among Black individuals. Taken
together, the 2019 PM ISA concluded that the evidence was adequate to conclude that race and ethnicity
modify PM2 5-related risk, and that non-White individuals, particularly Black individuals, are at increased
risk for PIVh s-related health effects, in part due to disparities in exposure. The following sections evaluate
recent studies pertaining to both PM2 5 exposure among different racial and ethnic groups
(Section 3.3.3.2.1) and PM2 5-related health risks among different racial and ethnic groups
(Section 3.3.3.2.2).
3.3.3.2.1 Exposure Disparity
Several recent studies within the U.S. and Canada evaluated the relationship between PM2 5
exposure and racial and ethnic disparities since the literature cutoff date of the 2019 ISA. These recent
studies add to the conclusions of the 2019 ISA, that there are disparities in PM2 5 exposure by race and
ethnicity. When considered together, these additional studies provide further evidence that non-White
communities are exposed to higher concentrations of PM2 5 compared to White communities (Figure
3-38—Race exposure, Table A-14—Race exposure).
Recent evidence further indicates that disparities in exposure persist by race and ethnicity.
Several multi-year cohort studies evaluated the potential disparity in long-term PM2 5 exposure by
race/ethnicity. The Heart Strategies Concentration on Risk Evaluation (HeartSCORE) study, conducted in
Pittsburg, PA between 2001-2014, observed that Black participants (16.1 (.ig/ni3) were exposed to slightly
higher concentrations of PM25 compared to White participants (15.7 (ig/m3) (Erqou et al.. 2018).
Similarly, the Veterans Cohort Study, conducted between 1976 and 2001 also showed that Black
participants (15.7 (ig/m3) were exposed to a substantially higher concentration of PM25 compared to
White participants (13.9 (.ig/ni3) (Lipfert and Wvzga. 2020). The NIH-AARP Diet and Health Study,
conducted between 1995 and 2011, also identified disparities in PM2 5 exposure by race and ethnicity.
When compared to White race (10.9 |_ig/m3). those of Black race (12.3 (.ig/ni3) experience higher
exposures to PM25, followed by Asian, Pacific Islander, or American Indian/Alaska Native (11.9 (.ig/ni3)
and Hispanic (11.4 (.ig/ni3) individuals Lim et al. (2018). A study conducted among the Medicare
population that moved outside of their original ZIP code between 2000 and 2012 observed that while
Post-move PM25 concentrations were lower among Black (Pre-move: 13.02 (ig/m3, Post-move:
12.12 (.ig/ni3) and White (Pre-move: 11.88 (ig/m3, Post-move: 11.15 (.ig/ni3) beneficiaries, a disparity in
exposure was still present, with Black individuals being exposed to elevated concentrations of PM2 5
compared to White individuals (Awad et al.. 2019). A study by Parker et al. (2018) using the Health
Interview Survey also showed that a greater number (37.2%) of non-Hispanic Black and Hispanic
(33.3%) individuals lived in the highest PM25 quartile, compared to White individuals (21.3%).
Additionally, most White (26.1%) and Hispanic (34.7%) individuals lived in the lowest PM25 quartile,
compared to only 10.5% of non-Hispanic Black individuals.
September 2021
3-139
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
Other recent studies used data from the U.S. Census or American Community Survey (ACS) to
gather area-level measures of race and ethnicity. A study by Rosofskv et al. (2018) observed that despite
declines in the PM2 5 concentrations during the study period, between 2003 and 2010 in Massachusetts,
the relative disparity in PM2 5 exposure by race/ethnicity remained, with non-Hispanic Black (2003:
11.7 (ig/m3, 2010: 8.4 (ig/m3), non-Hispanic Asian (2003: 11.6 (ig/m3, 2010: 8.2 (ig/m3), and Hispanic
(2003: 11.6 (ig/m3, 2010: 8.4 (.ig/ni3). populations experiencing higher PM2 5 exposures compared to
non-Hispanic White (2003: 11.1 (ig/m3, 2010: 7.8 (.ig/ni3) populations. (Kelly et al.. 2020) evaluated racial
and ethnic disparities using nine different exposure models. Despite differences in the absolute
concentrations estimated from each method, all exposure models demonstrated that areas of the U.S. with
a high percentage of Black individuals experienced higher PM2 5 exposures compared to areas with a
higher percentage of White individuals. Additionally, compared to White individuals, areas with a
predominant population of those classified as "other" race and Hispanic populations were also more
highly exposed to PM2 5. The result of (Kelly et al.. 2020) was supported in a study by Yitshak-Sade et al.
(2020) using data collected from the Massachusetts Department of Public Health showed that only 6.0%
of Black individuals were in the lowest quartile of PM2 5 exposure, while 8.1% were in the highest
quartile. Additionally, Lee (2019) in a study conducted in the state of California, demonstrated that
block-groups with a high percent (>75th percentile) of people of color (9.98 (.ig/ni3) compared to those
with a low percent (<25th percentile) of people of color (7.90 (ig/m3) were exposed to higher
concentrations of PM2 5.
Several recent studies used data from the National Emissions Inventory (NEI) to evaluate racial
and ethnic disparities in PM2 5 exposure. Richmond-Bryant et al. (2020) evaluated changes in burden due
to the closure of 92 coal-fired electricity generating units (EGUcf). Estimated from the American
Community Survey, the omission of the facilities resulted in an 11% reduction in absolute burden in
census tracts with a majority of White individuals. However, census tracks with a majority of Black
(6.6%) or Hispanic (4.4%) populations experienced less of a decrease in absolute burden. When
considering the proportional burden—the ratio of absolute burden in a racial subgroup to the absolute
burden of the total population—White census tracts had an overall decrease of 2.4%, while predominantly
Hispanic and Black census tracts had an increase in proportional burden of 4.5 and 2.2%, respectively. A
study by Tessum et al. (2019) estimated the disparities between the consumption of goods and services by
each racial and ethnic group that produce PM2 5 and the amount of PM2 5 each racial and ethnic group is
exposed to in order to estimate pollution inequity. Overall, the authors indicate that Black individuals are
exposed to the highest concentration of PM25 (6.0 (.ig/ni3). while only consuming 3.8 (ig/m3, meaning that
their estimated pollution inequity is 56%. Hispanic individuals were estimated to have a pollution
inequity of 63% (exposed to 5.5 (ig/m3, consuming 3.4 (.ig/ni3). while White individuals have an estimated
pollution inequity of-17% (exposed to 4.6 (ig/m3, consume 5.5 (.ig/ni3) (Figure 3-36). Another similar
study by Tessum et al. (2021) showed that people of color are consistently exposed to PM2 5 caused by
each emitter type in the U.S. The authors estimated Black (7.9 (.ig/ni3). Asian (7.7 (.ig/ni3). Hispanic
(7.2 (.ig/ni3) individuals are exposed to greater proportions of PM25 compared to White (5.9 (.ig/ni3)
individuals (Figure 3-37).
September 2021
3-140
DRAFT: Do Not Cite or Quote
-------
E
o>
3
a
a
to
a,
4-
3-
o-1
Source: Tessum et al. (2019). copyright permission pending.
Note: Within this figure, pollution inequity is the percent difference between a group's "exposed" and "caused" bars. Within each
group of bars, the emitters (A) and end uses (B) responsible for the exposure are depicted with gray lines showing relationships
among emitters and end uses.
Figure 3-36 Average PM2.5 experienced and caused, by racial-ethnic group.
Caused
- Black -
Exposed
Caused
Exposed
Hispanic
Caused
Exposed
¦ White/Other -
A) Emitters
Road Dustl
Res. Wood Comb. I I
Res. Gas Comb. I I
Res. Otherl 1
Off-Hiahwavl 1
Non-Coal Elec.l 1
B) End Uses
Trans.
Miscellaneous^
LD Gas Veh.L
Industrial H
HD Diesel Veh.f
Shelter
Services
Info.
] Goods
j Food
I Electricity
Construction I I
Comm. CooklngE
Coal Elec. Util.l
Agriculture |
September 2021
3-141
DRAFT: Do Not Cite or Quote
-------
A White
B poc
C Black
D Hispanic
E Asian
o i 6
^ 1 4
3 0)
0) Q. 2
Cl
X
tu 0
100 n
75 -
50 -
25 -
0 J
"
r~
r
- 40%
'
1
I
75%
f
1
i
78%
f
A
-66 0 66 133 200 -66 0 66 133 200
% Exposure disparity
0 66 133 200 -66 0
133 200 -66 0 66 133 200
F White
Coal elec. util.
Agriculture
Res. wood comb.
Road dust|
Residential olhe J
Comm. cooking J
Res, gas comb, r
Miscellaneous!
Off-hwy veh.|
& equip.
Construction I
Heavy duty
diesel veh.
Light duty
gas veh.
Industrial
G poc
Industrial
Light duty
gas veh.
Construction
Heavy duty
diesel veh.
Off-hwy veh.
& equip.
Res. gas comb.
Miscellaneous
Comm. cooking
Residential other
Road dust
Noncoal elec, util.
Res. wood comb.
Agriculture I
Coal elec. util.H
-66 -33 0 33 66 -66 -33 0
% Exposure disparity
H Black I Hispanic
Industrial
Coal elec. util.
Heavy duty
diesel veh.
Light duty
gas veh.
Construction
Miscellaneous
Res. gas comb.
Comm. cooking
Off-hwy veh.
and equip.
Residential other
Road dust
Agriculture
Noncoal e|ec, util,
Res. wood comb.
I
¥
Industrial
Light duty
gas veh.
Construction
Off-hwy veh.
& equip.
Heavy duty
diesel veh,
Miscellaneous
Comm. cooking
Residential other
Res, gas comb,
Road dust
Noncoal elec, util,
Res. wood comb.
Agriculturel
Coal elec. util.
J Asian
Light duty
gas veh.
Industrial
Res. gas comb.
Construction
Off-hwy veh.
& equip.
Heavy duty
diesel veh.
Miscellaneous
Res. wood comb.
Comm. cooking
Residential other
Noncoal elec, util,
Road dust
Agriculture
Coal elec. util
33 66 -66 -33
33 66 -66 -33
33 66 -66 -33
33 66
Source: Tessurn et al. (2021), copyright permission pending.
Note: Individual source type contributions to exposure presented on the y-axis and % exposure disparity presented on the x-axis.
Positive values are shaded red and negative values shaded blue. Dashed lines denote percent exposure caused by sources with
positive exposure disparity.
Figure 3-37 Source contributions to racial-ethnic disparity in PM2.5 exposure.
September 2021
3-142
DRAFT: Do Not Cite or Quote
-------
Study
TMikati ct al (2018)
tNaeluiiaii el al (2012)
tBravo et al (2016)
tBell & Ebisu (2012)
Population
US: American Community Survey, 2009-2013
US: NHIS Colioil, 2002-2005
US: Census (Eastern two-thirds), 2000
Racial/Ethnic Group
US: Census (tracts w/ PM ComponentMonitors),
2000
Hispanic
All
Black .
»rc
Black .
temc
Black .
w-
Black
femc
Reference Value
(pg7m3)
9.6
12.2
12.5
12.8
jBasu et al (2004) US: Birth records (California), 2000 Hispanic
*Eroqou et al (2018) US: HeartSCORE Study (Pittsburg, PA), 2001-2014 Black
~Kelly et al (2020) US: Census, 2010 C.MAQ
14.5
15.7
10.3
CAMX
9.6
9.9
*Leeetal (2019)
*Rosofsky et al (2018)
*Awadetal (2019)
*Lim et al (2018)
US: American Community Survey (California), 2016
US: Census and American Community Survey
(Massachusetts).. 2003-2010
US: Medicare, 2000-2012
US: NIII-AARP Diet and Health Study, 1995-2011
Black .
Black .
Black .
mrc
VNA ,
Black .
femc
•75th Percentile. %POC
SP Bay Area
San Joaquin valley
South Coast
503 PS
Hispanic
IK
Hispanic
Black Pre-Move
Black Posl-Move
9.3
9.1
11.1
7.8
*Lipfert et al (2020)
*Wang ct al (2020)
US: Veterans Cohort Study, 1976-2001
US: Medicare, 2000-2008
Black
Non-White
13.9
10.4
Ratio to Reference Exposure
Note: tStudies published since the 2009 PM ISA. *U.S. and Canadian studies published since the literature cutoff date (~January
2018) for the 2019 PM ISA. Circles represent ratio of each racial or ethnic group to the reference group (White individuals). Red text
and circles represent evidence included in the 2019 PM ISA; blue text and circles represent evidence not included in the 2019 PM
ISA. Reference concentrations in |jg/m3. NH: non-Hispanic. This figure builds upon Figure 12-2 in the 2019 PM ISA.
Figure 3-38 Difference in PM2.5 exposure by race.
3.3.3.2.2 Health Risk Disparity
1 Since the literature cutoff date of the 2019 ISA, several additional studies evaluated disparities in
2 the risk of PM2 5-related health effects, stratified by race and ethnicity. A small number of studies were
3 included in the 2009 PM ISA, that summarized racial and ethnic disparities in PM2.5 mortality risk. The
4 2019 PM ISA further identified several studies which provided evidence for an increased association
September 2021
3-143
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
between mortality and long-term exposure to PM25 among non-White groups. However, evidence of any
racial or ethnic disparities and PM2 5-related health outcomes was inconsistent. The most recent evidence
is consistent with differences in the risk of PM2 5-related health effects by race and ethnicity, as detailed in
the 2019 PM ISA. While the studies that evaluated all-cause or total (nonaccidental) mortality were
inconsistent, there was stronger evidence to indicate a greater risk of cause-specific mortality and some
other health endpoints among people of color. In particular, individuals of Black race consistently were
shown to have a greater risk for health outcomes associated with PM2 5 exposure. The following sections
evaluate recent studies pertaining to short-term (Table A-15) and long-term (Table A-16) PM2 5 exposure
and PM2 5-related health risks among different racial groups.
Short-Term PM2.5 Exposure
In a time-stratified case-crossover analysis, Yitshak-Sade et al. (2019a) examined the intersection
of greenspace, cardiovascular mortality, and PM2 5 exposure in Massachusetts. This study demonstrated
that among census block groups with a low percentage of White individuals, there was a 3.55% (95% CI:
1.49, 5.65) increase in cardiovascular mortality in less green areas, and a 2.47% (95% CI: 0.43, 4.56)
increase in cardiovascular mortality in more green areas associated with two-day average (lag 0-1) PM2 5
exposure. However, in census block groups with a high percentage of White individuals, less greenspace
was associated with a 1.14% (95% CI: -1.00, 3.33) increase, in cardiovascular mortality, while more
greenspace was associated with a 2.8% (95% CI: 0.62, 5.02) increase in cardiovascular mortality (Figure
3-39). These results indicate that regard less of greenspace, census block groups with low percentage of
White individuals have a higher risk of PM2 5-related cardiovascular mortality.
Another study examining short-term PM2 5 exposure and cardiovascular mortality conducted in
Massachusetts used four different metrics in which to evaluate race and ethnicity within the state
(Yitshak-Sade et al.. 2020). There was a higher percent change in cardiovascular mortality among Black
(4.78% [95% CI: -1.99, 12.02]) individuals compared to White (2.25% [95% CI: 0.80, 3.23]) associated
with the two-day moving average (lag 0-1) of PM2 5. When evaluating the racial composition by census
block group, there was a 1.62% (95% CI: 0.05, 3.22) increase in cardiovascular mortality in census block
groups with 0% Black (10th percentile) compared to a 3.35% (95% CI: 1.57, 5.16) increase in
cardiovascular mortality in census block groups with more than a 15% Black population (90th percentile)
(see Figure 3-39). This study also evaluated two novel measures of racial segregation. The first was the
Racial Residental Segregation (RRS) metric which quantified the concentration of non-Hispanic Black
and non-Hispanic White individuals in each census block group and could range between -1 (more Black
individuals) to 1 (more White individuals). In census block groups with more White individuals (RRS 0.5
to 1) there was a 1.84% (95% CI: 0.31, 3.40) increase in cardiovascular mortality, whereas in census
block groups with more Black individuals (RRS -1 to -0.5) there was a 15.37% (95% CI: 0.76, 31.99)
increase in cardiovascular mortality. The second indicator examined was the Index of Racial Dissimilarity
(IRD) which measured the dissimilarity between the distribution of non-Hispanic Black and non-Hispanic
White individuals within the census block group to the larger census tract. A higher IRD is indicative of
September 2021
3-144
DRAFT: Do Not Cite or Quote
-------
1
2
3
greater dissimilarity in the proportion of Black residents between the census block group and the census
tract. There was no evidence of differences in cardiovascular mortality risk attributed to short-term PRfej
exposure among any of the groups (Figure 3-40).
ov
ir<
o
s
u
o
r*
s
¦u
it
"S
a*
M
s
«
s
s-
7.0
6.5
6.0
5.5
5.0
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
-0.5
-1.0
0%
1.5%
15%
58%
89%
98%
(10th p)
(50th p)
(90th p)
(10th p)
(50th p)
(90th p)
% Black
% White
Source: Yitshak-Sade et al. (2020;. copyright permission pending.
Note: Results presented for a 10 [jg/m3 increase in PM2.5.
Figure 3-39 Percent change in cardiovascular disease mortality by PM2.5
exposure, stratified by census block group racial composition in
Massachusetts (2001-2011).
September 2021
3-145
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
0!
t
o
a
¦a
« o
n
Em U-,
= si-
it
B
35.0
30.0
25.0
20.0
15.0
10.0
5.0
0.0
-5.0
-10.0
-1 to -0.5
0,5 to -0.1
-0,1 to 0.1
!_^_i
0.1 to 0.5 0.5 to 1
i
RRS
8.17%
(10th p)
14.6%
(50th p)
tRD
27.6%
(90th p)
Source: Yitshak-Sade et al. (2020). copyright permission pending.
Note: Results percented for a 10 |jg/m3 increase in PM2.5.
RRS = Racial Residental Segregation, IRD = Index of Racial Dissimilarity.
Figure 3-40 Percent change in cardiovascular disease mortality by PM2.5
exposure, stratified by the racial residential segregation and
index of racial dissimilarity in Massachusetts (2001-2011).
Long-Term PM2.5 Exposure
A number of recent epidemiologic studies further evaluated whether race and ethnicity modify
the association between all-cause (non-accidental) mortality and long-term PM2.5 exposure. A study by
Parker et al. (2018) using National Health Interview Survey data evaluated all-cause mortality (excluding
unintentional injuries) and PM2.5 by race and ethnicity. This study reported a larger association, in terms
of magnitude, among Black (HR: 1.05 [95% CI: 1.03, 1.09]) and White (HR: 1.02 [95% CI: 1.00, 1.05])
individuals and a null association among Hispanic individuals (HR: 0.98 [95% CI: 0.94, 1.03]). However,
A wad et al. (2019) in a study of Medicare beneficiaries who moved out of their ZIP code did not report
results consistent with Parker et al. (2018). Since moving was essentially random, this created a natural
experiment that randomized an individual's exposure. Inverse probability weights were used to control for
several select covariates. Among Black movers, the HR for all-cause mortality was lower (HR: 1.06 [95%
CI: 1.04, 1.07]), compared to White movers (HR: 1.10 [95% CI: 1.10, 1.10]). Similarly, a recent study
using data from the Veterans Cohort Mortality Study from 1976-2001 evaluated mortality risk among
Black and White veterans. The RR were expressed in terms of the difference in the annual average and
the minimum concentration of PM2.5, and the results are interpreted as the change in mortality that would
September 2021
3-146
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
result if all cohort members were exposed to the minimum concentration of PM2 5. The mortality rate
using years 1976-2001 and PM2 5 measured between 1999-2001 was higher among White veterans
(RR: 1.05 [95% CI: 1.01, 1.10]) compared to Black veterans (RR: 0.82 [95% CI: 0.75, 0.89]). This effect
was less pronounced when only examining the cohort between 1997 and 2001 (White RR: 1.03 [95% CI:
0.91, 1.17], Black RR: 0.96 [95% CI: 0.76, 1.21]) (Lipfcrt and Wvzga. 2020). Another study of
non-accidental mortality among Medicare beneficiaries by Wang et al. (2020) observed positive, but
equal, relative risks (RR) among Black (RR: 1.02 [95% CI: 1.02, 1.02]) and White (RR: 1.03 [95% CI:
1.02, 1.03]) beneficiaries. Additionally, the North Carolina case-crossover study by Son et al. (2020) also
observed very little racial and ethnic differences for total mortality when stratified by race/ethnicity
(non-Hispanic White OR: 1.01 [1.01, 1.01]; non-Hispanic Black OR: 1.01 [1.00, 1.02]; Hispanic OR:
0.97 [0.93, 1.02]; non-Hispanic Asian OR: 1.01 [0.97, 1.02]; non-Hispanic Other OR: 1.01 [0.97, 1.04]).
Bennett et al. (2019) evaluated life-expectancy changes and PM2 5 among differing racial groups
in the U.S. In this study, counties with PM2 5 concentrations that exceeded 2.8 |ig/m3 and had a high
proportion of Black or African Americans were associated with lower life expectancy. This difference
was also greater among females compared to males (Figure 3-33). The study by Jorgenson et al. (2020)
that assessed the intersection between PM2 5 exposure, Black race, and income inequality also
demonstrated that states with a high percentage of Black populations had worse life expectancy at birth
associated with PM2 5. When including a three-way interaction between PM2 5, Black race, and income
inequality, the slope for PM2 5 and life expectancy becomes more negative as the percent of the population
that is Black and the income inequality increase. Overall, this study states that PM2 5 appears to have the
largest effect on life expectancy in states with a high level of income inequality and a larger percentage of
people of Black race (Figure 3-41).
September 2021
3-147
DRAFT: Do Not Cite or Quote
-------
90th Percentile of Top 10% Share = 50.8%
% Biack
Source; Joraenson et al. (2020). copyright permission pending.
Figure 3-41 Relationship between life expectancy and PM2.5 exposure by
income inequality and percent Black.
1 Recent epidemiologic studies also evaluated specific causes of death attributable to PMm
2 stratified by race and ethnicity. In an analysis of the NIH-AARP Diet and Health Study (1995-2011),
3 Lim et al. (2018) evaluated diabetes mortality. Black participants had an association larger in magnitude
4 between annual PM25 exposure and diabetes mortality (HR: 1.27 [95% CI: 1.02, 1.58]), compared to
September 2021
3-148
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
White participants (HR: 1.05 [95% CI: 0.96, 1.14]). However, other studies that examined whether there
are disparities in PMis-related cause-specific mortality do not report results consistent with Litn et al.
(2018). The study by Parker et al. (2018i that used the National Health Interview Survey (NHIS) between
1997-2001 to evaluate heart disease mortality and annual PM2 5 exposure reported the largest association
among White individuals (HR: 1.10 [95% CI: 1.05, 1.15]), with associations smaller in magnitude for
Black (HR: 1.04 [95% CI: 0.94, 1.15] and Hispanic (HR: 1.03 [95% CI: 0.95, 1.12] individuals, and no
evidence of an association for individuals of "other' races (HR: 0.90 [95% CI: 0.75, 1.07]). Lastly, the
study by Wang et al. (2020) evaluating over 53 million Medicare beneficiaries showed associations larger
in magnitude among White beneficiaries compared to non-White beneficiaries for cardiovascular
mortality (White RR: 1.05 [95% CI: 1.05, 1.06], non-White RR 1.03 [95% CI: 1.02, 1.03), heart disease
mortality (White RR: 1.07 [95% CI: 1.07, 1.08], non-White RR 1.03 [95% CI: 1.02, 1.04), and vascular
disease mortality (White RR: 1.07 [95% CI: 1.06, 1.08], non-White RR 1.05 [95% CI: 1.04, 1.06]).
In a study of post-menopausal women enrolled 111 the Women's Health Initiative (WHI), Honda et
al. (2017) estimated that the association between long-term exposure to PM2 5 and incident hypertension
was larger in magnitude among Asian/Pacific Islander (HR: 1.34 [95% CI: 1.00, 1.64]), non-White (HR:
1.27 [95% CI: 1.17, 1.38]), and Black participants (HR: 1.26 [95% CI: 1.06, 1.44]) compared to White
participants (HR: 1.15 [95% CI: 0.99, 1.35]). However, a cross-sectional study evaluating a cohort of
community center health patients located in 12 southeastern U.S. states evaluated self-reported
cardiometabolic disease (CMD) and long-term PM2 5 exposure reported no differences in the association
by race (Juarez et al.. 2020) (Figure 3-42).
Probability of CMD
0.20
0.15
O.IO
10
13
20
White Male with Hypertension
White Female with Hypertension
Black Male with Hypertension
Black Female with Hypertension
White Male
White Female
¦— Black Male
—Black Female
Source: Juarez et ai. (2020). copyright permission pending.
Figure 3-42 Probability of cardiometabolic disease and PM2.5 exposure,
stratified by race, gender, and hypertension status.
September 2021
3-149
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
3.3.3.3 Summary of Recent Evidence on At-Risk Populations in the Context
of Conclusions of the 2019 Integrated Science Assessment for
Particulate Matter
Within the 2019 PM ISA, evidence was evaluated that indicated some populations and lifestages
are at increased risk of a PM2 5-related health effect (2019 PM ISA, Chapter 12). These disparities
between populations and lifestages were in some cases found to be attributed to differences in health risks
as well as exposure. When considering proxies for SES, such as income or educational attainment, having
low income, or residing in a low-income area studies were evaluated that reported associations larger in
magnitude between mortality and long-term PM2 5 exposure, compared to populations with higher income
or living in higher income neighborhoods. However, there was inconsistent evidence of differential risk
when comparing across populations with low and high educational attainment. Additionally, evidence
was presented in the 2019 PM ISA that indicated a consistent disparity in PM2 5 exposure among different
racial and ethnic groups. This disparity also translated to PM-related health risks, specifically
demonstrating that Black individuals were at higher risk for PM2 5-related health outcomes, such as
mortality. Taken together, the 2019 PM ISA determined that the evidence was suggestive that people of
low SES and adequate to indicate that race and ethnicity, specifically non-White and Black populations,
are at increased risk of PM2 5-related health effects, in part due to disparities in exposure. Overall, the
recent studies support the conclusions of the 2019 PM ISA.
Recent epidemiologic studies published since the 2019 PM ISA supports the evidence that SES
may modify the association between PM2 5 exposure and PM2 5-related health risk. Studies
evaluating PM2 5-related health risks by SES add to the growing evidence presented in the 2019 PM ISA.
In addition to the proxy-SES metrics (e.g., income, educational attainment), several recent studies
explored composite measures of neighborhood SES, which consistently demonstrated a disparity in both
PM2 5 exposure and the risk of PM2 5-related health outcomes. Additionally, the recent evidence supported
the conclusions that lower SES is associated with cause-specific mortality and certain health endpoints
(i.e., MI and CHF), but less so for all-cause or total (non-accidental) mortality (Section 3.3.3.1).
Consistent with the evidence presented in the 2019 PM ISA, recent studies continue to support
and extend the evidence that disparities in PM2 5 exposure and health risk exist among different racial and
ethnic groups. Those of Black race, or who live in predominantly Black neighborhoods, are consistently
subjected to the higher PM2 5 exposures, especially when compared to non-Hispanic-White groups.
Recent studies also continue to support racial and ethnic disparities in the association between PM2 5
exposure and cause-specific mortality or certain health endpoints (i.e., incident hypertension), especially
when comparing Hispanic and non-Hispanic Black populations to non-Hispanic White populations.
However, similar to SES, there was less consistency when evaluating PM2 5 exposure and all-cause or
total (non-accidental) (Section 3.3.3.2).
September 2021
3-150
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
4.EVALUATION OF RECENT WELFARE EFFECTS
EVIDENCE
The 2019 PM ISA concluded a causal relationship for each of the three non-ecological welfare
effects categories evaluated: visibility effects, climate effects, and materials effects. However, the welfare
effects studies evaluated within this chapter represent only those studies most informative in considering
potential revisions to the PM NAAQS as defined by the scope of this Supplement (Section 1.2.2).
specifically studies that inform the relationship between PM and visibility impairment. Within this section
the evaluation of recent studies is performed in the context of the studies evaluated and scientific
conclusions presented in the 2019 PM ISA. As a result, within the following section, the summary and
causality determination from the 2019 PM ISA is presented prior to the evaluation of recent studies
published since the literature cutoff date of the 2019 PM ISA that examine the relationship between PM
and visibility impairment (Section 4.2). This approach allows for a full accounting of the evidence that
formed the basis of the key scientific conclusions in the 2019 PM ISA with respect to visibility
impairment and the identification of specific sections of the 2019 PM ISA that provide additional details
on the total evidence base being considered in the process of reconsidering the PM NAAQS.
The studies evaluated within the following sections represent only those studies most informative
in considering potential revisions to the PM NAAQS, i.e., that provide new information on public
preferences and/or methodologies or quantitative analyses of visibility impairment. As a result, the
summary section (Section 4.3) for visibility effects conveys how the evidence from recent studies fits
within the scientific conclusions of the 2019 PM ISA, and indicates whether recent evidence supports (is
consistent with), supports and extends (is consistent with and reduces uncertainties), or does not support
(is not consistent with) the causality determination for visibility effects in the 2019 PM ISA.
4.1 Summary of Evidence for Visibility Effects from 2019
Integrated Science Assessment for Particulate Matter
Overall, visibility in most regions of the U.S. has improved since the 2009 PM ISA, as indicated
by lower estimates of PM2 5 mass extinction. The greatest improvements have occurred in the eastern half
of the U.S., in regions with the poorest visibility. This has likely occurred because of a reduction in SO2
emissions, resulting in lower ammonium sulfate concentrations, because ammonium sulfate has
historically accounted for a larger fraction of PM2 5 mass than other PM2 5 components, and also because
ammonium sulfate is more effective than other PM2 5 components at scattering light. The resulting
decrease in PM2 5 in the eastern U.S. has resulted in better visibility.
September 2021
4-1
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Rural visibility impairment is greatest in eastern U.S. regions, including the Southeast, East
Coast, Mid-South, Central Great Plains, and Appalachian regions. In contrast, visibility is better on
average in most regions of the western U.S. Urban visibility is also generally better in the western U.S.
than in the eastern U.S., except for urban areas in California and Alaska. In part, this reflects the
difference in PM2 5 composition between the East and West, with a greater fraction being made up of
ammonium sulfate in the eastern U.S. and of particulate organic matter in the western U.S. The
effectiveness of light extinction by PM2 5 depends on composition and relative humidity, with low
scattering efficiency from PM10-2.5, moderate scattering efficiency by organic mass and sea salt, high
scattering efficiency by ammonium sulfate and ammonium nitrate, and high total extinction
(scattering + absorption) by light-absorbing carbon. However, the difference in extinction between the
eastern U.S. and western U.S. also reflects considerably higher PM25 concentrations in the eastern U.S.
and California than in the rest of the western U.S.
Altogether, new results and observations discussed in the 2019 PM ISA regarding atmospheric
visibility provide evidence that atmospheric visibility has improved as PM concentrations have decreased,
that regional and seasonal differences in atmospheric visibility parallel regional and seasonal PM
concentration patterns, and that regional differences in the relationship between PM and visibility are due
to differences in PM composition characteristics, rather than any factors beyond PM. These results
confirm a well-established relationship between PM and visibility summarized in the 2009 PM ISA and
earlier assessments. Overall, the evidence is sufficient to conclude that a causal relationship exists
between PM and visibility impairment.
4.2 Recent Studies that Inform the Relationship between PM
and Visibility Effects
Within the 2019 PM ISA (Section 13.2), studies that assessed the relationship between PM and
visibility impairment were evaluated. These included preference studies to assess acceptability of visual
air quality (Section 4.2.1). as well as field studies and computational methods to evaluate trends in visual
air quality (Section 4.2.2). Some studies published since the literature cutoff date for the 2019 PM ISA
provide additional insight into the PM-visibility impairment relationship.
4.2.1 Visibility Preference and Light Extinction
A recent study by Malm et al. (2019) reported results from a new visibility preference study and
evaluated the consistency between previous visibility preference studies. The main conclusion of this
study was that the level of acceptable visual air quality is more consistent across studies using metrics that
evaluate the distinction of an object from a background than using metrics that evaluate the greatest
distance at which an object can be observed. As described in the 2019 PM ISA, two fundamental
September 2021
4-2
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
characteristics of atmospheric visibility impairment are (1) a reduction in visual range, the greatest
distance through the atmosphere at which a prominent object can be identified, and (2) a reduction in
contrast, the sharpness with which an object can be distinguished from another object or background
(Malm. 2016). Both of these concepts can be expressed in terms of an atmospheric extinction coefficient
(bext), that relates the distance of an observed object to atmospheric light extinction following the
Beer-Lambert Law (U.S. EPA. 2019).
In practice, assessment of visibility impairment typically involves estimating the amount of light
extinction (bext) from measurements of PM species concentrations. Considerable effort has been devoted
to the development of an algorithm to estimate light extinction (6ext) with reasonable accuracy from PM
species monitoring data using data on dry mass extinction efficiency and hygroscopicity growth functions
for major species. Because it was developed based on measurements from the Interagency Monitoring of
Protected Visual Environments (IMPROVE) network, this algorithm is widely referred to as the
IMPROVE equation. Light extinction (bc:d) estimated from IMPROVE network data using the IMPROVE
equation is widely used to describe visibility impairment as well as to explore the relationship between
atmospheric visibility and PM2 5 concentrations. The 2009 PM ISA (U.S. EPA. 2019) reviewed evidence
that in polluted environments most of the light extinction (bext) responsible for visibility impairment is due
to scattering by PM2 5, although absorption by elemental carbon and some crustal materials as well as
scattering by coarse PM are important in some locations. Light extinction (6ext) was generally elevated in
urban centers compared to surrounding rural areas, particularly in the western U.S.
The demand for good visual air quality has been evaluated in multiple diverse locations using
visibility preference studies with similar protocols. In these studies, respondents are shown photographic
slides or hardcopy photographs of a single scene under various visibility conditions and asked to (1) rate
the visual air quality for each photograph on a scale from 1 (poor) to 7 (excellent) and (2) judge whether
the visual air quality depicted in the image was considered to be acceptable or unacceptable. Visibility
preference results from eight locations have been published between 1991 and 2019 using this approach
(Malm et al.. 2019). including five urban and two non-urban locations in the U.S. or Canada. Results from
the five urban locations were all reported before publication of the 2009 PM ISA (U.S. EPA. 2009).
where they are reviewed in detail. Results to these studies were generally reported in terms of light
extinction (6ext), or related metrics concerned with the distance at which an object can be seen, like visual
range. When results were compared between different locations, a wide range in acceptable values was
observed for these metrics. For example, median acceptable visual range values ranged from 20 km in
Washington, DC to 59 km in Denver (U.S. EPA. 2009).
Although no new visibility preference studies in the U.S. had been reported between the
publication of the 2009 PM ISA and the 2019 PM ISA (U.S. EPA. 2019). an additional U.S. study of the
Grand Canyon was published since that time, and previous studies that were summarized in the 2009 ISA
were reanalyzed using alternative visibility metrics that reduced the variability in acceptability between
settings and locations (Malm et al.. 2019). Figure 4-1 (Malm et al.. 2019) shows the percentage of
September 2021
4-3
DRAFT: Do Not Cite or Quote
-------
1 observers that indicated acceptable visual air quality as a function of atmospheric extinction coefficient
2 based on rankings of photographic images from six visibility preference studies conducted in Phoenix,
3 AZ (ADEO. 2003); Chilliwack and Abbotsford, BC (Prvor. 1996). Denver, CO (Ely et al.. 1991).
4 Washington, DC (Abt. 2001). and the Grand Canyon, AZ (Malm et al.. 2019).
400
Source: Malm et al. (2019). copyright permission pending.
Figure 4-1 Percent acceptability levels plotted against atmospheric
extinction coefficient for each of the images used in studies for
Washington, DC (WASH), Phoenix, AZ (PHX), Chilliwack, BC
(CHIL), Abbotsford, BC (ABBT), Denver, CO (DEN), and the Grand
Canyon, AZ (GRCA).
5 These results clearly demonstrate a large range in atmospheric extinction coefficient (bext) at a
6 given level of acceptability, indicating that metrics based on extinction are not universal indicators of
7 visibility preference levels. For example, considerably more atmospheric extinction coefficient was
8 needed for the Washington, DC scene to be regarded as unacceptable (192 Mm1) than for the Grand
September 2021
4-4
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
Canyon, AZ scene (23 Mm-1) (Malm et al.. 2019). For other locations the amount of extinction
corresponding to an unacceptable level of visibility impairment was intermediate between Washington,
DC and the Grand Canyon, AZ (Malm et al.. 2019). For context, urban monthly average bext derived from
2011-2014 Chemical Speciation Network (CSN) data presented in the 2019 PM ISA was within this
range for at least 1 month in all 31 U.S. regions analyzed (U.S. EPA. 2019).
Ideally, the relationship between the acceptability rating and the metric used for visibility
impairment would be independent of the scene being observed (Malm et al.. 2019). However, Figure 4-1
shows that when using extinction or any universal haze metric, the response was highly dependent on the
scene. Malm et al. (1981) showed that visual air quality judgments like those applied in visibility
preference studies are related to contrast of landscape features and thus are dependent on the integration
of haze over the sight paths between the observer and landscape features. Based on these results, Malm et
al. (2019) suggested that scene-dependent metrics like contrast that integrate the effects of bext along the
sight paths between observers and landscape features are better predictors of preference levels than the
universal metrics like extinction. The explanation for this is that extinction alone is not a measure of haze
(Malm et al.. 2019). but of light attenuation per unit distance, and visible haze is dependent on both
extinction and distance to a landscape feature. As a result, more haze is required to affect a nearby feature
than more distant features, and landscape features at different distances from an observer each have a
unique sensitivity to changes in atmospheric extinction, and consequently to PM mass concentration
(Malm et al.. 2019).
Figure 4-2 shows acceptability levels for the same studies in Figure 4-1 plotted against apparent
contrast of the distant feature most sensitive to haze. When the features approximately reach the visual
range, corresponding to a contrast between about -0.03 to -0.05, about 50% of observers rated the image
as not acceptable. When acceptability is expressed as a function of contrast in Figure 4-2. the response
was less dependent on scene, as indicated by a smaller difference between average acceptability in Grand
Canyon, AZ and Washington, DC than when acceptability is expressed as a function of extinction in
Figure 4-1. This is because most features in the Grand Canyon, AZ scene were more than 10 km distant,
while landscape elements in the Washington, DC scene were only a few kilometers apart, and foreground
features less than 1 km apart (Malm et al.. 2019).
September 2021
4-5
DRAFT: Do Not Cite or Quote
-------
-0.24 -0.20 -0.16 -0.12
Contrast
-0.08
-0.04
-0.00
Source: Malm et al <2019). copyright permission pending
Figure 4-2
Percent acceptability levels plotted against apparent contrast of
distant landscape features for each of the images used in studies
for Washington, DC (WASH), Phoenix, AZ (PHX), Chilliwack, BC
(CHIL), Abbotsford, BC (ABBT), Denver, CO (DEN), and the Grand
Canyon, AZ (GRCA).
Malm et al. (2019) concluded that visibility preference studies suggest that about 50% of
individuals would find visibility unacceptable if at any time the more distant landscape features nearly
disappear, and that this occurs when these features are near the visual range and have contrast levels of
approximately -0.02 to -0.05. Further, an acceptability level of 90% would require contrast levels to
remain above a level of about -0.01.
September 2021
4-6
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
4.2.2
Recent Advancements in Visibility Monitoring and Assessment
Hand et al. (2020) recently analyzed national and regional trends in light extinction based on
reconstructed total light extinction estimated from IMPROVE network speciated PM concentrations in
remote areas. On average the atmospheric extinction decreased by 1.8% per year from 1990 to 2018 and
by 2. 8% per year from 2002 to 2018. In the Eastern U.S., the atmospheric extinction coefficient decreased
by 4.3 % per year from 2002-2018 and was associated with major reductions of SO2 and NOx emissions
(Hand et al.. 2020). The reduction in atmospheric extinction reported by Hand et al. (2020) is supported
by the declining trend in sulfate concentrations detailed in the 2019 PM ISA (U.S. EPA. 2019). Hand et
al. (2020) concluded that regulation of SO2 and NOx emissions have improved atmospheric visibility, but
that the relative contribution of unregulated sources like biomass burning, dust, and international transport
to visibility impairment is increasing.
In addition, some recent studies suggest that this rapidly changing PM composition can impact
the accuracy of light extinction estimates. As described in detail in the 2019 PM ISA, light extinction is
estimated from IMPROVE network data using an algorithm that combines PM species mass, PM size,
PM species mass scattering coefficients, and relative humidity (Malm et al.. 1994). This algorithm was
modified with the goal of reducing bias that had been observed in applications of the original IMPROVE
equation by splitting major PM components between small and large size modes in recognition that
atmospheric PM generally follows a bimodal size distribution (Pitchford et al.. 2007). This approach has
been referred to as the revised IMPROVE equation (U.S. EPA. 2019) or the split component algorithm
(Prenni et al.. 2019). However, by the time of publication of the 2019 PM ISA, new studies had
concluded that the modified IMPROVE equation had not been generally successful in decreasing the bias
in atmospheric extinction estimates associated with the original equation (U.S. EPA. 2019).
Since the literature cutoff date of the 2019 PM ISA, additional bias due to changing PM
composition has been described (Prenni et al.. 2019). Estimates of atmospheric extinction based on the
modified IMPROVE equation were compared to light scattering directly measured with a nephelometer at
11 IMPROVE network sites from 2001-2016. The relationship between measured and reconstructed light
scattering had changed overtime, and light scattering was underestimated at many sites in recent years
(Prenni et al.. 2019). During the period when light scattering was underestimated, large decreases in
sulfate and organic mass concentrations also occurred. A 25% underprediction was also reported when
reconstructed extinction was compared to open-path cavity ringdown spectrometry measurements of light
extinction in the Great Smoky Mountains National Park during the summer of 2016 by Gordon et al.
(2018). The authors concluded that the accuracy of atmospheric extinction estimates from the IMPROVE
equation was reasonable at average relative humidity, but substantially lower at both higher and lower
humidities.
The importance of including relative humidity in estimating atmospheric extinction in urban areas
of the Eastern U.S. was demonstrated in the Baltimore, MD-Washington, DC area (Beversdorf et al..
2016). On days when PM was transported from the north, PM2 5 concentrations averaged 5.4 (ig/m3 and
September 2021
4-7
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
the variability of atmospheric extinction was controlled primarily by differences in PM concentrations.
However, substantially higher concentrations, averaging 18.4 (ig/m3 were observed for westerly PM
transport from the Ohio River Valley, PM was more hygroscopic, and the variability of atmospheric
extinction was controlled by both PM2 5 concentration and relative humidity (Bcversdorf et al.. 2016).
The observations of persistent and potentially increasing bias in the main method for estimating
atmospheric extinction from national monitoring network data is a continued concern, and research is
progressing on how to reduce it. Most recently, an alternative approach to the split component revised
IMPROVE algorithm was explored to evaluate the bias associated with component contributions to
extinction estimated by scaling to annual average mass by component, year, and site (Prcnni et al.. 2019).
This approach successfully reduced biases in extinction estimates from the split component revised
IMPROVE algorithm.
As described in the 2019 PM ISA, recent decreases in SO2 and NOx emissions have coincided
with increasing PM emissions from wildland fires (U.S. EPA. 2019). As a result, recent studies have
focused on visibility impairment specifically from fire related PM. Recent estimates of mass scattering
efficiencies for wildland fire smoke ranged from 2.50 to 4.76 m2/g, and increased with particle median
diameter, as expected from theoretical predictions (Laing et al.. 2016). Mass scattering efficiencies of fire
related PM were observed to increase by 56% during the first 2-3 hours after emission, at least in part
due to increasing particle size during atmospheric aging (Klcinman et al.. 2020). The large and rapid
change in mass scattering efficiencies during atmospheric aging presents a challenge for accurately
estimating atmospheric light extinction based on constant mass scattering coefficients, as in the
IMPROVE equation.
Recent studies have also introduced new instrumentation and measurement methods that could
help to reduce uncertainties in extinction. Multi-wavelength light attenuation methods used as a part of
thermal/optical carbon analysis in the IMPROVE and Chemical Speciation Networks (2019 PM ISA,
Section 2.4 and Section 13.2.4) were initiated in 2016 and are used to estimate brown carbon, which in
turn can be used to estimate biomass burning contributions to atmospheric light absorption (Chow et al..
2021). Application of a paired-wavelength method to estimate light absorption by brown carbon resulted
in a factor of two increase in the estimate of the contribution of brown carbon to light absorption in CSN
network samples from 2016-2017 (Chow et al.. 2021). These results suggest that a considerably greater
contribution of organic PM to light absorption by PM than could be estimated using network
thermal/optical methods in place before 2016. This is a highly relevant result given that organic matter
has recently overtaken sulfate as the most abundant PM2 5 component in many locations, and that
wildland fire emissions are an increasing source of PM (U.S. EPA. 2019).
September 2021
4-8
DRAFT: Do Not Cite or Quote
-------
1 Capabilities for estimating atmospheric extinction from photographic images have also
2 progressed. In the 2019 PM ISA photography is identified as one of several methods of measuring
3 atmospheric extinction (U.S. EPA. 2019). and results from image processing techniques had been shown
4 to be highly correlated with measured atmospheric extinction in hazy atmospheres under hazy conditions.
5 Additional, a recent study demonstrated that atmospheric extinction could also be estimated quantitatively
6 from photographic images under more pristine conditions, like webcams that are routinely operated at
7 National Parks (Malm et al.. 2018).
4.3 Summary of Recent Evidence in the Context of the 2019
Integrated Science Assessment for Particulate Matter
Causality Determination for Visibility Effects
8 Recent studies published since the 2019 PM ISA has addressed several existing research gaps and
9 emerging trends identified in the 2019 PM ISA. Analyzing visibility preference study results using
10 contrast as a metric greatly reduced the variability in acceptability between studies that reported results in
11 terms of atmospheric extinction or related metrics. Evaluation of uncertainties and development of
12 alternative approaches to estimating atmospheric extinction using the IMPROVE algorithm as well as
13 alternative methods for analyzing both light extinction and PM species have helped to meet new needs
14 introduced by rapidly decreasing sulfate and increasing fire related contributions to PM. New
15 measurements of physical and optical properties of wildfire smoke also provide useful new data for
16 understanding PM sources responsible for light extinction.
17 Some recent studies evaluated in this section provide the following new insights:
18 • A wide range in response to the level of acceptable visibility observed across different settings is
19 reduced by accounting for the distance between observer and landscape feature as a part of a
20 visibility metric. This was demonstrated by observation of a smaller variation across different
21 settings using apparent contrast than extinction.
22 • Impacts of the rapidly decreasing sulfate and increasing fire related contributions to PM have
23 been evaluated in recognition that the changing nature of PM composition in the U.S. is changing
24 the relationship between PM and visibility impairment.
25 • The changing relationship between PM and visibility impairment has led to increased bias and
26 spurred alternate approaches that have reduced bias in the IMPROVE algorithm used to estimate
27 light extinction.
28 Additional recent studies further support the conclusions in the 2019 PM ISA, specifically:
29 • In polluted environments most of the light extinction (bext) is due to mainly to scattering by PM2 5;
30 although absorption by elemental carbon and some crustal materials as well as scattering by
31 coarse PM are important in some locations.
32 • Light extinction (6ext) is generally elevated in urban centers compared to surrounding rural areas,
33 particularly in the western U.S.
September 2021
4-9
DRAFT: Do Not Cite or Quote
-------
1 • In practice, bext is estimated with reasonable accuracy using routinely available PM species
2 monitoring data using data on dry mass extinction efficiency and hygroscopicity growth functions
3 for major species. Mass extinction efficiencies can vary by a factor of 10 or more between
4 particulate species, which vary by region and season as well as by urban versus rural settings.
5 • Mass extinction efficiencies for sulfates, nitrates, and organics in rural areas tend to increase with
6 increasing concentrations due to shifts in the size distributions and more recent studies have
7 shown a similar dependency on concentration in urban and polluted environments.
September 2021
4-10
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
5.SUMMARY AND CONCLUSIONS
The 2019 Integrated Science Assessment for Particulate Matter (PM ISA) is a comprehensive,
systematic evaluation of the state of the science with respect to the health and welfare effects of PM that
built upon previous assessments conducted in support of the National Ambient Air Quality Standards
(NAAQS) for PM (U.S. EPA. 2019). While the 2019 PM ISA presented key scientific conclusions that
informed many policy-relevant questions, as detailed in Section 2. a subset of these conclusions was most
informative in the process of establishing the 2020 PM NAAQS (as discussed in Section 1.2). which is
under reconsideration. Specifically, extensive evidence spanning scientific disciplines supported the
conclusion of a causal relationship between both short- and long-term PM2 5 exposure and cardiovascular
effects and mortality. In addition, an assessment of those populations potentially at increased risk of a
PM-related health effect identified many populations and lifestages that experience both health risk and/or
exposure disparities with some of the strongest evidence being for non-White populations, with more
limited evidence for people of low socioeconomic status (SES). Lastly, in the assessment of welfare
effects, there is extensive evidence indicating a causal relationship between PM and visibility effects,
specifically visibility impairment. These topics are the basis of the scientific evaluation conducted in this
Supplement.
Recent studies published in the U.S. and Canada provide additional support for the conclusions of
the 2019 PM ISA. Overall: (a) recent studies support, and in some instances strengthen, the evidence
presented in the 2019 PM ISA; (b) many of the recent epidemiologic studies evaluated report positive
associations in areas with annual average or mean 24-hour avg PM2 5 concentrations similar to, or in many
cases lower, than those studies evaluated in the 2019 PM ISA; (c) some recent studies address key
scientific topics where the literature has evolved since the 2020 PM NAAQS review was completed,
specifically since the literature cutoff date for the 2019 PM ISA, including examining health effects at
near-ambient PM2 5 concentrations in experimental settings, and examining the association between PM2 5
exposure and COVID-19 infection and death, which both provide preliminary evidence to further inform
the PM2 5-health effects relationship; (d) recent studies support, and in some instances extend, the
evidence base indicating that non-White populations, specifically Black individuals, and low SES
individuals, experience disparities in both PM2 5-related health risks and exposures compared to
non-Hispanic White populations; and (e) recent studies further inform the role of PM on light extinction
and visibility impairment. In conclusion, the results of recent studies evaluated in this Supplement to the
2019 PM ISA support, and in some instances extend, the scientific conclusions of the 2019 PM ISA.
Overall, this Supplement to the 2019 PM ISA found the following:
September 2021
5-1
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
Cardiovascular Effects
Short-Term PM2.5 Exposure
• Recent U.S. and Canadian multicity studies conducted within diverse populations continue to
report positive associations between short-term PM2 5 exposure and emergency department (ED)
visits and hospital admissions for ischemic heart disease (IHD), myocardial infarction (MI) and
heart failure (HF) in studies with mean 24-hour avg PM2 5 concentrations ranging from
7.1-15.4 (ig/m3. Consistent with the evidence in the 2019 PM ISA, most recent studies report no
evidence of an association with stroke, regardless of stroke subtype. Further, recent
epidemiologic studies like those evaluated in the 2019 PM ISA often employed hybrid exposure
assessment models that allowed for a broader inclusion of geographic locations outside of the
traditional urban centers where ambient monitors are located. In addition, these studies report
evidence that continues to indicate an immediate effect of PM2 5 on cardiovascular-related
outcomes primarily within the first few days after exposure, and that associations generally
persisted in models adjusted for copollutants. Furthermore, recent epidemiologic studies that
conducted accountability analyses or employed causal modeling methods also report positive
associations across a number of statistical approaches, which further supports a relationship
between short-term PM2 5 exposure and cardiovascular effects.
Long-Term PM2.5 Exposure
• Consistent with studies evaluated in the 2019 PM ISA, some of the strongest evidence for
long-term PM2 5 exposure and cardiovascular effects comes from epidemiologic studies
examining cardiovascular mortality. Recent studies report consistent, positive associations for
cardiovascular mortality, specifically IHD and stroke mortality, across different cohorts at
varying spatial scales and across different exposure assessment and statistical methods with the
majority having annual PM25 concentrations ranging from 8.6 to 13.7 (ig/m3. In addition, recent
studies of cardiovascular morbidity, specifically coronary heart disease (CHD), stroke, and
atherosclerosis progression, most consistently report positive associations when focusing on
individuals with preexisting diseases and among patients followed after a cardiac event or
procedure, and not more diverse populations, which supports and extends the evidence presented
in the 2019 PM ISA. Recent studies of cardiovascular mortality and morbidity also indicate that
associations are relatively unchanged in copollutant models and that most assessments indicate a
linear, no-threshold concentration-response (C-R) relationship with initial evidence of
non-linearity at lower concentrations for some outcomes. Lastly, a few recent epidemiologic
studies that employed causal modeling methods, reduce some uncertainties related to potential
confounding bias and further support a relationship between long-term PM2 5 exposure and
cardiovascular effects.
Mortality
Short-Term PM2.5 Exposure
• Since the literature cut-off date for the 2019 PM ISA, relatively few multicity studies have been
conducted in the U.S. and Canada; however, these studies add to the extensive evidence based
evaluated in the 2019 PM ISA and previous assessments that reported consistent positive
associations across studies using different statistical models, exposure assessment approaches,
and methods for confounder control. These recent studies continue to report associations at mean
24-hour average concentrations ranging from 8.8 to 12.4 (ig/m3; support an immediate effect of
PM2 5 on mortality (i.e., at lag 0 to 1 day); and report that associations remain relatively
September 2021
5-2
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
unchanged in copollutant models. Additionally, there continues to be evidence that the
heterogeneity in city-to-city PM2 5-mortality risk estimates can partly be attributed to exposure
differences, such as housing characteristics. The assessment of the C-R relationship continues to
support a linear, no-threshold relationship with confidence in the shape at concentrations as low
as 5 (ig/m3. The relationship between short-term PM2 5 exposure and mortality is further
supported by recent epidemiologic studies that employed causal modeling methods and report
consistent, positive associations.
Long-Term PM2.5 Exposure
• Recent epidemiologic studies conducted in the U.S. and Canada consisting of cohorts with mean
annual PM2 5 concentrations generally below 12 (ig/m3, with the majority ranging from 5.9 to
11.65 (ig/m3, add to the large evidence base indicating consistent, positive associations between
long-term PM2 5 exposure and mortality detailed in the 2019 PM ISA. The reporting of consistent,
positive associations across studies examining various exposure windows, approaches for
confounder adjustment, and exposure assessment methods that used different sources of data and
were conducted at different spatial resolutions increases confidence in the relationship between
long-term PM2 5 exposure and mortality. In addition, recent studies further inform whether there
is evidence of copollutant confounding; although there were some differences across studies,
generally associations persisted in copollutant models. The assessment of the C-R relationship
continues to generally support a linear, no-threshold relationship with certainty down to 4 (ig/m3.
However, some uncertainties remain about the shape of the C-R curve at relatively low PM2 5
concentrations (<8 (.ig/ni3). with some recent studies providing evidence for either a sublinear,
linear, or supralinear relationship at these lower concentrations. Lastly, an extensive number of
epidemiologic studies that conducted accountability analyses or employed causal modeling
methods have been conducted since the literature cut-off date of the 2019 PM ISA; collectively,
these studies that used different statistical approaches and cohorts spanning diverse geographic
locations and populations provide additional support for the PM2 5-mortality relationship.
Additional Considerations Regarding the Health Effects of PM2.5
Experimental Studies at Near-Ambient PM2.5 Concentrations
• At the completion of the 2019 PM ISA, only a few controlled human exposure studies were
identified that had been conducted in Europe and examined health effects with near-ambient
PM2 5 concentrations (i.e., at concentrations around the 24-hour PM NAAQS of 35 (.ig/ni3). These
studies conducted in a population of older (55 years of age and older) overweight individuals
provided initial evidence for vascular changes and reductions in heart rate variability (HRV) at
low concentrations. A recent study in young, healthy participants also adds to the evidence base
indicating effects at near-ambient PM2 5 concentrations, specifically changes in lung function,
cardiac function, and inflammation. However, these results are inconsistent with the evidence for
both lung function and inflammation examined in controlled human exposure studies evaluated in
the 2019 PM ISA, which could be attributed to the higher ventilation rate and longer exposure
duration used compared to other studies.
COVID-19 Infection and Death
• With the onset of the COVID-19 pandemic, recent epidemiologic studies examined whether both
short-term and long-term PM2 5 exposure is associated with COVID-19 infection and death.
While some of these studies reported positive associations, these studies overall were subject to
methodological issues that may influence results. Specifically, many consisted of an ecological
September 2021
5-3
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
study design, studies were conducted during on ongoing pandemic while the etiology of
COVID-19 was still not understood (e.g., there are important differences in COVID-19 related
health such as by race and SES), and studies did not account for crucial factors that could
influence results (e.g., stay at home orders, social distancing, use of masks, and testing capacity).
Populations and Lifestages at Potentially Increased Risk of a PM-Related
Health Effect
Socioeconomic Status
• Recent studies that use a variety of metrics to represent SES including educational attainment and
income, along with studies that used composite metrics to represent neighborhood SES provide
additional support indicating there may be disparities in PM2 5 exposure and health risk by SES.
These studies indicate that the strongest evidence of a health risk disparity for low SES is for
cause-specific mortality and certain health endpoints (i.e., MI and CHF) when compared to higher
SES groups.
Race and Ethnicity
• Building upon the conclusions of the 2019 PM ISA, recent studies continue to support disparities
in PM2 5 exposure and health risks by race and ethnicity, with the strongest evidence for
non-White, specifically Black, populations. Black populations or individuals that live in
predominantly Black neighborhoods experience higher PM2 5 exposures, in comparison to
non-Hispanic White populations. Additionally, there is evidence of health risk disparities for both
Hispanic and non-Hispanic Black populations compared to non-Hispanic White populations for
cause-specific mortality and incident hypertension.
Visibility Effects
• Recent studies indicate that using contrast instead of light extinction as a metric greatly reduced
the variability in results in visibility preference studies. In addition, rapidly decreasing sulfate and
increasing fire related contributions to PM have led to a changing relationship between PM and
visibility impairment, likely impacting estimates of atmospheric extinction. In response, alternate
approaches to the application of the traditional IMPROVE algorithm for estimating atmospheric
light extinction have been developed that reduce the bias in atmospheric extinction estimates.
Parallel efforts have better characterized light extinction by major contributors to PM, particularly
for biomass burning.
September 2021
5-4
DRAFT: Do Not Cite or Quote
-------
Appendix A
Table A-1 Study-specific details for epidemiologic studies of short-term exposure to PM2.5 and cardiovascular
effects.
Study/Location,
Years
Study Population Exposure Assessment
Mean and Upper
Percentile
Concentrations
(|jg/m3)a
Outcome
Confounders
Considered
Copollutant
Examination
Wei et al. (2019)
Continental U.S.
2000-2012
Medicare
>65 yr
Daily concentration for
1 x 1 km grid cells using
validated satellite based
neural network model
calibrated using data from
monitors (n = 1,928), C-V
R2 = 0.83 overall,
0.78-0.88 by region (Pi et
al.. 2016)
Lag 0-1-day avg assigned
based on ZIP code of
residence.
NR
HA discharge data
recorded on Medicare
claims
Age, race, sex, Ml,
diet, time invariant
behavior factors,
ZIP code-level
SES, population
density, ethnicity,
access to parks,
food, drug stores,
day of week,
seasonality,
long-term trends.
Temperature
controlled using
cubic spline with up
to 9 df.
Correlation (r):
NA
Copollutant
models with: NA
Leiser et al. (2019)
Four counties in
Wasatch Front, UT
1999-2009
Medicare
n = 19,602 (2,032
cardiac events)
>65 yr
Daily average for ZIP code
centroids (n = 123) using
inverse distance weighting.
Lags 0, 1, 0-2, and
0-6 days.
Lag 0: 10.96
Lag 1: 10.95
3-day avg: 10.96
7-day avg: 10.79
Hospital re-admission
for IHD, Ml (ICD9 410),
HF (ICD9 428),
dysrhythmia/
arrhythmia (ICD 427)
Maximum daily
temperature, ZIP
code- level median
house-hold
income, Charlson
Comorbidity index,
enrollment in
Medicare part A
and B.
Correlation (r):
NA
Copollutant
models with: NA
September 2021
A-1
DRAFT: Do Not Cite or Quote
-------
Table A-1 (Continued): Study-specific details for epidemiologic studies of short-term exposure to PM2.5 and
cardiovascular effects.
Mean and Upper
Percentile
Study/Location, Concentrations Confounders Copollutant
Years Study Population Exposure Assessment (|jg/m3)a Outcome Considered Examination
Wing etal. (2017)
Nueces county, TX
2000-2012
BASIC
n = 317 events
>65 yr
Daily PM2.5 concentration
from centrally located
monitor. Lag 1 day.
Median: 7.7
IQR: 5.6 to 10.7
Passive and active
surveillance of IS
(i.e., acute onset
neurologic lasting
>24 h)
Temperature and
relative humidity,
individual level
characteristics and
time trends
controlled by
design.
Correlation (r):
NA
Copollutant
models with:
Ozone
Results didn't
change after
co-adjustment
Evans etal. (2017)
Monroe County, NY
2007-2012
Coronary syndrome/
unstable angina
patients (n = 362)
Mean age:
62.3 ± 12.9
Hourly average
concentrations, 1 monitor,
Eastside of Rochester
adjacent to two major
highways, patients reside
within 15 miles of monitor.
1 h avg.
Mean (IQR):
7.1 (7.62) all year;
8.18 (5.96)
Nov-Apr;
7.08 (7.80)
May-Oct
Upper (75th):
10.30 all year;
10.60 Nov-April
9.80 May-Oct
Upper (Max):
79.2 all year;
79.20 Nov-April;
64.01 May-Oct.
Physician-diagnosed
STEMI (>1 mm in >2
contiguous precordial
leads, or >2 adjacent
limb leads, or new or
presumed-new left
bundle branch block
with angina)
3 h avg temp, RH,
Copollutant
models with: NA
Correlation (r):
Delta-C: 0.31
BC: 0.57
NO2: 0.38
SO2: -0.01
Ozone: 0.11
CO: 0.29
September 2021
A-2
DRAFT: Do Not Cite or Quote
-------
Table A-1 (Continued): Study-specific details for epidemiologic studies of short-term exposure to PM2.5 and
cardiovascular effects.
Mean and Upper
Percentile
Study/Location, Concentrations Confounders Copollutant
Years Study Population Exposure Assessment (|jg/m3)a Outcome Considered Examination
Liu et al. (2020)
Calgary, Alberta
Canada
2004-2012
APPROACH
n = 6,142 HAs
among patients
admitted to
cardiology services
Hourly concentrations from Mean (SD): 9.79 Ml HAs captured in Temperature, RH Copollutant
3 monitors positioned to
represent background used
to compute daily average
PM2.5 concentration.
Long-term NO2
concentrations at ZIP code
of residential address
estimated using LUR
(Bertazzon et al.. 2015).
Lag 0, 1, 2 day and 3-,
5-day avgs.
(5.91)
Median: 7.00
IQR: 6.17
Upper: NR
registry of cardiac
catheterization patients
models with: NA
Correlation (r):
Os: -0.2
O3 max: 0.08
NO: 0.08
CO: 0.07
PM10: 0.43
Krall et al. (2018)
Atlanta, GA
(20 counties)
2002-2008
Birmingham, AL
(7 counties)
2004-2008
Dallas, TX
(12 counties)
2006-2008
Pittsburgh, PA
(3 counties)
2002-2008
St. Louis (16 counties)
MO and IL 2002-2007
Electronic billing
databases of ED
visits
Population-weighted
average estimates
estimated using 24 h avg
from ambient monitors
within each metropolitan
area with CMAQ predictions
(Friberq et al.. 2017: Friberq
et al.. 2016). A priori lag 0
with sensitivity analyses for
2-day avg.
Mean (SD):
Atlanta: 15.4 (7.1)
Birmingham: 14.5
(7.1)
Dallas: 10.8 (4.7)
Pittsburgh: 15
(8.6)
St. Louis: 13.8
(6.7)
IQR: 8.71
ED visits:
CHF (ICD9 428);
Cardiac dysrhythmia
(ICD9 427); IHD (ICD9
410-414); Stroke
(ICD9 433-437)
Weekday, season,
holidays,
meteorology (lag
Day 0 maximum
temperature, lag
Day 0-2 dew-point
temperature), the
hospitals reporting
data for each day,
cubic splines with
1 df for long-term
trends
Correlation (r):
Ozone; 0.47
EC: 0.5
OC: 0.68
SO4: 0.81
NH4: 0.77
(Correlations only
reported when
>0.4)
Copollutant
models with: NA
September 2021
A-3
DRAFT: Do Not Cite or Quote
-------
Table A-1 (Continued): Study-specific details for epidemiologic studies of short-term exposure to PM2.5 and
cardiovascular effects.
Mean and Upper
Percentile
Study/Location,
Concentrations
Confounders
Copollutant
Years
Study Population Exposure Assessment
(|jg/m3)a
Outcome
Considered
Examination
Ye et al. (2018)
24 h avg from 1 monitor
Mean (SD): 14.46
CVD ED visits
Temporal trends
Correlation (r):
Five counties in
(Jefferson Street). A priori
(7.69)
(i.e., IHD, Dysrhythmia,
and meteorology
CO: 0.47
Atlanta, GA
lag 0.
IQR: 9.28
and CHF)
(maximum
NO2: 0.50
SO2: 0.24
Aug 14, 1998-Dec 15,
75th: 18.21
temperature, cubic
function of
2013
(0.99, 1.01)
minimum and dew
Ozone: 0.44
Estimate from Fig 1
point temperature,
WS Fe: 0.65
day of week,
holiday, season,
Copollutant
hospital
models with:
participation
water-soluble Fe
period)
Fisher et al. (2019)
Contiguous U.S.
1999-2010
HPFS
Men
40-75 yr in 1986
n = 51,529
Validated kriging models to
estimate daily PM2.5
concentration (Liao et al..
2006) Data from >1,000
monitors used in model.
Lags up to 3 days prior to
stroke event and 4-day avg.
Mean (SD): 12.9
7.4
Self-reported stroke
adjudicated by
physician medical
record review Multiple
strokes at least 1 yr
apart included
OR Total Stroke
Lag 0-3 (avg):
0.94 (0.80, 1.10)
Temperature
controlled as a
linear term
(restricted cubic
splines in
sensitivity analysis)
with individual level
covariates and time
trends controlled
by design
Correlation (r):
NR
Copollutant
models with: NR
Lag 0
Lag 1
Lag 2
Lag 3
1.01 (0.90, 1.14)
0.92 (0.82, 1.03)
0.93 (0.82, 1.04)
1.00 (0.89, 1.12)
September 2021
A-4
DRAFT: Do Not Cite or Quote
-------
Table A-1 (Continued): Study-specific details for epidemiologic studies of short-term exposure to PM2.5 and
cardiovascular effects.
Mean and Upper
Percentile
Study/Location, Concentrations Confounders Copollutant
Years Study Population Exposure Assessment (|jg/m3)a Outcome Considered Examination
Sun et al. (2019)
U.S.
1993-2012
WHI
n = 5,417 confirmed
events
Post-menopausal
women age 50-79
at enrollment
Daily PM2.5 concentrations
estimated using Kriging
model (Liao et al.. 2006)
and assigned to residential
address of participant. PEs
0-0.27; SPEs: -0.11-0.04;
RMSSs~1.Lag 0, 0-1, and
0-2-day averages.
Sensitivity analyses to
evaluate 4, 5, and 6-day
moving averages.
Median: 10.8
75th: 15.3
95th: 26.1
IQR: 8.2
Self-reported stroke
(i.e., rapid onset of
persistent neurologic
deficit lasting >24 h).
Stroke type
Temperature and
RH, individual level
characteristics and
time trends
controlled by
adjudicated by trained design,
neurologists
Correlation (r):
PM10: 0.57
NO2: 0.44
NOx: 0.35
SO2: 0.30
Ozone: 0.20
Copollutant
models with: NR
Wvatt et al. (2020c)
U.S. nationwide
2008-2014
USRDS
Hemodialysis
patients
n = 351,294
Satellite derived PM2.5
concentration estimates
(AOD) integrated with
chemical transport model
predictions, meteorology,
land use variables for 1 km
grid cells (Pi et al.. 2016).
Lag 0 and examination
using unconstrained
distributive lag model.
Mean: 9.3
Range:
0.05-155.16
CVD, dysrhythmia, HF Temperature and Correlation (r):
30-day hospital
readmissions
(ICD-9 codes:
401-405, 410-411,
413, 426-27, 428)
RH. Individual level
characteristics,
county
characteristics
(i.e., population
size, risk
Characteristics)
day of the week,
seasonal and
long-term time
trends controlled
by design.
NR
Copollutant
models with: NR
September 2021
A-5
DRAFT: Do Not Cite or Quote
-------
Table A-1 (Continued): Study-specific details for epidemiologic studies of short-term exposure to PM2.5 and
cardiovascular effects.
Study/Location,
Years
Study Population Exposure Assessment
Mean and Upper
Percentile
Concentrations
(|jg/m3)a
Outcome
Confounders
Considered
Copollutant
Examination
deSouza et al. (2021) Medicaid adults (low Daily average at ZIP code Mean (SD) 11.5 First HA for CVD (ICD9 Air and dew-point Copollutant
Continental U.S.
2000-2012
income and/or
disabled)
n = 3,666,657 CVD
HAs
of residence. Predictions for
1 km2 grid cells integrated
remote sensing, chemical
transport model outputs,
meteorological and
land-use variables.
Ensemble model integrated
machine learning
algorithms. C-V R2 = 0.86
(Pi et al.. 2019). Lag
Day 0-1 used in all
analyses.
(7.3)
390-495); IHD (ICD9 temperature (daily correlation (r):
410-414); CHF (ICD9
428); AMI 410.9; IS:
(ICD9 434.91)
mean for 32 km2
grid cells in U.S;
individual level
factors, seasonality
and long-term
trends controlled
by design.
NR
Copollutant
models with:
Ozone
McClure et al. (2017)
Continental U.S.
Exposure: 2003-2011
Outcome: 2003-2007
to 2011
REGARDS
n = 30,239
n = 746 events
Integrated measurements
from monitors and satellite
derived PM2.5
concentrations (AOD)
estimated for 10 * 10 km
grid cells across the U.S.
(Al-Hamdan et al.. 2014).
Exposure assigned based
on residential address. Lags
0, 1, 2 days.
<12: n = 403
(53%)
12-150.4: n = 343
(47%)
Stroke (i.e., rapid onset
of persistent neurologic
deficit lasting >24 h)
defined by WHO
criteria (WHO. 1989)
ascertained by
self-report and medical
record review
Temperature and
RH with individual
characteristics and
time trends
controlled by
design
Copollutant
correlation (r):
NR
Copollutant
models with: NR
AMI = acute myocardial infarction; APPROACH = Alberta Provincial Project for Outcome Assessment in Coronary Heart Disease; Avg = average; h = hour; LUR = land use
regression; HA = hospital admission; STEMI = ST segment elevated myocardial infarction; RH = relative humidity; IHD = ischemic heart disease; CHF = congestive heart disease;
ICD-9 = International Classification of Disease 9th revision; HPFU = Health Professionals Follow-up Study; CMAQ = Community Multi-Scale Air Quality; REGARDS = REasons for
Geographic And Racial Differences in Stroke; SPE = standardized prediction error; PE = prediction error; RMSS = root mean square standardized; BASIC = Brain Attack Surveillance
in Corpus Christi: USRDS = U.S. Renal Data System; WS Fe = water soluble iron.
September 2021
A-6
DRAFT: Do Not Cite or Quote
-------
Table A-2 Study-specific details for epidemiologic studies using accountability analyses or causal modeling
methods to examine short-term exposure to PM2.5 and cardiovascular effects.
Study/Location
Years
Population
(Cohort)
Outcome
Exposure Assessment
PM2.5 Concentration
(Hg/m3)
Confounders
Copollutant
Examination
Albany, Bronx,
Buffalo,
Manhattan,
Queens, and
Rochester, NY
2005-2016
New York
Total
State
cardiovascular
Department of
disease; cardiac
Health
arrhythmia;
Statewide
cerebrovascular
Planning and
disease; chromic
Research
rheumatic heart
Cooperative
disease;
System
congestive heart
(SPARCS)
disease;
hypertension;
n = 1,922,918
ischemic heart
disease;
myocardial
infarction;
ischemic stroke
U.S. EPA Air Quality
System for each of the six
sites: Buffalo, Rochester,
Albany, and New York City
(Bronx, Manhattan,
Queens). Hourly PM2.5
concentrations were
measured using tapered
element oscillating
microbalance monitors and
24-h daily averages were
computed for each site and
each day for which
measurements of at least
75% of the hours that day
in that site were available
(18 h).
Case Period Median (25th,
75th Percentile):
Albany: 7.2 (4.4, 11.4)
Bronx: 8.8 (5.7, 13.8)
Buffalo: 8.5 (5.6, 12.6)
Manhattan: 10.5 (7.2,
15.3)
Queens: 8.0 (5.2, 12.5)
Rochester: 7.2 (4.5, 11.0)
Control Period Median
(25th, 75th Percentile):
Albany: 7.2 (4.4, 11.3)
Bronx: 8.8 (5.7, 13.8)
Buffalo: 8.5 (5.6, 12.6)
Manhattan: 10.5 (7.2,
15.3)
Queens: 7.9 (5.2, 12.5)
Rochester: 7.1 (4.5, 10.9)
Temperature; relative
humidity
Correlation (r): NA
Copollutant models
with: NA
September 2021
A-7
DRAFT: Do Not Cite or Quote
-------
Table A-2 (Continued): Study-specific details for epidemiologic studies using causal modeling methods to
examine short-term exposure to PM2.5 and cardiovascular effects.
Study/Location
Years
Population
(Cohort)
Outcome
Exposure Assessment
PM2.5 Concentration
(Hg/m3)
Confounders
Copollutant
Examination
Qiu et al. (2020) Medicare
Connecticut,
Maine,
Massachusetts,
New Hampshire,
Rhode Island, and
Vermont
2000-2012
Acute
myocardial
infarction
n = 156,717
Congestive
heart failure
n = 207,774
Ischemic
stroke
n = 170,663
Acute myocardial
infarction;
congestive heart
failure; ischemic
stroke
Daily ambient levels of
PM2.5 (24 h averaged,
|jg/m3) and ozone (8-h
maximum, ppb) were
predicted at a spatial
resolution of 1 km from a
machine learning algorithm
that combined satellite
remote censoring data,
chemical transport models,
land use, and meteorology,
using a neural network.
Daily averaged values
were constructed by
averaging the exposure
levels of all grid cells within
each individual ZIP code
Mean (SD):
Acute myocardial
infarction: 10.13 (6.48)
Congestive heart failure:
10.08 (6.42)
Ischemic stroke: 10.10
(6.47)
Temperature, relative
humidity
Age, sex, race,
smoking history,
cholesterol, BMI,
preexisting medical
conditions
Correlation (r): NA
Copollutant models
with: O3
Wang et al. (2019)
Rochester, NY
2005-2016
University of
Rochester
Medical Center
Cardiac
Catherization
Laboratory
n = 921
ST segment
elevation
myocardial
infarction (STEMI)
New York State
Department of
Environmental
Conservation air quality
monitoring site, where
PM2.5, SO2, O3, Co, and
black carbon were
measured continuously
throughout the study
period (2005-2016) for
patients residing within
15 miles of the monitoring
station
Mean (SD) for All Years
(2005-2016): 8.32 (7.17)
Mean (SD) for Before
(2005-2007): 10.37 (8.74)
Mean (SD) for During
(2008-2013): 8.31 (6.83)
Mean (SD) for After
(2014-2016): 6.67 (5.55)
Temperature and
relative humidity
Correlation (r): NA
Copollutant models
with: SO2, O3, CO,
BC
September 2021
A-8
DRAFT: Do Not Cite or Quote
-------
Table A-3 Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and cardiovascular
effects.
Mean and
Upper
Percentile
Study/Location Exposure Concentration Confounders
Years Population (Cohort) Outcome Assessment (|jg/m3) Considered Copollutant Examination
Rhinehart et al. (2020)
Allegheny County, PA
Jan 1, 2007
Rhinehart et al. (2020)
Dec 1, 2017
Resident of
Allegheny County,
PA and diagnosed
with AF (ECG,
ablation,
cardioversion [ICD10
427.31], Excluded
those <18 yr,
reporting history of
stroke, cardiothoracic
surgery or with no
record of follow-up.
n = 31,414
Diagnosis of
ischemic stroke
on uniform
electronic
health record
Exposure: 1-yravg
assigned across entire
study period
Spatial saturation
monitoring campaign
(June-July 2012 and
Jan-Mar 2013) and
LUR to estimate 1 -yr
avg concentration
within a 300 m buffer of
participants geocoded
address.
Mean: 10.6
Q4 (range):
11.11-15.74
Age, sex, race,
neighborhood
level income and
education
Correlation (r): NA
Copollutant models with:
NA
September 2021
A-9
DRAFT: Do Not Cite or Quote
-------
Table A-3 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
cardiovascular effects.
Mean and
Upper
Percentile
Study/Location Exposure Concentration Confounders
Years Population (Cohort) Outcome Assessment (|jg/m3) Considered Copollutant Examination
Chen et al. (2020)
Ontario Canada
Outcome: 2001-2016
Exposure: 2000-2016
ONPHEC (adults
35-85 yr residing
>5 yr in Ontario
Canada). Excluded
prevalent cases of
AMI
AMI
ascertained
using hospital
discharge
records
(validation
study: 89%
sensitivity/93%
specificity)
Annual PM2.5
concentration assigned
at centroid of postal
code of residence
reported for each year.
AOD and PM2.5
simulated by the
GEOS-Chem chemical
transport model. PM2.5
measurements
incorporated via
geographically
weighted regression.
Final surface with
1 x 1 km resolution for
Ontario (R
square = 0.76 with
ground level
measurements)
Mean: 8.61
See Figure 1
(Chen et al..
2020)
Age, sex,
income,
education
attainment,
percentage of
recent
immigrants,
unemployment
rate, urban/rural
residency,
indicators for
Greater Toronto
Area and
north/south
Correlation (r):
BC: 0.97
Mineral Dust: 0.94
See salt: 0.87
NOs: -0.87
Ammonium: 0.85
OM: 0.62
SO4: 0.73
Weaver et al. (2021)
2000-2004 to 2005-2008
JHS
n = 5,306 African
American adults
20-95 yr at baseline
(2000-2004)
Hypertension
(BP >140
diastolic BP
>90 mmHg) or
self-report of
hypertensive
medication use
Annual and 3-yr avg
PM2.5 concentration
estimated for geocoded
address. U.S. EPA
Bayesian space-time
downscaling fusion
model to estimate
census tract level PM2.5
concentration (Berrocal
et al.. 2012. 2010a. b).
Visit 1:
Annual average:
12.2 (IQR: 0.8)
3-yr avg: 12.4
(IQR: 0.4)
Visit 2:
12.1 (IQR: 0.8)
3-yr avg: 12.4
(IQR: 0.4)
Education,
smoking status,
nutritional status,
physical activity,
NSAID use, date
of measurement,
sex, BMI,
neighborhood
SES, age, and
food environment
within 1.5 miles
Correlation (r):
Os
Visit 1: PM2.5 (1 -yr);
PM2.5 (3-yr)
1 -yr: 0.12; 0.05
3-yr avg: -0.06; -0.03
Visit 2:
1 -yr: 0.07; 0.19
3-yr avg: -0.32; -0.36
Copollutant models
with: O3
September 2021
A-10
DRAFT: Do Not Cite or Quote
-------
Table A-3 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
cardiovascular effects.
Study/Location
Years
Population (Cohort) Outcome
Exposure
Assessment
Mean and
Upper
Percentile
Concentration
(Hg/m3)
Confounders
Considered
Copollutant Examination
Weaver et al. (2019)
Durham, Wake, or Orange
counties, NC
Exposure: 2000-2010
Outcome: 2001-2010
CATHGEN
n = 2,254 cardiac
catheterization
patients residing in 3
NC counties.
CAD index >23
(indicating 75%
coronary
vessel
occlusion)
Ml ascertained
using medical
records
Annual average PM2.5
concentration prior to
index case.
Daily mean PM2.4
estimated at 1 km
spatial resolution using
AOD with chemical
transport model, land
use variables and
meteorology (Pi et al..
2016)
Annual average:
12.7 (SD: 1.1)
Age, sex, BMI,
race, and
smoking status
Correlation (r): NR
Copollutant models
with: NR
Loop et al. (2018) REGARDS
U.S. Nationwide n = 17,126 Black and
2003-2007 (baseline) to 2012 White adults <-45 V
old)
Total CHD
(deaths and
nonfatal Ml
combined),
non-fatal Ml
Annual average PM2.5
concentration linked to
geocoded residential
address at baseline.
Daily PM2.5
concentration
estimated for 10 * 10
km grid using
ground-level monitoring
data, satellite
measurements of AOD.
Median: 13.6
75th: 14.8
1-yr mean
temperature,
season, race,
region,
urbanicity,
income,
education, age,
gender,
pack-years, BMI,
alcohol use,
physical activity,
and calendar
year
Correlation (r): NR
Copollutant models
with: NR
September 2021
A-ll
DRAFT: Do Not Cite or Quote
-------
Table A-3 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
cardiovascular effects.
Study/Location
Years
Population (Cohort) Outcome
Exposure
Assessment
Mean and
Upper
Percentile
Concentration
(Hg/m3)
Confounders
Considered
Copollutant Examination
Honda et al. (2017)
Exposure: 1980-2010
Outcome: 1993-1998
(recruitment) to 2010
WHI
n = 44,255
post-menopausal
women (average age
62 yr)
First self-report
of medication
for
hypertension
(SBP >140 mm
Hg, or DBP
>90 mm Hg)
Annual moving average Mean: 13.2
Age, BMI,
PM2.5 concentration
education,
estimated with daily
ethnicity,
concentrations from the
smoking status,
AQS and IMPROVE
physical activity,
networks using a
sodium intake,
universal kriging model
neighborhood
(CV R2 = 0.88)
SES, household
income,
employment
status, insurance
status, history of
high cholesterol,
history of
cardiovascular
disease, history
of diabetes,
clinical trial study
arm, and WHI
clinical site
Correlation (r): PM10:
0.56; PM10-2.5: 0.03
Copollutant models
with: NR
Duan et al. (2019b)
SWAN
cIMT by
Annual average 360
Mean: 16.5
Age, race,
Correlation (r):
Pittsburgh and Chicago, U.S.
n = 417 Black and
ultrasound and
days prior to clinic visit
(baseline)
education, SES,
Os:
White women (mean
age 51 yr at baseline)
plaque burden
(i.e., four
calculated from
monitors located within
75th: 17.1
(baseline)
BMI, and CVD
risk factors
Copollutant models:
with O3
levels, with 0
for no plaque
to 3 for a
plaque taking
20 km of residential
address
Daily PM2.5 values
retrieved from U.S.
(i.e., smoking,
total cholesterol,
HDL-c,
triglyceride,
up >50%
AQS
menopause
diameter of the
status, hormone
artery)
(Green et al.. 2015:
Ostro et al.. 2014)
use, fasting
glucose,
antidiabetic
medication, and
antihypertensive
medication)
September 2021
A-12
DRAFT: Do Not Cite or Quote
-------
Table A-3 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
cardiovascular effects.
Mean and
Upper
Percentile
Study/Location Exposure Concentration Confounders
Years Population (Cohort) Outcome Assessment (|jg/m3) Considered Copollutant Examination
Duan et al. (2019a)
Detroit, Ml; Oakland, CA;
Pittsburgh, PA; Chicago, IL;
and Newark, NJ.
Exposure: 1999-2005
Outcome: 2009-2013
SWAN
n = 1,188 women
Mean age 59.6 yr
Mean cIMT,
max cIMT, IAD
and plaque
presence or
severity
(i.e., four
levels, with 0
for no plaque
to 3 for a
plaque taking
up >50%
diameter of the
artery)
5-yr avg 360 days prior Mean (SD):
to each clinic visit 14.9(1.9)
calculated from 75^- <|g <|
monitors located within
20 km of residential Rgure 2
address
Daily PM2.5 values
retrieved from U.S.
AQS (Green et al..
2015: Ostro et al..
2014)
Age, race,
education, SES,
BMI, and CVD
risk factors
(i.e., smoking,
total cholesterol,
HDL-c,
triglyceride,
menopause
status, hormone
use, fasting
glucose,
antidiabetic
medication, and
antihypertensive
medication)
Correlation (r):
Os: 0.56
Copollutant models :
with O3
Keller et al. (2018)
Baltimore, MD
Jul 2000-Aug 2002 through
2012
Exposure: Feb and June
2012
MESA Air
n = 1,005
CAC
progression
(Agatston
units)
Spatiotemporal
prediction models used
to long-term average
PM2.5 concentration
from recruitment to
exam based on each
participants residential
history (Keller et al..
2015). C-V R2 = 0.84.
Mean (SD):
15.9 (0.80)
Age, sex,
race/ethnicity,
site, scanner
type, adiposity,
physical activity,
smoking,
employment,
total cholesterol,
high-density
lipoprotein level,
triglyceride level,
statin use, an
index of
neighborhood
SES, education,
and income
Correlation (r):
NR
Copollutant models: NR
September 2021
A-13
DRAFT: Do Not Cite or Quote
-------
Table A-3 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
cardiovascular effects.
Study/Location
Years
Population (Cohort) Outcome
Exposure
Assessment
Mean and
Upper
Percentile
Concentration
(Hg/m3)
Confounders
Considered
Copollutant Examination
Shin etal. (2019)
Ontario Canada
Exposure: 1998-2012
Outcome: Apr 2001-Mar
2015
ONPHEC
Ontario residents
(adults 35-85 yr)
n = 5,071,956
(those with history of
AF and stroke
excluded)
AF and Stroke
(1st hospital
admission for
ischemic or
hemorrhagic
stroke)
ascertained
using
administrative
databases
5-yr moving average
PM2.5 concentrations
estimated at residence
using AOD and PM2.5
simulated by the
GEOS-Chem chemical
transport model. Final
surface with 1 * 1 km
resolution for Ontario
(R square = 0.82 for
2004x2008 5-yr mean)
Mean: 9.8 (SD:
2.9)
IQR: 4.0
75th: 11.9
Max: 20
Age, sex, area
level risk factors
including SES,
and geographic
indicator
variables to
distinguish
whether a
participant's
residence was
located in the
north or south of
Ontario and
whether it was
urban or rural
Correlation (r):
NO2: 0.65
O3 0.275
Ox: 0.668
Copollutant models: with
NO2, O3, Ox
Baietal. (2019)
Exposure: 1998-2012
Outcome: Apr 2001-Mar
2015
ONPHEC
Ontario residents
(adults 35-85 yr)
n = 5,062,146 CHF
n = 5,141,172 AMI
(those with history of
CHF and Ml
excluded)
CHF and AMI
ascertained
using registries
based on
hospital
discharge data
3-yr moving
concentrations
assigned to postal code
participant residence
estimated using AOD
and PM2.5 simulated by
the GEOS-Chem
chemical transport
model. Final surface
with 1 x 1 km
resolution for Ontario
Mean: 9.6 (2.8)
IQR: 3.5
75th: 11.4
Max: 20
Age, sex, area
level risk factors
including SES,
and geographic
indicator
variables that
distinguished
participants
based on the
location of their
residence,
i.e., north or
south of Ontario
and urban vs.
rural
Correlation (r):
NO2: 0.4
O3 0.2
Ox: 0.4
Copollutant models: NR
September 2021
A-14
DRAFT: Do Not Cite or Quote
-------
Table A-3 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
cardiovascular effects.
Study/Location
Years
Population (Cohort) Outcome
Exposure
Assessment
Mean and
Upper
Percentile
Concentration
(Hg/m3)
Confounders
Considered
Copollutant Examination
Elliott et al. (2020)
Contiguous U.S.
Exposure: 1988-2007
Outcome: 1988-2008
NHS
n = 104,990
Women (35-55 at
baseline in 1976)
Ml (ICD-9 410) 24-mo moving average 13.7
and Stroke
(ICD-9
430-437)
Self-reported
physician
diagnosed
confirmed by
medical record
review
PM2.5 at residence
using spatiotemporal
prediction models for
contiguous U.S. (CV
R2: 0.77) (Yanoskv et
al.. 2014b).
(1988-2008)
80th: 16.5
Preexisting
health conditions,
smoking status,
healthy eating,
alcohol use,
income, and
education
Correlation (r):
NR
Copollutant models: NR
Danesh Yazdi et al. (2019)
Southeastern U.S. (Florida,
Alabama, Mississippi,
Georgia, North Carolina,
South Carolina, and
Tennessee)
2000-2012
Medicare data
n = 11,084,660
Older adults (65 plus
yr)
First hospital
admission for
Ml (ICD-9:
410), CHF
(ICD-9 428)
and Stroke
(ICD-9
430-438),
ascertained
using Medicare
data
Annual average PM2.5
concentrations
predicted using AOD
from satellites, land
use, and chemical
transport models to
assign daily exposure
in 1 x1 km grid cells (Di
et al.. 2016)
Cross validated
R2 = 0.86
State, sex, race,
year, eligibility for
Medicaid, and
census measures
of SES
Correlation (r):
NR
Copollutant models: NR
AF = atrial fibrillation; AMI = Acute Myocardial Infarction; AOD = aerosol optical depth; CAC = coronary artery calcium; CATHGEN = Catheterization Genetics study;
CHF = congestive heart failure; JHS = Jackson Heart Study; MESA = Multi-Ethnic Study of Atherosclerosis and Air Pollution; IAD = inter-adventitial diameter; NHS = Nurses Health
Study; ONPHEC = Ontario Population Health and Environment Cohort; Ox = redox weighted average of N02 and 03; REGARDS = REasons for Geographic and Racial Differences
in Stroke; SWAN = Study of Women's Health Across the Nation.
September 2021
A-15
DRAFT: Do Not Cite or Quote
-------
Table A-4 Study-specific details for epidemiologic studies using accountability analyses or causal modeling
methods to examine long-term exposure to PM2.5 and cardiovascular effects.
Study/Location
Years
Population
(Cohort)
Outcome
Exposure Assessment
PM2.5 Concentration
(Hg/m3)
Confounders
Copollutant
Examination
Danesh Yazdi et Medicare
al. (2021)
U.S.
2000-2016
n
63,006,793
Myocardial
infarction;
ischemic stroke;
atrial fibrillation,
and flutter
High-resolution
spatiotemporal ensemble
models, each of which
combined estimates from 3
different machine learning
algorithms, including a
neural network, a gradient
boosting machine, and a
random forest. The models
used hundreds of
predictors including land
use terms, chemical
transport model
predictions, meteorologic
variables, and satellite
measurements to estimate
daily levels of the
pollutants on a scale of
1 km x 1 km.
Mean: 10.21
Median: 10.05
Range: 0.01-30.92
Individual: sex, race, age,
Medicaid eligibility
ZIP code level: proportion of the
population >65 yr of age living
below the poverty line; population
density; median value of owner
occupied properties; proportion
of the population listed as Black;
median household income;
proportion of housing units
occupied by the owner;
proportion of the population
identified as Hispanic; proportion
of the population >65 yr of age
who had not graduated from high
school; lung cancer
hospitalization rate; mean BMI;
smoking rate; proportion of
Medicare beneficiaries with at
least 1 hemoglobin A1 c test in a
year; proportion of elderly
diabetic beneficiaries who had a
lipid panel test in a year;
proportion of beneficiaries who
had an eye examination in a
year; proportion of beneficiaries
with at least 1 ambulatory doctor
visit in a year; and proportion of
female beneficiaries who had a
mammogram during a 2-yr
period; region of residence
Correlation (r): NA
Copollutant models
with: O3, NO2
September 2021
A-16
DRAFT: Do Not Cite or Quote
-------
Table A-4 (Continued): Study-specific details for epidemiologic studies using causal modeling methods to
examine long-term exposure to PM2.5 and cardiovascular effects.
Study/Location population
Years (Cohort)
Outcome
Exposure Assessment
PM2.5 Concentration
(Hg/m3)
Confounders
Copollutant
Examination
Henneman et al. Medicare
(2019)
U.S.
2005-2012
Acute
myocardial
infarction;
cardiovascular
stroke; heart
failure; heart
rhythm
disorders;
ischemic heart
disease; all
cardiovascular
disease
Modeled national PM2.5
concentrations estimated
by combining GEOS-Chem
chemical transport
model-simulated PM2.5,
satellite-retrieved aerosol
optical depth, and
observed PM2.5
concentrations. Estimated
monthly 0.1 longitude * 0.1
latitude PM2.5
concentrations were
averaged to annual
concentrations and
spatially overlayed on U.S.
ZIP codes.
Coal exposure: HYSLPLIT
air parcel trajectory and
dispersion model
Mean (IQR):
2005: 10.0 (4.5)
2012: 7.2 (2.6)
Difference between
2012 and 2005: -3.2
(2.4)
Median age; median household
income; per-capita income; sex;
race; fraction of population by
county that smoked in 2000;
temperature; specific humidity
Correlation (r): NA
Copollutant models
with: NA
September 2021
A-17
DRAFT: Do Not Cite or Quote
-------
Table A-5 Study-specific details for epidemiologic studies of short-term exposure to PM2.5 and mortality.
Study/Location
Years
Study
Population
Mortality
Outcome
Exposure Assessment
Confounders in
Statistical
Model
Mean and Upper
Percentile
Concentrations
(Hg/m3)
Copollutant
Examination
Liu etal. (2019)
652 cities
worldwide; 107
U.S. cities; 25
Canadian cities
(U.S.:
1987-2006;
Canada:
1986-2011)
All ages
All cause Available PM2.5 data in the MCC
database. Measurements for air
pollutants from fixed-site monitoring
networks operated by local authorities
for each country.
DOW
RH
Temperature
Year
U.S.: 12.4
Canada: 9.3
Correlation (r): NA
Copollutant models
with: NA
Laviqne et al.
(2018)
U.S.
24 cities in
Canada
1998-2011
All ages
Nonaccidental:
n = 1,179,491
Cardiovascular:
n = 401,719
Respiratory:
n = 105,980
Nonaccidental
Cardiovascular
Respiratory
Daily (24-h) average PM2.5
concentrations from monitors in
Canada's NAPS network were used to
estimate exposures. Exposure
estimates were assigned to each study
participant based on the monitoring
station(s) located in participants' city of
residence. If PM2.5 measurements were
available from multiple monitors in a
single city, daily concentrations were
averaged across monitors.
RH (lag 0-2) Mean: 8.8
Temperature (lag Max: 98.2
0-2)
Correlation (r):
NO2: 0.53
Os: 0.03
Ox: 0.28
Copollutant models
with: Ox
Shin et al.
(2021b)
Canada
24 cities
2001-2012
>1 yr of age
65+
Respiratory Daily (24-h) average PM2.5
concentrations were calculated for each
study city using ambient monitoring data
available from Canada's NAPS. Daily
PM2.5 concentrations were averaged
across monitors within a city when
multiple monitoring sites were present.
DOW
Temperature (lag
0)
Year
Mean:
Warm
(April-September): £
Cold
(October-March): 6
Correlation (r): NA
Copollutant models
with: NA
September 2021
A-18
DRAFT: Do Not Cite or Quote
-------
Table A-5 (Continued): Study-specific details for epidemiologic studies of short-term exposure to PM2.5 and
mortality.
Study/Location
Years
Study
Population
Mortality
Outcome
Exposure Assessment
Confounders in
Statistical
Model
Mean and Upper
Percentile
Concentrations
(Hg/m3)
Copollutant
Examination
Shin et al.
(2021a)
Canada
22 cities
2001-2012
>1 yr of age
65+
All cause Daily (24-h) average PM2.5
concentrations were calculated for each
study city using ambient monitoring data
available from Canada's NAPS. Daily
PM2.5 concentrations were averaged
across monitors within a city when
multiple monitoring sites were present.
DOW
Temperature (lag
0)
Year
Mean
Warm
(April-September): £
Cold
(October-March): 6
Correlation (r):
O3 warm: 0.4
O3 cold: -0.3
Copollutant models
with: NA
Rappazzo et al. 18-64 yr of age Nonaccidental
(2019)
North Carolina
March
2013-February
2015
n = 399
out-of-hospital
sudden
unexpected
deaths
Hourly PM2.5 measurements from the
Wake County central site monitor were
obtained though the U.S. EPA's AQS
data mart. Daily concentrations were
calculated by averaging hourly
measurements over 24-h (midnight to
midnight).
Temperature (lag Mean: 10.93
°) 75th: 13.22
RH (lag 0) Max: 31.14
Correlation (r):
CO: 0.45
NO2: 0.36
Os: 0.22
SO2: 0.18
Copollutant models
with:
CO, NO2, Os, SO2
Baxter et al.
(2019)
U.S.
312 CBS As
1999-2005
All ages
Nonaccidental Daily (24-h) mean PM2.5 concentrations DOW
from population-based monitors were
obtained from the U.S. EPA's AQS.
Concentrations were averaged across
monitors within a CBSA/MD when
multiple monitoring sites were present
as detailed in Baxter et al. (2017).
Dew Point
Temperature (lag
0)
Temperature (lag
0; lag 1, 2, 3)
Year
Correlation (r): NA
Copollutant models
with: NA
AQS = air quality system; CBSA = core-based statistical area; DOW = day of week; MCC = Multi-City Multi-Country Collaborative Research Network; MD = metropolitan division;
NAPS = National Air Pollution Surveillance System.
September 2021
A-19
DRAFT: Do Not Cite or Quote
-------
Table A-6 Study-specific details for epidemiologic studies using causal modeling methods to examine
short-term exposure to PM2.5 and mortality.
PM2.5
Study/Location Population Exposure Concentrations
Years (Cohort) Outcome Assessment (|jg/m3) Confounders Copollutant Examination
Wei et al. (2020)
Massachusetts
2000-2012
Medicare
n = 1,503,572
All-cause mortality
Predicted daily
ambient PM2.5,
ozone, and
nitrogen dioxide
levels in each
1-km x 1-km grid
cell across the
contiguous U.S.
using well-validated
ensemble models
Mean (SD): 8.9 (5.4)
Range: 0.1-65.3
Temperature-air and dew
point
Individual: sex; race/ethnicity;
age; Medicaid eligibility
ZIP code level: annual
median household income;
median value of
owner-occupied housing
units; percentage of
population living in poverty;
percentage of the population
with less than a high school
education; population density;
home ownership rate
County level: annual
percentage of ever smokers;
percentage of obese people
Correlation (r): NA
Copollutant models with: O3,
NO2
September 2021
A-20
DRAFT: Do Not Cite or Quote
-------
Table A-6 (Continued): Study-specific details for epidemiologic studies using causal modeling methods to
examine short-term exposure to PM2.5 and mortality.
Study/Location population
Years (Cohort)
Outcome
Exposure
Assessment
PM2.5
Concentrations
(Hg/m3)
Confounders
Copollutant Examination
Wei et al.
(2021b)
Massachusetts
2000-2012
Medicare
n = 1,503,572
All-cause mortality
Daily
concentrations of
ambient PM2.5,
ozone, and
nitrogen dioxide at
1 km x 1 km grid
cells were
predicted using
geographically
weighted
regressions that
ensembled
predictions from
ensemble methods
Mean (SD): 8.9 (5.4)
Range: 0.1-65.3
Air surface temperature, dew
point temperature, and
relative humidity
Individual: sex, race, age,
Medicaid eligibility
Zip code level: median
household income, median
house value, percent of
owner-occupied homes,
percent of population living in
poverty, percent of population
below high school education,
population density, percent of
Blacks, and percent of
Hispanics, percent of persons
over age 65 with an annual
hemoglobin A1c test, an
annual low-density lipoprotein
test, and an annual eye exam
in each hospital catchment
area
County level: percent of ever
smokers, lung cancer rate,
and average BMI
Correlation (r): NA
Copollutant models with: O3,
NO2
Schwartz et al.
National
Mortality (daily
Obtained PM2.5 and
Mean: 12.8
Meteorologic: Daily mean
Correlation (r): NA
(2018a)
Center for
deaths)
NO2 from U.S.
25th Percentile:
7 5
temperature, wind speed, and
Copollutant models with: NO2
135 U.S. cities
Health
Statistics
n = 7,277,274
EPA's Air Quality
System
Technology
Transfer Network
75th Percentile:
16.1
sea-level pressure data
1999-2010
September 2021
A-21
DRAFT: Do Not Cite or Quote
-------
Table A-7 Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and mortality.
PM2.5
Study Mortality Confounders in Concentrations Copollutant
Study Population Outcomes Exposure Assessment Statistical Model (|jg/m3) Examination
Pope etal. (2019)
Contiguous U.S
PM2.5: 1999-2015
(1988-2015 in
sensitivity
analyses)
Follow-up:
1986-2015
NHIS
All cause
n = 1,599,329
Cardiopulmonary
deaths =
Cardiovascular
267,204
Cerebrovascular
18-84 yr of
Chronic lower
age
respiratory
Influenza/
pneumonia
Cancers
Influenza/
pneumonia
Lung cancer
Other/unknown
Primary analyses:
Population-weighted annual PM2.5
concentrations averaged over the
years 1999 to 2015 for census-tract
centroids. PM2.5 concentrations were
estimated from regulatory monitoring
data and constructed in a universal
kriging framework relying on
information from geographic
variables, including land use,
population, and satellite-derived
estimates of land use and air
pollution as described in Kim et al.
(2020). This method was shown to
have good model performance,
R2 = 0.78-0.90.
Sensitivity analyses: Census-tract
mean PM2.5 concentrations for the
years 1988 to 2015 were estimated
using imputed data from 1988-1998,
based on the relationship between
monitored PM10 and PM2.5
concentrations, and primary
modeled data from 1999-2015.
Approach is described in Kim et al.
(2020).
Age, sex,
race-ethnicity using
104 variables for all
interactive
combinations of 13 age
ranges, sex, and
race-ethnicity; income
(inflation-adjusted to
2015); education level;
marital status; rural
versus urban; U.S.
census region; survey
years
Subcohort analysis
also included smoking
status; BMI
Mean: 10.7
SD: 2.4
Range: 2.5-19.2
Correlation (r): NA
Copollutant models
with: NA
September 2021
A-22
DRAFT: Do Not Cite or Quote
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
mortality.
PM2.5
Study Mortality Confounders in Concentrations Copollutant
Study Population Outcomes Exposure Assessment Statistical Model (|jg/m3) Examination
Lefler et al.
(2019)
U.S.
PM2.5: 1988-2015
Follow-up:
1987-2015
NHIS—Subset All cause
n = 635,539
deaths =
106,385
18-84 yr of
age
Cardiopulmonary
Annual average PIVteswas modeled
using regulatory monitors and land
use data as described in Kim et al.
(2020). Exposure estimates were
assigned to home census tracts as
either 2-yr (i.e., cohort year and
previous year) or 5-yr (i.e., cohort
year and previous 4 yr) average
PM2 5 concentrations, 17-yr average
PM2.5 concentrations (1999-2015),
or 28-yr average PM2.5
concentrations (1988-2015).
Variables for all
interactive
combinations of 13 age
ranges, sex, and
race-ethnicity; marital
status;
inflation-adjusted
household income;
education; smoking
status; BMI; U.S.
census region; rural vs.
urban; suvey year
Mean: 10.67
SD: 2.37
IQR: 3.12
Correlation (r):
CO: 0.42
SO2: 0.41
PM10-2.5: 0.19
NO2: 0.56
Os: 0.33
Copollutant models
with: CO, PM10-2.5,
SO2, O3, NO2
Wang et al.
(2020)
U.S.
PM2 5: 2000-2008
Follow-up:
2000-2008
Medicare
n =
52,954,845
deaths =
15,843,982
65-120 yr of
age
Nonaccidental
Cardiovascular
IHD
Cerebrovascular
CHF
Respiratory
COPD
Pneumonia
Daily PM2 5 was estimated on a 6-km
grid using a spatiotemporal model
described in Yanoskv et al. (2014a).
Model inputs included monitored
PM25, meteorological and geospatial
covariates, and traffic-related PM
estimated using a Gaussian
line-source dispersion model.
Medicare beneficiaries were
matched to the grid point closest to
their ZIP code centroid and PM2.5
concentrations were averaged for
the 12-mo period prior to death.
Strata for age, race,
sex, and ZIP code; ZIP
code, and state SES
Mean: 10.3
SD: 3.2
Correlation (r):
NO2: 0.59
Os: 0.24
Copollutant models
with: O3
September 2021
A-23
DRAFT: Do Not Cite or Quote
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
mortality.
PM2.5
Study Mortality Confounders in Concentrations Copollutant
Study Population Outcomes Exposure Assessment Statistical Model (|jg/m3) Examination
Elliott et al. (2020) Nurses' Health Nonaccidental
U.S.
PM2.5: 1988-2007
Follow-up:
1988-2008
Study
n = 104,990
female
participants
deaths =
9,827
Average age
63.1 yr
24-mo average ambient PM2.5
exposures were estimated at
residential addresses using the
spatiotemporal prediction model
described in Yanoskv et al. (2014a).
Model predictions used publicly
available monitoring data, geospatial
predictors (road network data,
residential and urban land use,
density of PM2.5 and PM10
point-sources, elevation data) and
monthly average meteorological
data (windspeed, temperature,
precipitation).
Age; race; incident
cancer; family history
of Ml; smoking status;
pack-years; overall diet
quality; alcohol
consumption;
multivitamin use;
individual-level
Mean: 13.7
SD: 3.5
Correlation (r): NA
Copollutant models
with: NA
Wu et al. (2020a) Medicare
U.S.
PM2.5: 2000-2016
Follow-up:
2000-2016
All cause
n =
68,503,979
person-years
= 573,370,257
deaths =
27,106,639
65+ yr of age
An ensemble-based prediction
model was used to estimate daily
PM2.5 concentrations for a 1-km2
grid network across the contiguous
U.S. [discussed in Pi et al. (2019)1.
Grid cell predictions were
aggregated to ZIP codes, and
annual averages for each ZIP code
were calculated by averaging daily
concentrations. Annual average
PM2.5 concentrations in each ZIP
code were then assigned to
individuals who resided in that ZIP
code for each calendar year.
Age; race/ethnicity;
sex; Medicaid eligibility
County-level: average
BMI; smoking rate
ZIP code level:
proportion of Hispanic
and Black residents;
median household
income and home
value; proportion of
residents in poverty,
with high school
diploma, and own their
house; population
density; summer/winter
average max daily
temperature and RH
Geographic region of
U.S.; calendar year
Entire Cohort:
Mean: 9.8
SD: 3.2
Subset of cohort
<12.0:
Mean: 8.4
SD: 2.3
Correlation (r): NA
Copollutant models
with: NA
September 2021
A-24
DRAFT: Do Not Cite or Quote
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
mortality.
Study
Study
Population
Mortality
Outcomes
Exposure Assessment
Confounders in
Statistical Model
PM2.5
Concentrations
(Hg/m3)
Copollutant
Examination
Eum et al. (2018) Medicare
U.S. n =
PM2 5: 2000-2012 20,744,214
Follow-up: Dec deaths =
2000-Dec 2012 5,484,947
65+ yr of age
All cause
U.S. EPA Air Quality System (AQS)
monitors. Included monitors with at
least 8 calendar years of data,
having 9+ mos with 4+
measurements, which equaled 798
monitors. For each site, smoothed
time-series using linear regression
with thin plate splines with 4 df per
year. Gaps longer than 90 days
smoothed PM2.5 before and after
each gap separately. Predicted
values used to calculate moving
averages for PM2.5 for each month.
Yearly averages considered valid if
350+ days of data available.
Exposure assigned to individuals
that lived in ZIP codes with centroids
within 6 miles of a valid monitor.
County-level proportion
of non-Whites, current
smokers, diabetes,
asthma, individuals
possessing health care
plans, mean income,
and mean BMI
Mean: 11.65
SD: 3.09
Correlation (r): NA
Copollutant models
with: NA
September 2021
A-25
DRAFT: Do Not Cite or Quote
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
mortality.
Study
Study
Population
Mortality
Outcomes
Exposure Assessment
Confounders in
Statistical Model
PM2.5
Concentrations
(Hg/m3)
Copollutant
Examination
Crouse et al.
(2020)
Canada
PM2.5: 1998-2010
Follow-up:
2001-2011
CanCHEC
(2001)
n = 2,452,665
deaths =
191,555
25-89 yr of
age
Nonaccidental
Cardiovascular
Cardiometabolic
IHD
Cerebrovascular
Respiratory
Lung cancer
Base model: PM2.5 estimates at 1
km over 3-yr average (3-yr/1-km
model) with single-year lag assigned
to postal code of residence. As
detailed in van Donkelaar et al.
(2015) PM2.5 exposures derived from
AOD retrievals using GEOS-Chem
calibrated to surface measurements
by GWR.
Sensitivity analyses: Examined 1, 5,
and 10 km spatial scales and 1, 3,
and 8-yr temporal scales.
Age, sex
Community-level:
time-varying indicator
of size of each
subject's home
community from most
recent census;
CAN-Marg Index;
geographic airshed of
subject's residence
Individual-level:
aboriginal identity,
visible minority status,
marital status, highest
level of education,
employment status,
and household income
adequacy quintiles
1 -yr (Mean, Range)
1 km: 7.21 (0.0-20.0)
5 km: 6.59
(0.05-20.0)
10 km: 6.24
(0.21-20.0)
3-yr (Mean, Range)
1 km: 7.43
(0.00-20.0)
5 km: 6.79
(0.40-18.50)
10 km: 6.44
(0.60-18.16)
8-yr (Mean, Range)
1 km: 7.98
(0.35-18.36)
5 km: 7.27
(0.92-16.83)
10 km: 6.90
(0.79-16.40)
Correlation (r):
NO2: 0.49-0.62
Os: 0.44-0.58
Ox: 0.60-0.68
Copollutant
models with: O3,
NO2, Ox
September 2021
A-26
DRAFT: Do Not Cite or Quote
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
mortality.
Study
Study
Population
Mortality
Outcomes
Exposure Assessment
Confounders in
Statistical Model
PM2.5
Concentrations
(Hg/m3)
Copollutant
Examination
Pinault et al.
(2017)
U.S.
PM2.5:1998-2010
Follow-up:
5/15/1991-
12/31/2011
CanCHEC
(2001)
n = 2,448,500
deaths =
233,300
person-years
= 25,484,400
25-89 yr of
age
Nonaccidental
Cardiometabolic
Cardiovascular
IHD
Cerebrovascular
Respiratory
COPD
Pneumonia
Lung Cancer
PM2.5 estimates at 1 km over 3-yr
average (3-yr/1-km model) with
single-year lag assigned to postal
code of residence for years
1998-2012. As detailed in van
Donkelaar et al. (2015) PM2.5
exposures derived from AOD
retrievals using GEOS-Chem
calibrated to surface measurements
by GWR. PM2.5 concentrations
extended back to 1998 by applying
interannual variation of a published
PM2.5 data set (Bovs et al.. 2014).
Over North America R2 = 0.82 in
locations where PM2.5 <20 |jg/m3.
Age, sex
Individual-level:
aboriginal identity,
visible minority status,
marital status,
educational attainment,
income adequacy
quintile, labor force
status
Population center size,
airshed
Contextual: proportion
of persons 25 or older
who were unemployed,
proportion that had not
graduated high school,
proportion of persons
in low-income families
Mean (SD): 7.37
(2.60)
75th: 9.07
95th: 11.97
Max: 20.00
Correlation (r): NA
Copollutant models
with: NA
September 2021
A-27
DRAFT: Do Not Cite or Quote
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
mortality.
PM2.5
Study Mortality Confounders in Concentrations Copollutant
Study Population Outcomes Exposure Assessment Statistical Model (|jg/m3) Examination
Pappin et al.
(2019)
Canada
PM2.5: 1988-2015
Follow-up:
1991-2016
CanCHEC
(1991, 1996,
and 2001)
n = 8.5 million
deaths = 1.5
million
person-years
= 150,996,500
25-89 yr of
age
Nonaccidental PM2.5 estimates at 1 km over 3-yr
average (3-yr/1-km model) with
single-year lag assigned to postal
code of residence for years
1998-2012. As detailed in van
Donkelaar et al. (2015) PM2.5
exposures derived from AOD
retrievals using GEOS-Chem
calibrated to surface measurements
by GWR. For PM2.5 concentrations
prior to 1998, back casting method
employed that applied observed
trends in ground monitoring data for
PM2.5 to adjust pre-gridded PM2.5
estimates (Menq et al.. 2019).
Age, sex, immigrant
status
Contextual: CAN-Marg
Index, population size
of home
community/city, urban
form, regional airshed
Individual-level:
income, education,
occupational class,
Indigenous status,
visible minority status,
employment status,
marital status
Mean (SD):
1991
1996
2001
7.95 (3.28)
7.18 (2.70)
6.68 (2.24)
99th Percentile:
1991
1996
2001
17.26
15.00
12.30
Correlation (r): NA
Copollutant models
with: O3, NO2, Ox
Range (min-max):
1991
1996
0.37-20.00
0.37-20.00
2001: 0.37-18.50
Christidis et al.
(2019)
Canada
PM2.5: 1998-2015
Follow-up:
2000-2012
(linked to postal
code history
1981-2016)
mCCHS
n = 452,700
deaths =
50,700
person-years
= 4,452,700
Nonaccidental PM2.5 estimates at 1 km2 over 3-yr
average (3-yr/1-km model) with
single-year lag assigned to postal
code of residence. As detailed in van
Donkelaar et al. (2015) PM2.5
exposures derived from AOD
retrievals using GEOS-Chem
calibrated to surface measurements
by GWR. Spatial variation from
modeled surface used with simulate
PM2.5 and constrained with local
ground-based monitors to estimate
PM2.5 concentrations through 2015
as detailed in Menq et al. (2019).
Age, sex, immigrant
status, visible minority
identity, Indigenous
identity, marital status,
educational attainment,
employment status,
income quintile,
alcohol consumption,
smoking behavior, fruit
and vegetable
consumption, leisure
exercise frequency,
BMI, CAN-Marg Index,
ethnic concentration,
community/city size,
urban form, airshed
Mean (SD): 5.9
(2.0)
75th: 7.1
95th: 9.7
Max: 17.2
Correlation (r): NA
Copollutant models
with: O3, NO2, Ox
September 2021
A-28
DRAFT: Do Not Cite or Quote
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
mortality.
Study
Study
Population
Mortality
Outcomes
Exposure Assessment
Confounders in
Statistical Model
PM2.5
Concentrations
(Hg/m3)
Copollutant
Examination
Erickson et al.
CanCHEC
Nonaccidental
(2020)
(2001)
Cardiovascular
Canada
n = 3,101,605
Cardiometabolic
PM2.5: 1998-2106
immigrants (n)
IHD
Follow-up:
= 684,400
Cerebrovascular
5/15/2001-
deaths (non-
Respiratory
12/31/2016
immigrants) =
323,430
COPD
deaths
Lung cancer
(immigrants) =
87,775
PM2.5 estimates at 1 km2 over 3-yr
average (3-yr/1-km model) with
single-year lag assigned to postal
code of residence for years
1998-2012. As detailed in van
Donkelaar et al. (2015) PM2.5
exposures derived from AOD
retrievals using GEOS-Chem
calibrated to surface measurements
by GWR. Spatial variation from
modeled surface used with simulate
PM2.5 and constrained with local
ground-based monitors to estimate
PM2.5 concentrations through 2015
as detailed in Menq et al. (2019).
PM2.5 concentrations extended back
to 1998 by applying interannual
variation of a published PM2.5 data
set (Bovs et al.. 2014).
Age, sex, visible
minority status,
Aboriginal identity,
marital status,
educational attainment,
income adequacy
quintiles, employment
status, occupational
classification, age at
immigration,
geographic region of
birth, CAN-Marg Index,
community size, urban
form, regional airshed
Mean (SD)
Non-immigrant:
7.53 (2.65)
Immigrant:
Pre-1971: 9.13
(2.53)
1971-1980
(2.28)
1981-1990
(2.13)
1991-2001
(2.02)
9.28
9.54
9.69
Correlation (r): NA
Copollutant models
with: NA
Erickson et al.
(2019)
Canada
PM2.5: 1998-2012
Follow-up:
2001-2011
CanCHEC
(2001)
n = 2,468,190
CCHS
n = 80,630
Nonaccidental
Cardiovascular
IHD
Lung cancer
PM2.5 estimates at 1 km over 3-yr
average (3-yr/1-km model) with
single-year lag assigned to postal
code of residence for years
1998-2012. As detailed in van
Donkelaar et al. (2015). PM2.5
exposures derived from AOD
retrievals using GEOS-Chem
calibrated to surface measurements
by GWR.
Age, sex, visible
minority status,
Aboriginal status,
marital status,
educational attainment,
income quintile, labor
force status, CMA-size,
airshed, CAN-Marg
Index
CanCHEC
Mean: 8.40
CCHS:
Mean: 6.70
Correlation (r): NA
Copollutant models
with: NA
September 2021
A-29
DRAFT: Do Not Cite or Quote
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
mortality.
PM2.5
Study Mortality Confounders in Concentrations Copollutant
Study Population Outcomes Exposure Assessment Statistical Model (|jg/m3) Examination
Zhang et al.
(2021) Study
Ontario, Canada n = 88,615
PM2.5: 2000-2016 deaths =
Ontario Health Nonaccidental
Cardiovascular
Respiratory
Follow-up:
2009-2017
7,488
30+ yr of age
PM2.5 estimates at 1 km2 over 3-yr
average (5-yr/1-km model) with
single-year lag assigned to postal
code of residence. As detailed in van
Donkelaar et al. (2015) PM2.5
exposures derived from AOD
retrievals using GEOS-Chem
calibrated to surface measurements
by GWR. Additional methodological
changes including treating
topographical changes and urban
land cover as separate predictors
improved R2 to 0.80.
Age, sex, ethnicity,
place of birth,
educational level,
marital status, annual
household income,
BMI, daily intake of
fruits and vegetables,
physical activity,
smoking habit, drinking
habit, environmental
exposure to smoke,
urban residency,
percentage of recent
immigrants,
percentage of
population >15 with
educational attainment
lower than high school,
and percentage of
population >15
unemployed and
income quintile
Mean (SD): 7.1
(1.5)
75th: 8.8
Correlation (r): NO2
= 0.63
Copollutant models
with: NO2
Cakmak et al.
(2018)
Canada
PM2.5:1998-2011
Follow-up:
1991-2011
CanCHEC
(1991)
n = 2,291,250
deaths =
522,305
>25 yr of age
IHD
COPD
Lung cancer
PM2.5 estimates from median
satellite-derived concentrations from
1998-2011 at 10 km * 10 km
resolution as detailed in van
Donkelaar et al. (2010). Changes in
PM2.5 from 1998 to 2005 inferred
from MISR and SeaWiFS satellites
(Boys et al.. 2014). Seven-year
average exposures applied to each
participant.
Age, sex, aboriginal
ancestry, visible
minority status, marital
status, education level,
occupational level,
immigrant status,
income quintile
Smoking data and BMI
from CCHS
Mean (Range)
(SD):
3.8 (1.2)—7.4 (2.2)
Correlation (r): O3:
-0.007 to 0.698
Copollutant models
with: O3
September 2021
A-30
DRAFT: Do Not Cite or Quote
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
mortality.
Study
Study
Population
Mortality
Outcomes
Exposure Assessment
Confounders in
Statistical Model
PM2.5
Concentrations
(Hg/m3)
Copollutant
Examination
Haves et al.
(2020)
U.S.
PM2.5: 1980-2010
Follow-up:
1995-2011
NIH-AARP
Diet and
Health Study
n = 565,477;
deaths =
135,289
50-71 yr of
age
Nonaccidental
Cardiovascular
Respiratory
Spatiotemporal prediction model
detailed in Kim et al. (2017) that
provided mean annual estimates of
PM2.5 for each census tract. PM2.5
data prior to 1999 estimated using
extrapolation based on PM2.5 data in
FRM/IMPROVE; PM2.5 sulphate data
in Clean Air Status and Trends
Network; and visibility data across
the Weather-Bureau-Army-Navy
network. Annual average PM2.5
concentrations assigned at census
tract level lagged by 1 yr in
time-dependent manner.
Age, race/ethnicity,
education level, marital
status, BMI, alcohol
consumption, smoking
status, city/state,
characteristics of
census tract at
enrollment (median
income, percentage
not completing high
school)
Median: 13.3
Range: 2.9-28.0
Correlation (r): NA
Copollutant models
with: NA
Sensitivity analyses: (1) Follow-up
period starting in 2000; and
(2) distance to residence from AQS
PM2.5 monitoring site.
Chen et al. (2020) ONPHEC Cardiovascular
Ontario, Canada n = 5,264,985
PM2 5: 2000-2016 deaths =
Follow-up: 305,335
2001-2016 35-85 yr of
age
Annual average PM2.5 estimated at 1
km2 from multiple satellite retrievals
of AOD combined with geophysical
relationship between AOD and PM2.5
simulated by GEOS-Chem, which
were then calibrated with surface
measurements by GWR as detailed
in van Donkelaar et al. (2019).
Annual estimates closely agree with
ground measurements across North
America (R2 = 0.76).
Age, sex, urban/rural
residency, north/south,
percentage of
population >15 with
educational attainment
lower than high school,
percentage of recent
immigrants,
unemployment rate,
income quintile, urban
vs. other areas
Mean: 8.61
Correlation (r):
Organic mass: 0.62
Sulphate: 0.73
Ammonium: 0.85
Nitrate: 0.9
Sea salt: 0.87
Mineral dust: 0.92
Black carbon: 0.97
Copollutant models
with: NA
September 2021
A-31
DRAFT: Do Not Cite or Quote
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
mortality.
Study
Study
Population
Mortality
Outcomes
Exposure Assessment
Confounders in
Statistical Model
PM2.5
Concentrations
(Hg/m3)
Copollutant
Examination
Pinault et al.
CanCHEC
(2018)
(2001)
Canada
n = 2,448,500
CanCHEC
CVD deaths =
PM2.5:1998-2012
123,500
Follow-up:
5/15/2001-
mCCHS
12/31/2011
n = 270,600
mCCHS
CVD deaths =
PM2.5:1998-2012
12,400
Follow-up:
2001-2008
Cardiovascular PM2.5 estimates at 1 km2 over 3-yr
average (3-yr/1-km model) with
single-year lag assigned to postal
code of residence for years 1998 -
2012. As detailed in van Donkelaar
et al. (2015) PM2.5 exposures
derived from AOD retrievals using
GEOS-Chem calibrated to surface
measurements by GWR. PM2.5
concentrations extended back to
1998 by applying interannual
variation of a publishing PM2.5 data
set (Bovs et al.. 2014). Estimates
were correlated with ground
measurements, R2 = 0.82.
Age, sex, population
center size, airshed,
aboriginal identity,
visible minority status,
educational attainment,
labor force status,
income adequacy
quintile, percentage of
persons >25
unemployed or had not
graduated from high
school, overall
percentage of people
in low-income families
2001 CanCHEC
Mean (SE): 7.37
(2.60)
mCCHS
Mean (SE): 6.37
(2.65)
Correlation (r): NA
Copollutant models
with: NA
Lim et al. (2018)
U.S.
PM2.5: 2002-2010
Follow-up:
1995-2011
NIH-AARP
Diet and
Health Study
n = 549,735
Diabetes
deaths =
3,597
50-71 yr of
age
Diabetes
Long-term exposure to PM2.swas Age, sex, region, Mean (SD): 11.0 Correlation (R2)
estimated usina annual averaae race/ethnicitv. (2.7) Mn.-na
estimated using annual average
concentrations for the Years 2002 to
2010 from a spatiotemporal
prediction model as described in Kim
et al. (2017). Average PM2.5
concentrations were assigned based
on residential census tract centroids.
race/ethnicity,
education level, marital
status, BMI, alcohol
consumption, smoking
status, diet, median
census tract household
income, percent of
census tract population
less than high school
education
(2.7)
Range: 2.8-21.2
NO2: 0.6
Os: 0.01
Copollutant models
with: NA
September 2021
A-32
DRAFT: Do Not Cite or Quote
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
mortality.
PM2.5
Study Mortality Confounders in Concentrations Copollutant
Study Population Outcomes Exposure Assessment Statistical Model (|jg/m3) Examination
Ward-Caviness et
al. (2020)
North Carolina,
U.S.
PM2.5: 2003-2016
Follow-up: July
2004-December
2016
North Carolina
hospital-based
cohort of
individuals
diagnosed
with heart
failure; using
electronic
health records
n = 23,302
contributed to
analyses (out
of total of 35,
084 HF
patients)
Deaths =
4,496
Mean age =
66.9 yr
(SD = 15.2
years;
IQR = 22.2 yr)
All cause
Exposures are estimated using
nearest PM2.5 monitor and using
Harvard's 1 km * 1 km modeled
PM2.5 surface.
Monitors: Daily PM2.5 values were
obtained from the U.S. EPA National
Ambient Air Quality Standards
(NAAQS) ground-based monitoring
network (July 1, 2003-December
31, 2016). Annual average PM2.5
concentration assigned to nearest
monitor of geocoded address for the
365 days preceding the time of initial
heart failure, based on the patient's
electronic health records. Sensitivity
analyses were restricted to
participants <30 km from a monitor.
Modeled: An ensemble-based
prediction model was used to
estimate daily PM2.5 concentrations
for a 1-km2 grid network across the
contiguous U.S.; (model validation:
r2 = 0.89 for Middle Atlantic region
of U.S. for 2000-2015) [discussed in
Pi et al. (2019)1. Daily PM2.5
estimates for North Carolina were
extracted for the years 2003-2016
and annual averages were
calculated.
Age, sex, race,
distance to nearest
monitor,
neighborhood-level
socioeconomic
variables (household
below federal poverty
line, median home
value, median
household income,
urbanicity, households
receiving public
assistance)
Monitors:
Mean: 10.2
SD: 2.11
IQR: 3.36
Modeled:
Mean: 10.3
SD: 1.70
IQR: 2.45
Correlation (r): NA
Copollutant models
with: NA
September 2021
A-33
DRAFT: Do Not Cite or Quote
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term exposure to PM2.5 and
mortality.
Study
Study
Population
Mortality
Outcomes
Exposure Assessment
Confounders in
Statistical Model
PM2.5
Concentrations
(Hg/m3)
Copollutant
Examination
Malik etal. (2019)
U.S. -31
hospitals
PM2.5: 2002-2007
Follow-up:
2003-2013
TRIUMPH and
PREMIER
cohorts
n = 5,640
deaths = NA
Mean age =
59.9
(SD = 12.7)
All cause
Daily average PM2.5 concentration
estimated at census tract centroid of
patient using U.S. EPA's
downscaled CMAQ, which is a
Bayesian space-time fusion model
that bias corrects modeled output
data with monitored data as detailed
in (Berrocal et al.. 2012). Annual
average PM2.5 concentrations
estimated for year prior to
myocardial infarction.
Age, sex, race,
smoking status, date of
enrollment, SES
(patients' education,
insurance status,
history of avoiding care
because of costs,
end-of-the-month
financial resources)
Mean: 11.96
SD: 2.11
Range: 4.3-20.5
Correlation (r): O3:
-0.02
Copollutant models
with: O3
Bennett et al.
(2019)
U.S. (1,339
county units)
PM2.5: 1999-2015
Follow-up: NA
Vital
Registration
data from
NCHS
Deaths = 41.9
million
All cause
Annual average PM2.5 estimated
from integrated geographic model
described in Kim et al. (2020). PM2.5
modeled through application of
universal kriging approach to
monitoring data and geographic
variables, and satellite-derived
estimates of PM2.5. PM2.5
concentrations predicted to 2010
census block centroids in contiguous
U.S. and aggregated by
population-weighting to county level.
Per capita income;
percentage of
population
Black/African
American, graduated
from high school, live
in urban areas,
unemployed; proxy for
cumulative smoking,
mean temperature and
RH
1999
Median: 12.7
75h: 14.6
99th: 19.7
2015
Median: 7.7
75th: 8.5
99th: 10.1
Correlation (r):
NO2: 0.50
Os: 0.55
Copollutant models
with: NA
AOD = aerosol optical depth; BMI = body mass index; CanCHEC = Canadian Census Health and Environment Cohort; CAN-Marg = Canadian Marginalization Index;
CCHEC = Canadian Census Health and Environment Cohort; CMA = census metropolitan area size; CMAQ = Community Multi-scale Air Quality model; GEOS-Chem = Goddard
Earth Observing System-Chem; GWR = geographically weighted regression; mCCHS = Canadian Community Health Survey—Mortality cohort; NCHS = National Center for Health
Statistics; NHIS = National Health Interview Survey; ONPHEC = Ontario Population Health and Environment Cohort; Ox = redox weighted average of N02 and 03;
PREMIER = Prospective Registry Evaluating Myocardial Infarction: Events and Recovery; RH = relative humidity; SD = Standard Deviation; SES = socioeconomic status;
TRIUMPH = Translational Research Investigating Underlying Disparities in Acute Myocardial Infarction.
September 2021
A-34
DRAFT: Do Not Cite or Quote
-------
Table A-8 Study-specific details for epidemiologic studies using accountability analyses or causal modeling
methods to examine long-term exposure to PM2.5 and mortality.
Study/Location
PM2.5
Population
Exposure
Concentrations
Years
(Cohort)
Outcome
Assessment
(Hg/m3)
Confounders
Copollutant Examination
Peterson et al.
National
Cardiovascular
Estimated annual
Weighted annual
County: median household
Correlation (r): NA
(2020)
Center for
mortality
average PM2.s-total
trend mass
income, percent of
Copollutant models with: NA
Health
and component
concentration
non-White population, and
U.S.
Statistics
concentrations
(standard error):
population; age-standardized
n = 2,132
(sulfates, nitrates,
0.134 (0.001)
annual COPD mortality rates
counties
elemental carbon, and
(to account for the
1990-2010
organic carbon)
between 1990 and
2010 from CMAQ
cumulative burden of
smoking and annual smoking
rates)
Wei et al. (2020)
Medicare
All-cause
Used predicted daily
Mean (SD): 9.0
Temperature—air and dew
Correlation (r): NA
mortality
ambient PM2.5, ozone,
(1.9)
point
Copollutant models with: O3,
Massachusetts
n = 1,503,572
and nitrogen dioxide
levels in each
Range: 3.3-16.4
Individual: sex;
NO2
race/ethnicity; age; Medicaid
1-km x 1-km grid cell
eligibility
2000-2012
across the contiguous
U.S. using
well-validated
ZIP code level: annual
median household income;
ensemble models
median value of
owner-occupied housing
units; percentage of
population living in poverty;
percentage of the population
with less than a high school
education; population
density; home ownership
rate
County level: annual
percentage of ever smokers;
percentage of obese people
September 2021
A-35
DRAFT: Do Not Cite or Quote
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using causal modeling methods to
examine long-term exposure to PM2.5 and mortality.
PM2.5
Study/Location Population Exposure Concentrations
Years (Cohort) Outcome Assessment (|jg/m3) Confounders Copollutant Examination
Wei et al.
(2021b)
Massachusetts
2000-2012
Medicare All-cause
n = 1,503,572 mortality
Daily concentrations
of ambient PM2.5,
ozone, and nitrogen
dioxide levels at
1 km x 1 km grid cells
were predicted using
geographically
weighted regressions
that ensembled
predictions from
ensemble methods
Mean (SD): 9.0
(1.9)
Range: 3.3-16.4
Air surface temperature, dew Correlation (r): NA
point temperature, and
relative humidity
Individual: sex, race, age,
Medicaid eligibility
ZIP code level: median
household income, median
house value, percent of
owner-occupied homes,
percent of population living in
poverty, percent of
population below high school
education, population
density, percent of Blacks,
and percent of Hispanics,
percent of persons over age
65 with an annual
hemoglobin A1c test, an
annual low-density
lipoprotein test, and an
annual eye exam in each
hospital catchment area
County level: percent of ever
smokers, lung cancer rate,
and average BMI
Copollutant models with: O3,
NO2
September 2021
A-36
DRAFT: Do Not Cite or Quote
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using causal modeling methods to
examine long-term exposure to PM2.5 and mortality.
PM2.5
Study/Location Population Exposure Concentrations
Years (Cohort) Outcome Assessment (|jg/m3) Confounders Copollutant Examination
Wu et al.
(2020a)
U.S.
2000-2016
Medicare
68,503,979
Mortality
Estimated daily PM2.5
levels at a high
spatiotemporal
resolution using a
1-km2 grid network
across the contiguous
U.S. and a
well-validated
ensemble-based
prediction model
Mean (SD): 9.8
(3.2)
Meteorologic variables: ZIP
code level summer and
winter averages of maximum
daily temperatures; relative
humidity
Individual: age,
race/ethnicity, sex, Medicare
eligibility
ZIP code level: proportion of
Hispanic residents;
proportion of Black resident;
median household income;
median home value;
proportion of residents in
poverty; proportion of
residents with a high school
diploma; population density;
proportion of residents that
own their house
County level: average BMI
and smoking rate
Indicator variables: four
census geographic regions
(Northeast, South, Midwest,
and West); calendar years
(2000-2016)
Correlation (r): NA
Copollutant models with: NA
September 2021
A-37
DRAFT: Do Not Cite or Quote
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using causal modeling methods to
examine long-term exposure to PM2.5 and mortality.
Study/Location population
Years (Cohort)
Outcome
Exposure
Assessment
PM2.5
Concentrations
(Hg/m3)
Confounders
Copollutant Examination
Wu et al. (2019) Medicare
New England n = 3,300,000
(Vermont, New
Hampshire,
Connecticut,
Massachusetts,
Rhode Island,
and Maine)
2000-2012
All-cause PM2.5 exposures are
mortality determined at each
1 km x 1 km grid cell
using a
spatiotemporal
prediction model
which uses multiple
different sources as
input
Mean: 9.35 Individual: Age, sex, race,
and Medicare eligibility
Area-level: BMI; percent ever
smoke; percent Hispanic
population; percent Black
population; median
household income; median
value of housing; percent
below poverty level; percent
below high school education;
percent of owner-occupied
housing; population density;
percent with LDL-C test;
percent with 1 ambulatory
visit; percent with
hemoglobin A1c test;
temperature; relative
humidity
Correlation (r): NA
Copollutant models with: O3
Corriqan et al.
(2018)
U.S.
2000-2010
National
Center for
Health
Statistics
n = 619
counties (total)
n = 486
counties
(attainment)
n = 133
counties (non-
attainment)
Cardiovascular
mortality
U.S. EPA's Air Quality
System (AQS)
monitoring
sites—calculated
annual average PM2.5
concentrations at
each monitoring site
for each year between
2000 and 2010; then
averaged the annual
means across the
monitors located in
the same county to
calculate annual
averages for counties
Before (2000-2004)
Mean (IQR): 12.0
(3.9)
After (2005-2010)
Mean (IQR): 10.8
(3.3)
Temperature
County: total income,
percent with at least a high
school diploma (of
population 25 yr and older),
percent Hispanic (of total
population), and percent
Black (of total population)
Correlation (r): NA
Copollutant models with: NA
September 2021
A-38
DRAFT: Do Not Cite or Quote
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using causal modeling methods to
examine long-term exposure to PM2.5 and mortality.
Study/Location population
Years (Cohort)
Outcome
Exposure
Assessment
PM2.5
Concentrations
(Hg/m3)
Confounders
Copollutant Examination
Yitshak-Sade et Medicare Mortality
al. (2019b)
Maine, New
Hampshire,
Vermont,
Massachusetts,
Rhode Island,
Connecticut,
New York, New
Jersey,
Delaware,
Pennsylvania,
Maryland,
Washington, DC,
Virginia, and
West Virginia
15,401,064
Highly spatially
resolved PM2.5 data
(1 x 1 km spatial
resolution) from a
hybrid satellite-based
model incorporating
daily satellite remote
sensing Aerosol Optic
Depth data and
classic land-use
regression
methodologies
Range of Mean
Annual PM2.5
Concentrations:
6.5-14.5
Temperature
Individual: age,
race/ethnicity, sex, Medicare
eligibility
Correlation (r): NA
Copollutant models with: NA
2000-2013
September 2021
A-39
DRAFT: Do Not Cite or Quote
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using causal modeling methods to
examine long-term exposure to PM2.5 and mortality.
PM2.5
Study/Location Population Exposure Concentrations
Years (Cohort) Outcome Assessment (|jg/m3) Confounders Copollutant Examination
Henneman et al. Medicare
(2019)
U.S.
2005-2012
All-cause Modeled national
mortality PM2.5 concentrations
estimated by
combining
GEOS-Chem
chemical transport
model-simulated
PM2.5,
satellite-retrieved
aerosol optical depth,
and observed PM2.5
concentrations.
Estimated monthly 0.1
longitude * 0.1
latitude PM2.5
concentrations were
averaged to annual
concentrations and
spatially overlayed on
U.S. ZIP codes
Coal exposure:
HYSLPLIT air parcel
trajectory and
dispersion model
Mean (IQR):
2005: 10.0 (4.5)
2012: 7.2 (2.6)
Difference between
2012 and 2005:
-3.2 (2.4)
Median age; median
household income;
per-capita income; sex; race;
fraction of population by
county that smoked in 2000;
temperature; specific
humidity
Correlation (r): NA
Copollutant models with: NA
Sanders et al.
(2020)
U.S.
2000-2013
Medicare
n = 137
counties (non-
attainment)
n = 467
counites
(attainment)
Mortality
Daily 24-h average
PM2.5 was calculated
for each county as a
simple average of all
monitors within a
county (if multiple
monitors exist)
Mean (SD): 10.84
(3.06)
Mean for
Non-attainment
<2006: 15.29
Attainment <2006:
10.99
Mean for
Non-attainment
>2006: 11.96
Attainment >2006:
9.33
Temperature (daily minimum
and maximum)
Total precipitation
Income per capita, share of
population employed,
migration
Correlation (r): NA
Copollutant models with: NA
September 2021
A-40
DRAFT: Do Not Cite or Quote
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using causal modeling methods to
examine long-term exposure to PM2.5 and mortality.
Study/Location population
Years (Cohort)
Outcome
Exposure
Assessment
PM2.5
Concentrations
(Hg/m3)
Confounders
Copollutant Examination
Fan and Wang
(2020)
U.S.
1999-2013
Medicare
n = 770
county-month
observation for
treated
n = 7504
county-month
observation for
controls
Mortality
Ambient PM2.5
monitoring data from
the U.S. EPA Air
Quality System (AQS)
and power plants
were selected from a
list of coal-fired power
plants from Clean Air
Watch
Mean (SD): 12.04
(3.78)
Temperature, dew point,
barometric pressure
County: median household
income, poverty rate,
percentage of non-Hispanic
Whites in the population, and
percentage of population
with college degree
Correlation (r): NA
Copollutant models with: NA
September 2021
A-41
DRAFT: Do Not Cite or Quote
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using causal modeling methods to
examine long-term exposure to PM2.5 and mortality.
PM2.5
Study/Location Population Exposure Concentrations
Years (Cohort) Outcome Assessment (|jg/m3) Confounders Copollutant Examination
Schwartz et al.
(2021)
U.S.
2000-2016
Medicare
n =
623,036,820
person-years
of follow-up
Mortality
Used a validated
prediction model
calibrated to
measurements at
almost 2000 U.S. EPA
Air Quality System
monitoring stations
using an ensemble of
machine learners that
provided daily
estimates for a 1 km
grid of the contiguous
U.S.
Mean (SD): 10.3
(3.1)
Median: 9.8
IQR: 7.9, 12.0
Individual: age, sex, ZIP
code, Medicaid eligibility
ZIP code: percent of people
>65 living in poverty, median
household income, median
house value, percent of
owner occupied homes,
percent Black, percent
Hispanic, population density,
and education, percentage of
Medicare participants who
had a hemoglobin A1 c test, a
low-density lipoprotein
cholesterol (LDLC) test, a
mammogram, an eye exam,
and a visit to an annual
checkup for each year in
each hospital catchment
area, distance from each ZIP
code centroid to the nearest
hospital, hospitalization rate
for lung cancer (proxy for
long-term smoking)
County: percentage of
people who ever smoked
and BMI
Meteorologic: average
temperature in the warm
months (April-September)
and in the cold months
(October-March)
Correlation (r): NA
Copollutant models with: NA
September 2021
A-42
DRAFT: Do Not Cite or Quote
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using causal modeling methods to
examine long-term exposure to PM2.5 and mortality.
PM2.5
Study/Location Population Exposure Concentrations
Years (Cohort) Outcome Assessment (|jg/m3) Confounders Copollutant Examination
Schwartz et al.
(2018b)
Northeastern
and Mid-Atlantic
States (Maine,
New Hampshire,
Vermont,
Massachusetts,
Rhode Island,
Connecticut,
New York, New
Jersey,
Delaware,
Pennsylvania,
Maryland,
Washington, DC,
Virginia, and
West Virginia)
2000-2013
Medicare
n =
129,341,959
person-years
n = 6,334,905
deaths
Life expectancy
Estimate annual
average
concentrations of
PM2.5 at each ZIP
code, which uses a
hybrid model that
integrates land use,
meteorological, and
satellite remote
sensing data
Mean: 10.3
Median: 10.4
25th Percentile: 9.2
75th Percentile:
11.4
Individual: age, sex, race,
ZIP code of residence for
that year, Medicaid eligibility
ZIP code: percentage of the
population that was Black,
Hispanic, >65 yr of age living
in poverty, living in
owner-occupied housing,
and with less than a high
school education as well as
median household income,
median value of
owner-occupied housing,
and population density;
hospitalization rate for lung
cancer (to capture long-term
smoking); percentage of
Medicare participants who
had a hemoglobin A1 c test, a
low-density lipoprotein
cholesterol (LDL-C) test, a
mammogram, and a visit to a
primary care physician for
each year in each hospital
catchment area
County: percentage of
people who ever smoked
and BMI scores
Correlation (r): NA
Copollutant models with: NA
September 2021
A-43
DRAFT: Do Not Cite or Quote
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using causal modeling methods to
examine long-term exposure to PM2.5 and mortality.
Study/Location population
Years (Cohort)
Outcome
Exposure
Assessment
PM2.5
Concentrations
(Hg/m3)
Confounders
Copollutant Examination
Awad et al.
(2019)
U.S.
2000-2012
Medicare
n =
12,095,504
Mortality
Mean annual
exposure to PM2.5 for
each enrollee at
his/her residential ZIP
code for each year
between 2000 and
2012 was estimated
using a neural
network-based hybrid
prediction mode. Daily
predictions were
generated and then
averaged over the
calendar year for the 4
grids closest to the
centroid of the ZIP
code of residence
Mean in the year of
move: 11.88 for
Whites and 13.02
for Blacks
Mean in second
year after move:
11.15 for Whites
and 12.12 for
Blacks
Individual: age, sex, race,
ZIP code of residence,
Medicaid eligibility;
hospitalizations for
Alzheimer's disease, acute
myocardial infarction,
diabetes Mellitus, heart
failure, Parkinson's disease,
pneumonia, other respiratory
diseases, ischemic stroke,
unstable angina, vascular
dementia, chronic
obstructive pulmonary
disease, and lung cancer
ZIP code: median household
income, population density,
percentage Black,
percentage of
owner-occupied housing
units, median value of
owner-occupied housing,
percentage above age 65
living below the poverty
level, and percentage above
age of 65 with less than high
school education
Correlation (r): NA
Copollutant models with: NA
Hiqbee et al.
(2020)
U.S.
1986-2015
National
Center for
Health
Statistics
n = 635,539
All-cause
mortality
Cardiopulmonary
mortality
Annual pollution
exposures were
estimated for each
census block using
national regulatory
monitoring data from
1999-2015 within a
universal kriging
model employing land
use regression
methods and
hundreds of variables
Mean (SD): 10.7
(2.4)
Range: 2.5-19.2
Individual: age, sex,
race/ethnicity, educational
attainment, marital status,
income level, urban-rural
designation, census tract,
interview date, mortality
status, smoking status, BMI,
and date of death
Correlation (r): NA
Copollutant models with: NA
September 2021
A-44
DRAFT: Do Not Cite or Quote
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using causal modeling methods to
examine long-term exposure to PM2.5 and mortality.
PM2.5
Study/Location Population Exposure Concentrations
Years (Cohort) Outcome Assessment (|jg/m3) Confounders Copollutant Examination
Wei et al.
(2021a)
U.S.
2000-2016
Medicare
74,537,533
All-cause The daily
mortality concentrations of
ambient PM2.5, O3,
and NO2 at
1 km x 1 km grid cells
across the contiguous
U.S. were predicted
and validated using
hybrid models that
ensembled predictions
from random forest,
gradient boosting, and
neural network
Mean (SD): 9.85
(3.17)
Meteorologic: Air
temperature, humidity,
average temperature in
warm (April-September) and
cold seasons
(January-March, plus
October-December)
Individual: sex, race, age,
Medicaid eligibility
ZIP code: percentage of
Blacks, percentage of
Hispanics, median
household income, median
value of owner occupied
housing, percentage of
Americans aged 65 and
older living below the poverty
threshold, percentage of
Americans with less than
high school education,
percentage of owner
occupied housing units, and
population density;
percentage of Medicare
participants who had a
hemoglobin A1c test, a
low-density lipoprotein
cholesterol (LDLC) test, a
mammogram, and an eye
exam to a primary care
physician for each year in
each hospital catchment
area
County: BMI and percentage
of ever smokers
Correlation (r): NA
Copollutant models with: O3
and NO2
September 2021
A-45
DRAFT: Do Not Cite or Quote
-------
Table A-9 Study-specific details for epidemiologic studies of short-term PM2.5 exposure and COVID-19.
Study/Location Exposure Assessment and Long-Term Mean
Years Outcome and Upper Percentile Concentrations (|jg/m3) Covariates in Statistical Model Copollutant Examination
Adhikari and Yin Daily confirmed Daily average PM2.5 collected and averaged from Lagged outcome and day trend Correlation (r): O3 = -0.82
(2020) COVID-19 cases two stationary monitors. Copollutant models with:
Queens County, and deaths Mean (SD): 4 73 (2.39) NA
NY Median: 4.1
March1. Range: 0.65 to 11.15
2020—April 20,
2020
September 2021
A-46
DRAFT: Do Not Cite or Quote
-------
Table A-10 Study-specific details for epidemiologic studies of long-term PM2.5 exposure and COVID-19.
Study/Location
Years Outcome Exposure Assessment Covariates in Statistical Model Copollutant Examination
Chakrabartv et RO
al. (2021)
U.S.
70-79, % of population >80, number of
people tested, hospital beds, ICU beds,
liquid asset poverty rate, total healthcare
and social services workers, total essential
workers, fraction of total healthcare and
social services workers, fraction of total
essential workers, average household size,
average family size, number of households,
number of family, households with elderly
resident, renter-occupied housing units,
lapse in states issuance of stay-at-home
order, residents in 2 or more unit structures,
median income for single earner household,
median income for working age individuals,
% White, % African-American, % Asian, %
Native American Indian, % Pacific Islander,
% Other race, ozone, NO2, relative humidity,
SO2.
PM2.5 concentrations were aggregated at
the state-level using monthly estimates
based on stationary monitors, model inputs,
and satellite observations between the
years 2012-2017.
population, population density, % of
population <9, % of population 10-19, % of
population 20-29 % of population 30-39, %
of population 40-49, % of population 50-59,
% of population 60-69, % of population
Correlation (r): NA
Copollutant models with: NA
Liang et al.
(2020)
3,122 U.S.
Counties January
22, 2020 to July
17, 2020
U.S. COVID-19 Daily ambient PM2.5 concentrations
deaths estimated using an ensemble machine
n = 138 552 learning model at the 1 km * 1 km grid
level for 2010-2016. Annual mean
concentrations were aggregated to county
level.
5th percentile: 3.8
95th percentile: 10.4
County-level number of cases per 1,000
people, social deprivation index, population
density, % of residents >60, % males, %
race and ethnicity, body mass index,
smoking rate, number of regular hospital
beds per 1,000 people, number of intensive
unit beds per 1,000 people, number of
medical doctors per 1,000 people, average
mobility (March and July 17, 2020), average
temperature and humidity, state-level
COVID-19 test positive rate as of July 17,
2020, and spatial smoother with 5 degrees
of freedom for both latitude and longitude.
Correlation (r): NA
Copollutant models with: NO2,
O3
September 2021
A-47
DRAFT: Do Not Cite or Quote
-------
Table A-10 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure and
COVID-19.
Study/Location
Years
Outcome
Exposure Assessment
Covariates in Statistical Model
Copollutant Examination
Mendv et al.
COVID-19
(2021)
hospitalizations
University of
n =1,128 COVID
Cincinnati
cases
Hospital System
n = 310
March 13,
hospitalizations
2020-July 5,
2020
Stieb et al.
COVID-19
(2020)
infections
111 Canadian
n = 73,390
health regions
up to May 13,
2020
10-yr average and 10-yr maximum
concentrations of PM2.5 were aggregated
for each ZIP code between 2008-2017
within the study area.
Average Mean ± SD: 11.34 ± 0.70
Maximal Mean ± SD: 13.83 ± 1.03
Age, race, sex, median household income,
smoking status, obesity, diabetes, asthma,
COPD, cardiovascular disease, chronic
kidney disease, neoplasm/history of
neoplasm
Correlation (r): NA
Copollutant models with: NA
Annual PM2.5 concentrations estimated
using satellite imagery, chemical transport
models, and ground observations for years
2000-2016 on a 0.01° * 0.01° grid.
Concentrations were aggregated to 2018
health region boundaries.
Mean (SD): 6.1 (2.1)
Temperature (minimum, maximum),
population density, % >65, % with income
¦slowest income cutoff, % Black, % asthma,
% COPD, % hypertension, % diabetes, %
physically active, % overweight, % obese, %
smokers, days since first case, days since
peak incidence, greenspace
Correlation (r): NA
Copollutant models with: NA
Wu et al. (2020b)
3,089 U.S.
Counties
Up to June 18,
2020
COVID-19 deaths
n = 116,747
Daily PM2.5 concentrations estimated using
atmospheric chemistry and machine
learning models for the years 2000-2016
on 0.01° x 0.01° grid. Daily concentrations
were aggregated to county level and
averaged.
Mean (SD): 8.4 (2.5)
Days since first COVID-19 case reported,
population density, % of population >65, %
of population 45-64, % of population 15-44,
% of population in poverty, median
household income, % Black residents, %
Hispanic residents, % of adult population
with
-------
Table A-11 Study-specific details for epidemiologic studies examining socioeconomic status and PM2.5
exposure.
Study/Location
Years
Exposure Assessment
Mean Concentration (|jg/m3
Copollutant Examination
Han et al. (2017)
Houston, TX
June 2013-November
2013
PM2.5 concentrations collected from two stationary Mean (SD):
monitors placed in a single Low SES community, |_ow 3^3. 11 3 (2 90)
and a High SES community. ,
a 3 High SES: 9.6 (2.93)
Correlation (r): NA
Copollutant models with: NA
Lee (2019)
California
2016
PM2.5 concentrations estimated from both
stationary monitors and satellite
Statewide
Mean (SD): 8.09 (3.25)
Poverty: High: 9.7, Low: 8.7
Education: High: 9.9, Low: 8.5
Correlation (r): NA
Copollutant models with: NA
Lee and Park (2020)
California
2012-2014
PM2.5 concentrations estimated from stationary
monitors
Mean (SD):
Overall: 9.4 (8.0)
High Vulnerability: 10.8 (8.0)
Low Vulnerability: 6.8 (5.5)
Correlation (r): NA
Copollutant models with: NA
Richmond-Bryant et al.
(2020)
U.S.
2008, 2011, 2014
PM2.5 emissions from fossil-fuel Electricity
Generating Units obtained from the National
Emissions Inventory
NR
Correlation (r): NA
Copollutant models with: NA
September 2021
A-49
DRAFT: Do Not Cite or Quote
-------
Table A-11 (Continued): Study-specific details for epidemiologic studies examining socioeconomic status and
PM2.5 exposure.
Study/Location
Years
Exposure Assessment
Mean Concentration (|jg/m3
Copollutant Examination
Rosofskv et al. (2018)
Massachusetts
2003-2010
PM2.5 concentrations estimated at 1 km * 1
grid using a modeling approach which
incorporated satellite, land use, and
meteorological data aggregated to yearly
averages.
km Population-weighted annual average
PM2.5 range:
Education:
Masters: 2003:11.2, 2010: 8,0
Bachelors: 2003: 11.2, 2010: 8.0
High School-Post Secondary:
2003: 11.1, 2010: 7.9
$75,000: 2003:11.1, 2010:7.9
cn
0
1
<0
cn
0
0
0
2003:
11.1,
2010:
(JO
O
CO
cn
1
<0
cn
0
0
0
0
2003:
11.2,
2010:
B.O
w
K)
0
1
<0
CO
cn
0
0
0
2003:
11.2,
2010:
B.1
<$20,000: 2003: 11.4, 2010: 8.2
Correlation (r): NA
Copollutant models with: NA
Tanzer et al. (2019)
Pittsburgh, PA
April 2017-May 2018
PM2.5 measured using Met-One Neighborhood PM
Monitors (NPMs) and small subset measured
using PurpleAir PA-II, as part of a Rea-time
Affordable Multi-Pollutant (RAMP) package
Annual average range: 7.5 to 25.i
EJ Communities: 10.6 (1.0)
Non-EJ Communities: 10.3 (1.5)
Correlation (r): 0.32 (0.16-0.56) SO2
Copollutant models with: NA
Weaver et al. (2019)
2001-2010
Duke University Medical
Center Wake, Durham,
and Orange Counties in
North Carolina
Daily average PM2.5 concentrations were
estimated using a hybrid model on a 1 * 1 km grid
Mean (SD): 12.7 (1.1)
Cluster 1
12.9 (1.1)
Cluster 2
13.2 (1.0)
Cluster 3
12.8 (1.1)
Cluster 4
12.8 (1.1)
Cluster 5
12.2 (1.2)
Cluster 6
11.9 (1.0)
Correlation (r): NA
Copollutant models with: NA
September 2021
A-50
DRAFT: Do Not Cite or Quote
-------
Table A-12 Study-specific details for epidemiologic studies examining short-term PM2.5 exposure and health risk
disparity by socioeconomic status.
Study/
Location
Years
Population
(Cohort)
Outcome
Exposure Assessment
and Mean
Concentration (|jg/m3)
Select Results
Covariates in Statistical
Model
Copollutant
Examination
Yitshak-Sade et
al. (2019a)
Massachusetts
2001-2011
Case-crossover
Massa- Cardio-
chusetts vascular
Department mortality
of Public
Health
n = 179,986
Daily average PM2.5
estimated from a model
incorporating aerosol
optical depth and
monitored PM2.5 at a 1 *
1 km grid
Mean: 10.2
Maximum: 17.4
Change in CV Mortality
High % High School Educated
Green: 2.64% (0.46, 4.68)
Less Green: 1.42% (0.72, 3.62)
High % without high school diploma
Green: 2.64% [0.60, 4.72]
Less Green: 3.31% [1.26, 5.41]
High % White
Green: 2.80% (0.62, 5.02)
Less Green: 1.14% (-1.00, 3.33)
Low % White
Green: 2.47% (0.43, 4.56)
Less Green: 3.55% (1.49, 5.65)
Temperature and day of
the week
Correlation (r): NA
Copollutant models
with: NA
Son et al. (2020)
North Carolina
2002-2013
Case-crossover
Death
Records
n = 775,338
Total Mortality Daily average PM2.5
concentrations estimated
from the CMAQ
downscaler model on a
12x12 grid
Mean (SD): 11.4 (5.7)
Maximum: 70.8
OR (95% CI):
Income
<$41,500: 1.01 (1.01, 1.01)
>$41,500: 1.01 (1.00, 1.01)
Education
<12 yr: 1.01 (1.01, 1.02)
High School: 1.01 (1.00, 1.01)
1-4 yr of college: 1.01 (1.00, 1.01)
>5 yr of college: 1.01 (0.99, 1.02)
Unknown: 1.00 (0.98, 1.03)
Residential greenness,
proximity to water bodies,
median household
income, and classification
of urbanicity
Correlation (r): O3:
0.48
Copollutant models
with: NA
September 2021
A-51
DRAFT: Do Not Cite or Quote
-------
Table A-13 Study-specific details for epidemiologic studies examining long-term PM2.5 exposure and health risk
disparity by socioeconomic status.
Study/
Location
Years
Population
(Cohort)
Outcome
Exposure Assessment
and Long-Term Mean
and Upper Percentile
Concentrations (|jg/m3)
Select Results
Covariates in
Statistical Model
Copollutant
Examination
Bai et al.
(2019)
Ontario,
Canada
Canadian
Ontario
Population
and Health
Cohort
Congestive
Heart Failure
(n = 5,062,146)
Acute
Myocardial
Infarction
(n = 5,141,172)
Annual average PM2.5
concentrations estimated
using the GEOS-Chem
CTM model and a
geographically weighted
regression model
Mean (SD): 9.6 (2.8)
Max: 20.0
Income
HR (95% CI)
CHF
Lowest: 1.12 (1.10, 1.13)
Lower: 1.09 (1.06, 1.10)
Middle: 1.03 (1.01, 1.04)
Upper: 1.04 (1.03, 1.06)
Uppermost: 1.01 (1.00, 1.04)
AMI
Lowest: 1.12 (1.09, 1.15)
Lower: 1.09 (1.06, 1.12)
Middle: 1.04 (1.03, 1.07)
Upper: 1.03 (1.01, 1.06)
Uppermost: 1.03 (1.00, 1.06)
Age, sex, area level risk
factors including SES,
and geographic
indicator variables that
distinguished
participants based on
whether their residence
in the north or south of
Ontario and whether it
was urban or rural
Correlation (r): NA
Copollutant models
with: NA
Bennett et al.
(2019)
U.S.
2015
National
Center for
Health
Statistics
n = 41.9
million
Life expectancy
loss
Annual average PM2.5
concentrations estimated
using an integrated
geographic regression
model
Median: 7.7
99th percentile: 10.1
PM2.5 associated with lower life
expectancy among counties with
lower income, higher percent
poverty, and those with a low %
who graduated from high school.
See Figure
Income, % in poverty,
Black race, >high
school, urbanization,
unemployment,
cumulative smoking,
mean temperature,
relative humidity,
county-specific random
intercepts
Correlation (r): NA
Copollutant models
with: NA
September 2021
A-52
DRAFT: Do Not Cite or Quote
-------
Table A-13 (Continued): Study-specific details for epidemiologic studies examining long-term PM2.5 exposure and
health risk disparity by socioeconomic status.
Study/
Location
Years
Population
(Cohort)
Outcome
Exposure Assessment
and Long-Term Mean
and Upper Percentile
Concentrations (|jg/m3)
Select Results
Covariates in
Statistical Model
Copollutant
Examination
Bevan et al.
(2021)
U.S.
2000-2016
National
Center for
Health and
Statistics
n =
5,769,315
Cardiovascular
Mortality
Annual zip-level average
PM2.5 concentrations
estimated using aerosol
optical depth calibrated to
ground-based
observations
Mean (SD):
6.51 (1.54)
Social Deprivation Index (SDI)
(1-100)
No. deaths/100,000 (95% CI)
SDI 1-25: 39.1
(32.1, 46.1)
SDI 26-50: 48.2
(38.5, 57.9)
SDI 51-75: 71.0
(57.6, 84.4)
SDI 76-100: 52.0
(30.9, 73.1)
Smoking, diabetes
obesity, physical
inactivity, and
urbanization
Correlation (r): NA
Copollutant models
with: NA
Crouse et al.
(2019)
Canada
2001-2011
Canadian
Census
Health
Environment
Cohort
Non-accidental
Mortality
Cardio-
metabolic
Mortality
Cardiovascular
Mortality
Annual average PM2.5
concentrations estimated
from satellite derived
annual estimates on a
1 x 1 km grid
Mean (SD): 8.4 (2.7)
95th percentile: 13.3
For all mortality outcomes, the
strongest associations were among
the group with the lowest amount of
deprivation and the lowest amount
of greenspace. See Figure.
Aboriginal identity,
visible minority status,
marital status, highest
level of education,
employment status, and
household income
adequacy quintiles.
Correlation (r): NA
Copollutant models
with: O3
Jorqenson et
al. (2020)
U.S.
2000-2014
U.S. Mortality
Database
Life-expectancy
at birth
PM2.5 concentrations
estimated from stationary
monitors, averaged at the
state level
Mean: 10.55
Maximum: 19.02
PM2.5 more detrimental in states
with high percent of income
inequality. See Figure.
Income share of top
10%, % Black, total
population, median
household income,
median age, % college
degree or higher
Correlation (r): NA
Copollutant models
with: NA
Schulz et al.
(2018)
Detroit
Metropolitan
Area
2013
n = 171,000 Cardio-
pulmonary
Mortality
Annual average PM2.5 Both PM2.5 and social vulnerability Age, gender,
concentrations estimated
from downscaler and
CMAQ by census tract
Mean: 9.6
Maximum: 10
were independently related to CV
mortality in the same model.
race/ethnicity,
educational attainment,
death attributable to
smoking, and marital
status.
Correlation (r): NA
Copollutant models
with: NA
September 2021
A-53
DRAFT: Do Not Cite or Quote
-------
Table A-13 (Continued): Study-specific details for epidemiologic studies examining long-term PM2.5 exposure and
health risk disparity by socioeconomic status.
Study/
Location
Years
Population
(Cohort)
Outcome
Exposure Assessment
and Long-Term Mean
and Upper Percentile
Concentrations (|jg/m3)
Select Results
Covariates in
Statistical Model
Copollutant
Examination
Shin et al.
(2019)
Ontario,
Canada
Canadian
Ontario
Population
and Health
Cohort
n =
5,071,956
Atrial Fibrillation
Stroke
Annual average PM2.5
concentrations estimated
using AOD and PM2.5
simulated by the
GEOS-Chem CTM at a
1 x 1 km grid
Mean (SD):
9.8 (2.9)
IQR: 4.0
Maximum: 20
Income
HR (95% CI)
Atrial Fibrillation
Lowest: 1.06 (1.04, 1.08)
Lower: 1.05 (1.03, 1.07)
Middle: 1.02 (1.00, 1.04)
Upper: 1.02 (1.01, 1.04)
Uppermost: 0.99 (0.97, 1.02)
Stroke
Lowest: 1.08 (1.04, 1.13)
Lower: 1.08 (1.05, 1.11)
Middle: 1.04 (1.00, 1.08)
Upper: 1.01 (0.99, 1.04)
Uppermost: 1.01 (0.99, 1.04)
Age, sex, area-level
SES (education, recent
immigrants,
unemployment rate, and
income quintile),
urban/rural area, and
northern/southern
Ontario
Correlation (r):
NO2: 0.65
O3 0.275
Ox: 0.668
Copollutant models
: with NO2, O3, Ox
Wang et al.
(2020)
U.S.
2000-2008
Medicare,
>65
n =
52,954,845
Mortality Daily average PM2.5
concentrations estimated
from a validated
spatiotemporal
generalized additive
model on a 6 km grid
Mean (SD):
10.32 (3.15)
Associations were null by income,
(see Figure)
Age, sex, race, and ZIP
code with additional
control for ZIP code and
state SES
Correlation (r):
NO2: 0.59, Os: 0.24
Copollutant models
with: O3
September 2021
A-54
DRAFT: Do Not Cite or Quote
-------
Table A-13 (Continued): Study-specific details for epidemiologic studies examining long-term PM2.5 exposure and
health risk disparity by socioeconomic status.
Study/
Location
Years
Population
(Cohort)
Outcome
Exposure Assessment
and Long-Term Mean
and Upper Percentile
Concentrations (|jg/m3)
Select Results
Covariates in
Statistical Model
Copollutant
Examination
Weaver et al.
(2019)
2001-2010
Duke
University
Medical Center
Wake,
Durham, and
Orange
Counties in
North Carolina
Catheterizati
on Genetics
(CATHGEN)
study
Coronary Artery
Disease,
Myocardial
Infarction,
hypertension,
Diabetes
Mellitus
Daily average PM2.5
concentrations were
estimated using a hybrid
model on a 1 x 1 km grid
Mean (SD):
12.7 (1.1)
Hypertension
OR (95% CI)
Cluster 1
Cluster 2
Cluster 3
2.70 (0.95, 7.59)
11.86 (2.10, 67.21)
0.70 (0.86, 1.07)
(see Figure)
Odds ratios were
adjusted for age, sex,
BMI, race, and smoking
status
Correlation (r): NA
Copollutant models
with: NA
Wvatt et al.
(2020b)
3,132 counties
in the U.S.
1990-2010
National
Center for
Health
Statistics
Cardiovascular
mortality
Annual average PM2.5
concentrations were
estimated using
CMAQNR
In the 1990's, counties with highest Age, baseline year Correlation (r): NA
social deprivation benefited least, PM2.5 and CMR for each Copollutant models
but by 2010, counties with highest county with'NA
social deprivation benefited the
most by a reduction in PM2.5 (see
Figure)
September 2021
A-55
DRAFT: Do Not Cite or Quote
-------
Table A-13 (Continued): Study-specific details for epidemiologic studies examining long-term PM2.5 exposure and
health risk disparity by socioeconomic status.
Study/
Location
Years
Population
(Cohort)
Outcome
Exposure Assessment
and Long-Term Mean
and Upper Percentile
Concentrations (|jg/m3)
Select Results
Covariates in
Statistical Model
Copollutant
Examination
Zhang et al.
(2021)
Ontario,
Canada
2009-2017
Ontario
Health Study
n = 88,615
Total
non-accidental
mortality,
cardiovascular
mortality,
respiratory
mortality
Annual average PM2.5
estimated from model
incorporating satellite,
chemical transport, and
ground-level observation
data at a 1 x 1 km grid
Mean: 7.8
75th percentile: 8.8
Household Income
Hazard Ratio
(95% CI)
Non-Accidental
<$25,000: 1.61 (1.23, 2.12)
$25,000-$49,999: 1.42 (1.18,
$50,000-$74,999: 1.14 (0.94,
$75,000-$99,999: 1.24 (0.97,
>$100,000: 1.00 (0.82, 1.21)
Cardiovascular
<$25,000: 4.58 (2.48, 8.47)
$25,000-$49,999: 1.34 (0.91,
$50,000-$74,999: 1.10 (0.73,
$75,000-$99,999: 1.11 (0.63,
>$100,000: 1.46 (0.94, 2.25)
Respiratory
<$25,000: 3.20 (1.40, 7.34)
$25,000-$49,999
$50,000-$74,999
$75,000-$99,999
0.84 (0.45,
1.33 (0.63,
1.47 (0.45,
1.72)
1.38)
1.60)
1.97)
1.68)
1.98)
1.57)
2.79)
4.74)
Age, sex, ethnicity,
survey year, Canadian
born, educational level,
marital status, BMI, fruit
and vegetable intake,
smoking, alcohol
drinking, physical
activity, environmental
exposure to tobacco
smoke at home or in the
workplace, urban/rural,
south/north, and
neighborhood SES
characteristics (percent
recent immigrants,
percent population >15
unemployed, percent
population >15 with
educational level lower
than high school, and
income quintile)
Correlation (r): NA
Copollutant models
with: NO2
>$100,000: 4.48 (1.69, 11.83)
September 2021
A-56
DRAFT: Do Not Cite or Quote
-------
Table A-14 Study-specific details for epidemiologic studies examining race/ethnicity and PM2.5 exposure.
Study/Location
Years
Exposure Assessment
Mean Concentration (|jg/m3
Copollutant Examination
Awad et al. (2019)
U.S.
2000-2012
Annual average PM2.5 concentrations
estimated using a neural network-
based hybrid model between
2000-2012 on a 1 x 1 km grid
White
Mean: Pre-move: 11.88, Post-move: 11.15
99th Percentile: Pre-move: 20.24, Post-move: 18.08
Black
Pre-move: 13.02, Post-move: 12.12
99th Percentile: Pre-move: 20.68, Post-move: 18.95
Correlation (r): NA
Copollutant models with: NA
Erqou et al. (2018)
PM2.5 concentrations estimated using
Mean (SE)
Correlation (r): NA
Pittsburg, PA
land use regression models for the
300 m buffer surrounding an
individual's residence for the year prior
to enrollment in the study
Overall: 15.7 (0.77)
Copollutant models with: NA
2001-2014
Black: 16.1 (0.75)
White: 15.7 (0.73)
Kelly et al. (2020)
U.S.
2011
PM2.5 concentrations estimated from 9 Population-weighted average (range from models)
different exposure models which can
be described as either geophysical
process, interpolation-based,
Bayesian statistical regression,
satellite-AOD-based, or machine
learning models
NH-White: 9-10.3
Hispanic: 9.8-11.4
NH-Other: 9.4-11.5
NH-Black: 10.1-12.1
Correlation (r): NA
Copollutant models with: NA
Lee (2019)
California
2016
PM2.5 concentrations estimated from
both stationary monitors and satellite
Mean (SD): 8.09 (3.25)
25th-75th percentiles = 5.77-9.76
Correlation (r): NA
Copollutant models with: NA
September 2021
A-57
DRAFT: Do Not Cite or Quote
-------
Table A-14 (Continued): Study-specific details for epidemiologic studies examining race/ethnicity and PM2.5
exposure.
Study/Location
Years
Exposure Assessment
Mean Concentration (|jg/m3)
Copollutant Examination
Lievanos (2019)
California
2009-2011
PM2.5 concentrations estimated from
stationary monitors for each census
tract centroid using kriging
Annual mean concentrations percentile ranges:
<20th: 2.30-7.71
20-40th: 7.71-9.11
40-60th: 9.11-10.86
60-80th: 10.86-12.52
>80th: 12.54-17.04
Correlation (r): NA
Copollutant models with: NA
Lim et al. (2018)
U.S.
1995-2011
Annual average PM2.5 concentrations
estimated using a spatiotemporal
prediction model at the census tract
Mean (SD)
Overall: 11.0 (2.7)
White: 10.9 (2.7)
Black: 12.3 (2.4)
Hispanic: 11.4 (3.5)
Asian, Pacific Islander, or American Indian/Alaska Native:
11.9 (3.0)
Maximum: 21.2
Correlation (r): NO2 = 0.60,
o3 = 0.01
Copollutant models with: NA
Lipfert and Wvzqa
(2020)
U.S.
1976-2001
Average PM2.5 concentrations
estimated from stationary monitors,
averaged at the county level
Mean at cohort entry:
White: 13.9
Black: 15.7
Correlation (r):
White:
SO42: 0.68, NO2: 0.55,
peak CO: -0.19,
peakOs: 0.57, peakS02: 0.20,
PM10: 0.45
Black:
SO42: 0.50, NO2: 0.58,
peak CO: 0.16, peak O3: 0.19,
peakSCh: 0.43, PM10: 0.49
Copollutant models with: NA
Parker et al. (2018) Annual average PM2 5 concentrations Median: 11.8 Correlation (r): NA
U.S. estimated from stationary monitors 90th percentile: 14.7 Copollutant models with: NA
1997-2011
September 2021
A-58
DRAFT: Do Not Cite or Quote
-------
Table A-14 (Continued): Study-specific details for epidemiologic studies examining race/ethnicity and PM2.5
exposure.
Study/Location
Years
Exposure Assessment
Mean Concentration (|jg/m3
Copollutant Examination
Richmond-Bryant et al.
(2020)
U.S.
2008, 2011, 2014
PM2.5 emissions from fossil-fuel
Electricity Generating Units obtained
from the National Emissions Inventory
NR
Correlation (r): NA
Copollutant models with: NA
Rosofskv et al. (2018)
Massachusetts
2003-2010
PM2.5 concentrations estimated at
1 km x 1 km grid using a modeling
approach which incorporated satellite,
land use, and meteorological data
aggregated to yearly averages
Population-weighted annual average PM2.5 range:
2003: 11.1 to 11.7
2010: 7.8 to 8.4
Correlation (r): NA
Copollutant models with: NA
Tessum et al. (2019)
U.S.
2002-2015
PM2.5 concentrations estimated using
a modeling approach based on the
InMAP model and incorporating 2014
National Emissions Inventory data as
well as biogenic and wildfire emission
sources
Average Exposure:
Black: 6.0
Hispanic: 5.5
White: 4.6
Correlation (r): NA
Copollutant models with: NA
Tessum et al. (2021)
2014
PM2.5 concentrations estimated using
a modeling approach based on the
InMAP model and incorporating 2014
National Emissions Inventory data
Population average exposure from all domestic
anthropogenic sources
People of Color: 7.4
Black: 7.9
Hispanic: 7.2
Asian: 7.7
White: 5.9
Correlation (r): NA
Copollutant models with: NA
Yitshak-Sade et al.
(2020)
Massachusetts
2001-2011
Daily average PM2.5 estimated from a
model incorporating aerosol optical
depth and monitored PM2.5 at a
1 x 1 km grid
NR
Correlation (r): NA
Copollutant models with: NA
September 2021
A-59
DRAFT: Do Not Cite or Quote
-------
Table A-15 Study-specific details for epidemiologic studies examining short-term PM2.5 exposure and health risk
disparity by race/ethnicity.
Study/Location population
Years (Cohort)
Outcome
Exposure
Assessment
Short-term Mean
Concentration
(Hg/m3)
Select Results
Covariates in
Statistical Model
Copollutant
Examination
Yitshak-Sade et
al. (2019a)
Massachusetts
2001-2011
Case-crossover
Massachusetts
Department of
Public Health
n = 179,986
Cardiovascular
mortality
Daily average PM2.5
estimated from a
model incorporating
aerosol optical depth
and monitored PM2.5 at
a 1 x 1 km grid
Mean: 10.2
Maximum: 17.4
Change in CV mortality
High % White
Green: 2.80% (0.62, 5.02)
Less green: 1.14% (-1.00, 3.33)
Low % White
Green: 2.47% (0.43, 4.56)
Less green: 3.55% (1.49, 5.65)
Temperature and day
of the week
Correlation (r): NA
Copollutant models
with: NA
Yitshak-Sade et
al. (2020)
Massachusetts
2001-2011
Massachusetts
Department of
Public Health
n = 130,863
Cardiovascular
mortality
Daily average PM2.5
estimated from a
model incorporating
aerosol optical depth
and monitored PM2.5 at
a 1 x 1 km grid
NR
Change in CV mortality
Individual
Black: 4.78% (-1.99, 12.02)
White: 2.25% (0.80, 3.23)
Block group
Low % Black: 1.62% (0.05, 3.22)
High % Black: 3.35% (1.57, 5.16)
Racial Residental Segregation
(RRS)
More White (RRS 0.5, 1):
1.84% (0.31, 3.40)
More Black (RRS -1, -0.5):
15.37% (0.76, 31.99)
Index of Racial Dissimilarity (IRD)
No difference—see figure
Temperature and day
of the week
Correlation (r): NA
Copollutant models
with: NA
September 2021
A-60
DRAFT: Do Not Cite or Quote
-------
Table A-16 Study-specific details for epidemiologic studies examining long-term PM2.5 exposure and health risk
disparity by race/ethnicity.
Study/Location population
Years (Cohort) Outcome
Exposure Assessment and
Long-Term Mean and Upper
Percentile Concentrations
(Hg/m3)
Select Results
Covariates
Copollutant
Examination
Awad et al.
(2019)
U.S.
2000-2012
Medicare
n =
12,095,504
All-cause Annual average PM2.5
mortality concentrations estimated using
a neural network-based hybrid
model between 2000-2012 on
a 1 x 1 km grid
White
Mean: Pre-move: 11.88,
Post-move: 11.15
99th Percentile: Pre-move:
20.24, Post-move: 18.08
Black
Pre-move: 13.02, Post-move:
12.12
99th Percentile: Pre-move:
20.68, Post-move: 18.95
HR (95% CI)
Black movers: 1.06 (1.04, 1.07)
White movers: 1.10 (1.10, 1.10)
Age, race, sex, Medicaid
eligibility, calendar year,
hospitalization before move
for: Alzheimer's disease,
AMI, COPD, CV disease,
diabetes, heart failure, lung
cancer, Parkinson's
Disease, pneumonia, any
respiratory illness, stroke,
unstable angina, vascular
dementia and ZIP code level
variables for the new ZIP
code including: median
household income, % Black,
% Hispanic, % of owner
occupied housing units,
population density, and
median value of owner
occupied housing
Correlation (r):
NA
Copollutant
models with:
NA
Bennett et al.
(2019)
U.S.
2015
National
Center for
Health
Statistics
n = 41.9
million
Life Annual average PM2.5
expectancy concentrations estimated using
loss an integrated geographic
regression model
Median: 7.7
99th percentile: 10.1
PM2.5 associated with lower life
expectancy among counties
with, higher percent of Black or
African Americans.
Income, % in poverty, Black
race, >high school,
urbanization, unemployment,
cumulative smoking, mean
temperature, relative
humidity, county-specific
random intercepts
Correlation (r):
NA
Copollutant
models with:
NA
September 2021
A-61
DRAFT: Do Not Cite or Quote
-------
Table A-16 (Continued): Study-specific details for epidemiologic studies examining long-term PM2.5 exposure and
health risk disparity by race/ethnicity.
Study/Location population
Years (Cohort) Outcome
Exposure Assessment and
Long-Term Mean and Upper
Percentile Concentrations
(Hg/m3)
Select Results
Covariates
Copollutant
Examination
Joraenson et al. U.S.
(2020) Mortality
U 3 Database
2000-2014
Life PM2.5 concentrations estimated
expectancy from stationary monitors,
at birth averaged at the state level
Mean: 10.55
Maximum: 19.02
PM2.5 more detrimental in states
with high percent of population
of Black race.
Income share of top 10%, %
Black, total population,
median household income,
median age, % college
degree or higher
Correlation (r):
NA
Copollutant
models with:
NA
Juarez et al.
(2020)
Southeastern
U.S.
2002-2009
Southern Cardio- Annual average PM2.5
Community metabolic concentrations estimated from
Cohort Study disease a continuous, spatial surface
n = 72 215 model using ground-level
ambient air measures and
satellite-derived
measurements
Mean: 13.5
90th percentile: 15.8
Race did not modify the
relationship between PM2.5 and
CMD.
Age, history of tobacco
use/smoking status, air
quality outdoors, educational
level, household income,
marital status, gender,
residence location, race
Correlation (r):
NA
Copollutant
models with:
NA
Honda et al.
(2017)
U.S.
1993-2010
Women's
Health
Initiative
n = 23,656
Hyper-
tension
Mean (SD): 13.2 (3.0)
HR (95% CI)
White: 1.15 (0.99, 1.35)
Black: 1.26 (1.06, 1.44)
Non-White: 1.27 (1.17, 1.38)
Asian/Pacific Islander: 1.34
(1.00, 1.64)
Hispanic/Latino:
1.18 (0.99, 1.38)
Age, BMI, education,
ethnicity, smoking status,
physical activity, sodium
intake, neighborhood SES,
household income,
employment status,
insurance status, history of
high cholesterol, history of
cardiovascular disease,
history of diabetes, clinical
trial study arm and WHI
clinical site
Correlation (r):
PM10: 0.56
PM10-2.5: 0.03
Copollutant
models with:
NA
September 2021
A-62
DRAFT: Do Not Cite or Quote
-------
Table A-16 (Continued): Study-specific details for epidemiologic studies examining long-term PM2.5 exposure and
health risk disparity by race/ethnicity.
Study/Location
Years
Population
(Cohort)
Outcome
Exposure Assessment and
Long-Term Mean and Upper
Percentile Concentrations
(Hg/m3)
Select Results
Covariates
Copollutant
Examination
~metal. (2018)
U.S.
1995-2011
NIH-AARP Diabetes
Diet and Mortality
Health Study
n = 549,735
Annual average PM2.5
concentrations estimated using
a spatiotemporal prediction
model at the census tract
Mean (SD): 11.0 (2.7)
Maximum: 21.2
HR (95% CI)
Black: 1.27 (1.02, 1.58)
White: 1.05 (0.96, 1.14)
Age sex, region, race, or
ethnic group, level of
education, marital status,
BMI, alcohol consumption,
smoking status, diet, median
census tract household
income, % of census tract
population with < a high
school education
Correlation (r):
NO2 = 0.60,
O3 = 0.01
Copollutant
models with:
NA
Lipfert and
Wvzaa (2020)
U.S.
1976-2001
Veterans Mortality Average PM2.5 concentrations
Cohort estimated from stationary
Mortality monitors, averaged at the
Study county level
n = 70,000 White: 13.9
Black: 15.7
RR (95% CI)
1976-2001
White: 1.05 (1.01, 1.10)
Black: 0.82 (0.75, 0.89)
1997-2001
White: 1.03 (0.91, 1.17)
Black: 0.96 (0.76, 1.21)
individual age, race,
smoking, height, body mass
index, blood pressure,
county-wide climate and ZIP
code level socioeconomic
indicators
Correlation (r):
White:
SO42: 0.68,
NO2:0.55,
peak CO:
-0.19,
peak O3: 0.57,
peak SO2:
0.20, PM10:
0.45
Black:
SO42: 0.50,
NO2: 0.58,
peak CO: 0.16,
peak O3: 0.19,
peak SO2:
0.43, PM10:
0.49
Copollutant
models with:
NA
September 2021
A-63
DRAFT: Do Not Cite or Quote
-------
Table A-16 (Continued): Study-specific details for epidemiologic studies examining long-term PM2.5 exposure and
health risk disparity by race/ethnicity.
Study/Location
Years
Population
(Cohort)
Outcome
Exposure Assessment and
Long-Term Mean and Upper
Percentile Concentrations
(Hg/m3)
Select Results
Covariates
Copollutant
Examination
Parker et al.
(2018)
U.S.
1997-2011
National
Health
Interview
Survey
n = 657,238
All-cause
mortality
(excluding
unintentional
injuries),
heart
disease
mortality
Annual average PM2.5
concentrations estimated from
stationary monitors
Median: 11.8
90th percentile: 14.7
HR (95% CI)
All-cause
Black: 1.05 (1.03, 1.09)
White: 1.03 (1.02, 1.03)
Hispanic: 0.98 (0.94, 1.03)
Heart disease
Black: 1.04 (0.94, 1.15)
White: 1.10 (1.05, 1.15)
Hispanic: 1.03 (0.95, 1.12)
Other: 0.90 (0.75, 1.07)
Sex, family income as a % of Correlation (r):
the poverty threshold,
marital status, education,
county-level income, region
of county, urbanization, and
survey year
NA
Copollutant
models with:
NA
Son et al. (2020)
North Carolina
2002-2013
Death Total
Records Mortality
n = 775,338
Daily average PM2.5
concentrations estimated from
the CMAQ downscaler model
on a 12 ><12 grid
Mean (SD): 11.4 (5.7)
Maximum: 70.8
OR (95% CI)
NH-White: 1.01 (1.01, 1.01)
NH-Black: 1.01 (1.00, 1.02)
Hispanic: 0.97 (0.93, 1.02)
NH-Asian: 1.01 (0.97, 1.02)
NH-Other 1.01 (0.97, 1.04)
Case crossover
Correlation (r):
Os: 0.48
Copollutant
models with:
NA
Wang et al.
(2020)
U.S.
2000-2008
Medicare, Mortality
>65
n =
52,954,845
Daily average PM2.5
concentrations estimated from
a validated spatiotemporal
generalized additive model on
a 6 km grid
Mean (SD): 10.32 (3.15)
Associations were null by race.
Age, sex, race, and ZIP
code with additional control
for ZIP code and state SES
Correlation (r):
NO2: 0.59, Os:
0.24
Copollutant
models with:
Os
September 2021
A-64
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
References
Abt (Abt Associates Inc). (2001). Assessing public opinions on visibility impairment due to air pollution: Summary
report. Research Triangle Park, NC: U.S. Environmental Protection Agency.
Adar. SD: Filigrana. PA: Clements. N: Peel. JL. (2014). Ambient coarse particulate matter and human health: A
systematic review and meta-analysis [Review]. Curr Environ Health Rep 1: 258-274.
http://dx.doi.org/10.1007/s40572-014-0Q22-z
ADEO (Arizona Department of Environmental Quality). (2003). Final report-recommendation for a Phoenix area
visibility index. Phoenix, Arizona.
Adhikari. A: Yin. J. (2020). Short-Term Effects of Ambient Ozone, PM2.5, and Meteorological Factors on COVID-
19 Confirmed Cases and Deaths in Queens, New York. Int J Environ Res Public Health 17.
http://dx.doi.org/10.3390/iierphl7114047
Al-Hamdan. MZ: Crosson. WL: Economou. SA: Estes. MG. Jr: Estes. S: Hemmings. SN: Kent. ST: Puckett. M:
Ouattrochi. DA: Rickman. PL: Wade. GM: McClure. LA. (2014). Environmental public health
applications using remotely sensed data. Geocarto International 29: 85-98.
http://dx.doi.org/10.1080/10106049.2012.7152Q9
Atkinson. RW: Kang. S: Anderson. HR: Mills. IC: Walton. HA. (2014). Epidemiological time series studies of
PM2.5 and daily mortality and hospital admissions: a systematic review and meta-analysis [Review].
Thorax 69: 660-665. http://dx.doi.org/10.1136/thoraxinl-2013-204492
Awad. YA: Pi. O: Wang. Y: Choirat. C: Coull. BA: Zanobetti. A: Schwartz. J. (2019). Change in PM2.5 exposure
and mortality among Medicare recipients: Combining a semi-randomized approach and inverse probability
weights in a low exposure population. Environmental Epidemiology 3: e054.
http://dx.doi.org/10.1097/EE9.0000000000000Q54
Bai. L: Shin. S: Burnett. RT: Kwong. JC: Hvstad. P: van Ponkelaar. A: Goldberg. MS: Lavigne. E: Copes. R:
Martin. RV: Kopp. A: Chen. H. (2019). Exposure to ambient air pollution and the incidence of congestive
heart failure and acute myocardial infarction: A population-based study of 5.1 million Canadian adults
living in Ontario. Environ Int 132: 105004. http://dx.doi.Org/10.1016/i.envint.2019.105004
Barnett. AG: Williams. GM: Schwartz. J: Best. TL: Neller. AH: Petroeschevskv. AL: Simpson. RW. (2006). The
effects of air pollution on hospitalizations for cardiovascular disease in elderly people in Australian and
New Zealand cities. Environ Health Perspect 114: 1018-1023. http://dx.doi.org/10.1289/ehp.8674
Baxter. LK: Crooks. JL: Sacks. JP. (2017). Influence of exposure differences on city-to-city heterogeneity in
PM2.5-mortality associations in US cities. Environ Health 16: 1. http://dx.doi.org/10.1186/sl2940-Q16-
0208-v
Baxter. LK: Pionisio. K: Pradeep. P: Rappazzo. K: Neas. L. (2019). Human exposure factors as potential
determinants of the heterogeneity in city-specific associations between PM2.5 and mortality. J Expo Sci
Environ Epidemiol 29: 557-567. http://dx.doi.org/10.1038/s41370-018-008Q-7
Baxter. LK: Sacks. JP. (2014). Clustering cities with similar fine particulate matter exposure characteristics based
on residential infiltration and in-vehicle commuting factors. Sci Total Environ 470-471: 631-638.
http://dx.doi.Org/10.1016/i.scitotenv.2013.10.019
Bell. ML: Ebisu. K: Peng. RP: Walker. J: Samet. JM: Zeger. SL: Pominic. F. (2008). Seasonal and regional short-
term effects of fine particles on hospital admissions in 202 U.S. counties, 1999-2005. Am J Epidemiol 168:
1301-1310. http://dx.doi.org/10.1093/aie/kwn252
Bennett. JE: Tamura-Wicks. H: Parks. RM: Burnett. RT: Pope. CA: Bechle. MJ: Marshall. JP: Panaei. G: Ezzati.
M. (2019). Particulate matter air pollution and national and county life expectancy loss in the USA: A
spatiotemporal analysis. PLoS Med 16: el002856. http://dx.doi.org/10.1371/iournal.pmed.10Q2856
Berrocal. VJ: Gelfand. AE: Holland. PM. (2010a). A bivariate space-time downscaler under space and time
misalignment. Ann Appl Stat 4: 1942-1975. http://dx.doi.org/10.1214/10-AQAS351
Berrocal. VJ: Gelfand. AE: Holland. PM. (2010b). A spatio-temporal downscaler for output from numerical models.
J Agric Biol Environ Stat 15: 176-197. http://dx.doi.org/10.1007/sl3253-009-00Q4-z
Berrocal. VJ: Gelfand. AE: Holland. PM. (2012). Space-time data fusion under error in computer model output: An
application to modeling air quality. Biometrics 68: 837-848. http://dx.doi.org/10.1111/i. 1541-
0420.2011.01725.x
September 2021
R-l
PRAFT: Po Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
Bertazzon. S: Johnsoa M: Eccles. K: Kaplan. GG. (2015). Accounting for spatial effects in land use regression for
urban air pollution modeling. Spat Spatio-temporal Epidemiol 14-15: 9-21.
http://dx.doi.Org/10.1016/i.sste.2015.06.002
Bevaa GH: Freedman. DA: Lee. EK: Raiagopalaa S: Al-Kindi. SG. (2021). Association between ambient air
pollution and county-level cardiovascular mortality in the United States by social deprivation index. Am
Heart J 235: 125-131. http://dx.doi.Org/10.1016/i.ahi.2021.02.005
Beversdorf. AJ: Ziemba. LP: Chen. G: Corr. CA: Crawford. JH: Diskin. GS: Moore. RH: Thornhill. KL: Winstead.
EL: Anderson. BE. (2016). The impacts of aerosol loading, composition, and water uptake on aerosol
extinction variability in the Baltimore-Washington, DC region. Atmos ChemPhys 16: 1003-1015.
http://dx.doi.org/10.5194/acp-16-1003-2016
Bourdrel. T: Annesi-Maesano. I: Alahmad. B: Maesano. CN: Bind. MA. (2021). The impact of outdoor air pollution
on COVID-19: a review of evidence from in vitro, animal, and human studies [Review]. 30.
http://dx.doi.org/10.1183/16000617.0242-202Q
Bovs. BL: Martin. RV: van Donkelaar. A: MacDonell. RJ: Hsu. NC: Cooper. MJ: Yantosca. RM: Lu. Z: Streets.
DG: Zhang. O: Wang. SW. (2014). Fifteen-year global time series of satellite-derived fine particulate
matter. Environ Sci Technol 48: 11109-11118. http://dx.doi.org/10.1021/es502113p
Brauer. M: Brook. JR: Christidis. T: Chu. Y: Crouse. PL: Erickson. A: Hvstad. P: Li. C: Martin. RV: Meng. J:
Pappin. AJ: Pinault. LL: Tiepkema. M: van Donkelaar. A: Weichenthal. S: Burnett. RT. (2019). Mortality-
air pollution associations in low-exposure environments (MAPLE): Phase 1 (pp. 1-87). Boston, MA:
Health Effects Institute.
Cakmak. S: Hebbern. C: Pinault. L: Lavigne. E: Vanos. J: Crouse. PL: Tiepkema. M. (2018). Associations between
long-term PM2.5 and ozone exposure and mortality in the Canadian Census Health and Environment
Cohort (CANCHEC), by spatial synoptic classification zone. Environ Int 111: 200-211.
http://dx.doi.Org/10.1016/i.envint.2017.ll.030
Chakrabartv. RK: Beeler. P: Liu. P: Goswami. S: Harvey. RD: Pervez. S: van Donkelaar. A: Martin. RV. (2021).
Ambient PM2.5 exposure and rapid spread of COVID-19 in the United States. Sci Total Environ 760:
143391. http://dx.doi.Org/10.1016/i.scitotenv.2020.143391
Chen. H: Zhang. ZL: van Donkelaar. A: Bai. L: Martin. RV: Lavigne. E: Kwong. JC: Burnett. RT. (2020).
Understanding the Joint Impacts of Fine Particulate Matter Concentration and Composition on the
Incidence and Mortality of Cardiovascular Disease: A Component-Adjusted Approach. Environ Sci
Technol 54: 4388-4399. http://dx.doi.org/10.1021/acs.est.9b06861
Chi. GC: Haiat. A: Bird. CE: Cullen. MR: Griffin. BA: Miller. KA: Shih. RA: Stefanick. ML: Vedal. S: Whitsel.
EA: Kaufman. JD. (2016). Individual and neighborhood socioeconomic status and the association between
air pollution and cardiovascular disease. Environ Health Perspect 124: 1840-1847.
http://dx.doi.org/10.1289/EHP199
Chow. JC: Chen. LWA: Wang. X: Green. MC: Watson. JG. (2021). Improved estimation of PM2.5 brown carbon
contributions to filter light attenuation. Particuology 56: 1-9. http://dx.doi.Org/10.1016/i.partic.2021.01.001
Christidis. T: Erickson. AC: Pappin. AJ: Crouse. PL: Pinault. LL: Weichenthal. SA: Brook. JR: van Donkelaar. A:
Hvstad. P: Martin. RV: Tiepkema . M: Burnett. RT: Brauer. M. (2019). Low concentrations of fine particle
air pollution and mortality in the Canadian Community Health Survey cohort. Environ Health 18: 84.
http://dx.doi.org/10.1186/sl2940-019-0518-v
Corrigan. AE: Becker. MM: Neas. LM: Cascio. WE: Rappold. AG. (2018). Fine particulate matters: The impact of
air quality standards on cardiovascular mortality. Environ Res 161: 364-369.
http://dx.doi.Org/10.1016/i.envres.2017.ll.025
Crouse. D: Philip. S: van Donkelaar. A: Martin. RV: Jessiman. B: Peters. PA: Weichenthal. S: Brook. J. R.:
Hubbell. B: Burnett. RT. (2016). A new method to jointly estimate the mortality risk of long-term exposure
to fine particulate matter and its components. Sci Rep 6: 18916. http://dx.doi.org/10.1038/srepl8916
Crouse. PL: Erickson. AC: Christidis. T: Pinault. L: van Donkelaar. A: Li. C: Meng. J: Martin. RV: Tiepkema. M:
Hvstad. P: Burnett. R: Pappia A: Brauer. M: Weichenthal. S. (2020). Evaluating the Sensitivity of PM2.5-
Mortality Associations to the Spatial and Temporal Scale of Exposure Assessment. Epidemiology 31: 168-
176. http://dx.doi.org/10.1097/EPE.00000000000Q1136
Crouse. PL: Peters. PA: Hvstad. P: Brook. JR: van Ponkelaar. A: Martin. RV: Villeneuve. PJ: Jerrett. M: Goldberg.
MS: Pope. CA: Brauer. M: Brook. RP: Robichaud. A: Menard. R: Burnett. RT. (2015). Ambient PM 2.5, O
3, and NO 2 exposures and associations with mortality over 16 years of follow-up in the Canadian Census
Health and Environment Cohort (CanCHEC). Environ Health Perspect 123: 1180-1186.
http://dx.doi.org/10.1289/ehp. 1409276
September 2021
R-2
PRAFT: Po Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
Crouse. PL: Peters. PA: van Donkelaar. A: Goldberg. MS: Villeneuve. PJ: Brion. Q: Khan. S: Atari. DO: Jerrett. M:
Pope. CA: Brauer. M: Brook. JR: Martin. RV: Stieb. D: Burnett. RT. (2012). Risk of nonaccidental and
cardiovascular mortality in relation to long-term exposure to low concentrations of fine particulate matter: a
Canadian national-level cohort study. Environ Health Perspect 120: 708-714.
http://dx.doi.org/10.1289/ehp. 1104049
Crouse. PL: Pinault. L: Balram. A: Brauer. M: Burnett. RT: Martin. RV: van Donkelaar. A: Villeneuve. PJ:
Weichenthal. S. (2019). Complex relationships between greenness, air pollution, and mortality in a
population-based Canadian cohort. Environ Int 128: 292-300.
http://dx.doi.Org/10.1016/i.envint.2019.04.047
Pai. L: Zanobetti. A: Koutrakis. P: Schwartz. JP. (2014). Associations of fine particulate matter species with
mortality in the United States: a multicity time-series analysis. Environ Health Perspect 122: 837-842.
http://dx.doi.org/10.1289/ehp. 1307568
Panesh Yazdi. M: Wang. Y: Pi. O: Wei. YG: Reauia. WJ: Shi. LH: Sabath. MB: Pominici. F: Coull. BA: Evans.
JS: Koutrakis. P: Schwartz. JP. (2021). Long-term association of air pollution and hospital admissions
among Medicare participants using a doubly robust additive model. Circulation 143: 1584-1596.
http://dx.doi.org/10.1161/CIRCULATIQNAHA.120.050252
Panesh Yazdi. M: Wang. Y: Pi. O: Zanobetti. A: Schwartz. J. (2019). Long-term exposure to PM2.5 and ozone and
hospital admissions of Medicare participants in the Southeast USA. Environ Int 130: 104879.
http://dx.doi.Org/10.1016/i.envint.2019.05.073
deSouza. P: Braun. P: Parks. RM: Schwartz. J: Pominici. F: Kioumourtzoglou. MA. (2021). Nationwide Study of
Short-term Exposure to Fine Particulate Matter and Cardiovascular Hospitalizations Among Medicaid
Enrollees. Epidemiology 32: 6-13. http://dx.doi.org/10.1097/EPE.00000000000Q1265
Pi. O: Amini. H: Shi. L: Kloog. I: Silvern. R: Kelly. J: Sabath. MB: Choirat. C: Koutrakis. P: Lvapustin. A: Wang.
Y: Micklev. LJ: Schwartz. J. (2019). An ensemble-based model of PM2.5 concentration across the
contiguous United States with high spatiotemporal resolution. Environ Int 130: 104909.
http://dx.doi.Org/10.1016/i.envint.2019.104909
Pi. O: Pai. L: Wang. Y: Zanobetti. A: Choirat. C: Schwartz. JP: Pominici. F. (2017a). Association of short-term
exposure to air pollution with mortality in older adults. JAMA 318: 2446-2456.
http://dx.doi.org/10.1001/iama.2017.17923
Pi. O: Kloog. I: Koutrakis. P: Lvapustia A: Wang. Y: Schwartz. J. (2016). Assessing PM2.5 exposures with high
spatiotemporal resolution across the Continental United States. Environ Sci Technol 50: 4712-4721.
http://dx.doi.org/10.1021/acs.est.5b06121
Pi. O: Wang. Y: Zanobetti. A: Wang. Y: Koutrakis. P: Choirat. C: Pominici. F: Schwartz. JP. (2017b). Air
pollution and mortality in the Medicare population. N Engl J Med 376: 2513-2522.
http://dx.doi.org/10.1056/NEJMoal7Q2747
Pominici. F: Peng. RP: Bell. ML: Pham. L: McPermott. A: Zeger. SL: Samet. JL. (2006). Fine particulate air
pollution and hospital admission for cardiovascular and respiratory diseases. JAMA 295: 1127-1134.
http://dx.doi.org/10.100 l/iama.295.10.1127
Pominici. F: Schwartz. J: Pi. O: Braun. P: Choirat. C: Zanobetti. A. (2019). Assessing adverse health effects of
long-term exposure to low levels of ambient air pollution: Phase 1. Boston, MA: Health Effects Institute.
https://www.healtheffects.org/svstem/files/dominici-rr-200-report.pdf
Porans. KS: Wilker. EH: Li. W: Rice. MB: Liungman. PL: Schwartz. J: Coull. BA: Kloog. I: Koutrakis. P:
P'Agostino. RB: Massaro. JM: Hoffmann. U: O'Ponnell. CJ: Mittleman. MA. (2016). Residential
proximity to major roads, exposure to fine particulate matter, and coronary artery calcium: The
Framingham Heart Study. Arterioscler Thromb Vase Biol 36: 1679-1685.
http://dx.doi.org/10.1161/ATVBAHA. 116.307141
Puan. C: Talbott. E: Brooks. M: Park. SK: Broadwin. R: Matthews. K: Barinas-Mitchell. E. (2019a). Five-year
exposure to PM2.5 and ozone and subclinical atherosclerosis in late midlife women: The Study of Women's
Health Across the Nation. Int J Hyg Environ Health 222: 168-176.
http://dx.doi.Org/10.1016/i.iiheh.2018.09.001
Puan. C: Talbott. EO: Broadwin. R: Brooks. M: Matthews. K: Barinas-Mitchell. E. (2019b). Residential exposure to
PM2.5 and ozone and progression of subclinical atherosclerosis among women transitioning through
menopause: The study of women's health across the nation. J Womens Health 28: 802-811.
http://dx.doi.org/10.1089/iwh.2Q18.7182
September 2021
R-3
PRAFT: Po Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
Elliott. EG: Laden. F: James. P: Rimm. EB: Rexrode. KM: Hart. JE. (2020). Interaction between Long-Term
Exposure to Fine Particulate Matter and Physical Activity, and Risk of Cardiovascular Disease and Overall
Mortality in US Women. Environ Health Perspect 128: 127012. http://dx.doi.org/10.1289/EHP7402
Ely. DW: Learv. JT: Stewart. TR: Ross. DM. (1991). The establishment of the Denver Visibility Standard. Denver,
Colorado: Colorado Department of Health.
EPA Press Office (U.S. Environmental Protection Agency Press Office). (2021). EPA to reexamine health standards
for harmful soot that previous administration left unchanged. Available online at
https://www.epa.gov/newsreleases/epa-reexamine-health-standards-harmful-soot-previous-administration-
left-unchanged
Erickson. AC: Brauer. M: Christidis. T: Pinault. L: Crouse. PL: van Donkelaar. A: Weichenthal. S: Pappin. A:
Tiepkema. M: Martin. RV: Brook. JR: Hvstad. P: Burnett. RT. (2019). Evaluation of a method to indirectly
adjust for unmeasured covariates in the association between fine particulate matter and mortality. Environ
Res 175: 108-116. http://dx.doi.Org/10.1016/i.envres.2019.05.010
Erickson. AC: Christidis. T: Pappin. A: Brook. JR: Crouse. PL: Hvstad. P: Li. C: Martia RV: Meng. J: Pinault. L:
von Donkelaar. A: Weichenthal. S: Tiepkema. M: Burnett. RT: Brauer. M. (2020). Pisease assimilation:
The mortality impacts of fine particulate matter on immigrants to Canada. Health Rep 31: 14-26.
http://dx.doi.org/10.25318/82-003-x2020003000Q2-eng
Erqou. S: Cloughertv. JE: Olafiranve. O: Magnani. JW: Aiver. A: Tripathv. S: Kinnee. E: Kip. KE: Reis. SE. (2018).
Particulate matter air pollution and racial differences in cardiovascular disease risk. Arterioscler Thromb
Vase Biol 38: 935-942. http://dx.doi.org/10.1161/ATVBAHA. 117.310305
Eum. K: Suh. HH: Pun. V: Maniourides. J. (2018). Impact of long-term temporal trends in fine particulate matter
(PM2.5) on association of annual PM2.5 exposure and mortality: an analysis of over 20 million Medicare
beneficiaries. Environmental Epidemiology 2: e009. http://dx.doi.org/10.1097/EE9.00000000000000Q9
Evans. KA: Hopke. PK: Utell. MJ: Kane. C: Thurstoa SW: Ling. FS: Chalupa. P: Rich. PO. (2017). Triggering of
ST-elevation myocardial infarction by ambient wood smoke and other particulate and gaseous pollutants. J
Expo Sci Environ Epidemiol 27: 198-206. http://dx.doi.org/10.1038/ies.2016.15
Fan. M: Wang. Y. (2020). The impact of PM2.5 on mortality in older adults: evidence from retirement of coal-fired
power plants in the United States. Environ Health 19: 28. http://dx.doi.org/10.1186/sl2940-020-00573-2
Fisher. JA: Puett. RC: Laden. F: Wellenius. GA: Sapkota. A: Liao. P: Yanoskv. JP: Carter-Pokras. O: He. X: Hart.
JE. (2019). Case-crossover analysis of short-term particulate matter exposures and stroke in the health
professionals follow-up study. Environ Int 124: 153-160. http://dx.doi.Org/10.1016/i.envint.2018.12.044
Franklin. M: Koutrakis. P: Schwartz. J. (2008). The role of particle composition on the association between PM2.5
and mortality. Epidemiology 19: 680-689. http://dx.doi.org/10.1097/EPE.0b013e3181812bb7
Friberg. MP: Kahn. RA: Holmes. HA: Chang. HH: Sarnat. SE: Tolbert. PE: Russell. AG: Mulholland. JA. (2017).
Paily ambient air pollution metrics for five cities: Evaluation of data-fusion-based estimates and
uncertainties. Atmos Environ 158: 36-50. http://dx.doi.Org/10.1016/i.atmosenv.2017.03.022
Friberg. MP: Zhai. X: Holmes. HA: Chang. HH: Strickland. MJ: Sarnat. SE: Tolbert. PE: Russell. AG: Mulholland.
JA. (2016). Method for fusing observational data and chemical transport model simulations to estimate
spatiotemporally resolved ambient air pollution - Supplemental material [Supplemental Pata]. Environ Sci
Technol 50.
Gan. WO: Allen. RW: Brauer. M: Pavies. HW: Mancini. GB: Lear. SA. (2014). Long-term exposure to traffic-
related air pollution and progression of carotid artery atherosclerosis: a prospective cohort study. BMJ
Open 4: e004743. http://dx.doi.org/10.1136/bmiopen-2013-004743
Gardner. B: Ling. F: Hopke. PK: Frampton. MW: Utell. MJ: Zareba. W: Cameron. SJ: Chalupa. P: Kane. C:
Kulandhaisamv. S: Topf. MC: Rich. PO. (2014). Ambient fine particulate air pollution triggers ST-
elevation myocardial infarction, but not non-ST elevation myocardial infarction: A case-crossover study.
Part Fibre Toxicol 11:1. http://dx.doi.org/10.1186/1743-8977-ll-l
Goldberger. JJ: Cain. ME: Hohnloser. SH: Kadish. AH: Knight. BP: Lauer. MS: Maron. BJ: Page. RL: Passman.
RS: Siscovick. P: Stevenson. WG: Zipes. PP. (2008). American Heart Association/American College of
Cardiology Foundation/Heart Rhythm Society Scientific Statement on Noninvasive Risk Stratification
Techniques for Identifying Patients at Risk for Sudden Cardiac Peath. A scientific statement from the
American Heart Association Council on Clinical Cardiology Committee on Electrocardiography and
Arrhythmias and Council on Epidemiology and Prevention. J Am Coll Cardiol 52: 1179-1199.
http://dx.doi.Org/10.1016/i.iacc.2008.05.003
September 2021
R-4
PRAFT: Po Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
Gordon. TP: Prenni. AJ: Renfro. J. R.: Mcclure. E: Hicks. B: Onasch. TB: Freedman. A: Mcmeeking. GR: Chen. P.
(2018). Open-path, closed-path, and reconstructed aerosol extinction at a rural site. J Air Waste Manag
Assoc 68: 824-835. http://dx.doi.org/10.1080/10962247.2018.1452801
Green. R: Broadwin. R: Malig. B: Basu. R: Gold. EB: Oi. L: Sternfeld. B: Bromberger. JT: Greendale. GA: Kravitz.
HM: Tomev. K: Matthews. K: Derby. C: Jackson. EA: Green. R: Ostro. B. (2015). Long-and short-term
exposure to air pollution and inflammatory/hemostatic markers in midlife women. Epidemiology 27: 211-
220. http://dx.doi.org/10.1097/EDE.0000000000000421
Greven. S: Dominici. F: Zeger. S. (2011). An Approach to the Estimation of Chronic Air Pollution Effects Using
Spatio-Temporal Information. J Am Stat Assoc 106: 396-406. http://dx.doi.Org/10.l 198/iasa.201 l.ap09392
Han. I: Guo. Y: Afshar. M: Stock. TH: Svmanski. E. (2017). Comparison of trace elements in size-fractionated
particles in two communities with contrasting socioeconomic status in Houston, TX. Environ Monit Assess
189: 67. http://dx.doi.org/10.1007/slQ661-017-5780-2
Hand. JL: Prenni. AJ: Copeland. S: Schichtel. BA: Malm. WC. (2020). Thirty years of the Clean Air Act
Amendments: Impacts on haze in remote regions of the United States (1990-2018). Atmos Environ 243:
117865. http://dx.doi.Org/10.1016/i.atmosenv.2020.117865
Hart. JE: Liao. X: Hong. B: Puett. RC: Yanoskv. JD: Suh. H: Kioumourtzoglou. MA: Spiegelman. D: Laden. F.
(2015a). The association of long-term exposure to PM2.5 on all-cause mortality in the Nurses' Health Study
and the impact of measurement-error correction. Environ Health 14: 38. http://dx.doi.org/10.1186/sl2940-
015-0027-6
Hart. JE: Puett. RC: Rexrode. KM: Albert. CM: Laden. F. (2015b). Effect modification of long-term air pollution
exposures and the risk of incident cardiovascular disease in US women. J Am Heart Assoc 4.
http://dx.doi.org/10.1161/JAHA.115.0Q2301
Hartiala. J: Breton. CV: Tang. WH: Lurmann. F: Hazea SL: Gilliland. FD: Allavee. H. (2016). Ambient air
pollution is associated with the severity of coronary atherosclerosis and incident myocardial infarction in
patients undergoing elective cardiac evaluation. J Am Heart Assoc 5.
http://dx.doi.org/10.1161/JAHA. 116.003947
Haves. RB: Lim. C: Zhang. YL: Cromar. K: Shao. YZ: Reynolds. HR: Silverman. DT: Jones. RR: Park. Y: Jerrett.
M: Ahn. J: Thurston. GD. (2020). PM2.5 air pollution and cause-specific cardiovascular disease mortality.
Int J Epidemiol 49: 25-35. http://dx.doi.org/10.1093/iie/dvzl 14
Hemmingsen. JG: Jantzen. K: Moller. P: Loft. S. (2015a). No oxidative stress or DNA damage in peripheral blood
mononuclear cells after exposure to particles from urban street air in overweight elderly. Mutagenesis 30:
635-642. http://dx.doi.org/10.1093/mutage/gev027
Hemmingsen. JG: Rissler. J: Lvkkesfeldt. J: Sallsten. G: Kristiansen. J: Moller. P: Loft. S. (2015b). Controlled
exposure to particulate matter from urban street air is associated with decreased vasodilation and heart rate
variability in overweight and older adults. Part Fibre Toxicol 12: 6. http://dx.doi.org/10.1186/sl2989-Q15-
0081-9
Henneman. LRF: Choirat. C: Zigler. ACM. (2019). Accountability assessment of health improvements in the United
States associated with reduced coal emissions between 2005 and 2012. Epidemiology 30: 477-485.
http://dx.doi.org/10.1097/EDE.0000000000001Q24
Higbee. JD: Lefler. JS: Burnett. RT: Ezzati. M: Marshall. JD: Kim. SY: Bechle. M: Robinson. AL: Pope. CA.
(2020). Estimating long-term pollution exposure effects through inverse probability weighting methods
with Cox proportional hazards models. Environmental Epidemiology 4: e085.
http://dx.doi.org/10.1097/EE9.0000000000000Q85
Higgins. JP. (2008). Commentary: Heterogeneity in meta-analysis should be expected and appropriately quantified
[Comment]. Int J Epidemiol 37: 1158-1160. http://dx.doi.org/10.1093/iie/dvn204
Honda. T: Eliot. MN: Eaton. CB: Whitsel. E: Stewart. JD: Mu. L: Suh. H: Szpiro. A: Kaufman. JD: Vedal. S:
Wellenius. GA. (2017). Long-term exposure to residential ambient fine and coarse particulate matter and
incident hypertension in post-menopausal women. Environ Int 105: 79-85.
http://dx.doi.Org/10.1016/i.envint.2017.05.009
Host. S: Larrieu. S: Pascal. L: Blanchard. M: Declerca. C: Fabre. P: Jusot. JF: Chardon. B: Le Tertre. A: Wagner. V:
Prouvost. H: Lefranc. A. (2008). Short-term associations between fine and coarse particles and hospital
admissions for cardiorespiratory diseases in six French cities. Occup Environ Med 65: 544-551.
http://dx.doi.org/10.1136/oem.2007.036194
Janes. H: Dominici. F: Zeger. SL. (2007). Trends in air pollution and mortality: an approach to the assessment of
unmeasured confounding. Epidemiology 18: 416-423. http://dx.doi.org/10.1097/EDE.0b013e31806462e9
September 2021
R-5
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
Janssen. NAH: Fischer. P: Marra. M: Ameling. C: Cassee. FR. (2013). Short-term effects of PM2.5, PM10 and
PM2.5-10 on daily mortality in the Netherlands. Sci Total Environ 463: 20-26.
http://dx.doi.Org/10.1016/i.scitotenv.2013.05.062
Jerrett. M: Turner. MC: Beckerman. BS: Pope. CA: van Donkelaar. A: Martin. RV: Serre. M: Crouse. D: Gapstur.
SM: Krewski. D: Diver. WR: Cooean. PF: Thurston. GD: Burnett. RT. (2016). Comparing the health
effects of ambient particulate matter estimated using ground-based versus remote sensing exposure
estimates. Environ Health Perspect 125: 552-559. http://dx.doi.org/10.1289/EHP575
Jorgenson. AK: Hill. TP: Clark. B: Thombs. RP: Ore. P: Balistreri. KS: Givens. JE. (2020). Power, proximity, and
physiology: does income inequality and racial composition amplify the impacts of air pollution on life
expectancy in the United States? Environ Res Lett 15: 024013. http://dx.doi.org/10.1088/1748-
9326/ab6789
Juarez. PD: Tabatabai. M: Valdez. RB: Hood. DB: Im. W: Mouton. C: Colen. C: Al-Hamdan. MZ: Matthews-
Juarez. P: Lichtveld. MY: Sarpong. D: Ramesh. A: Langston. MA: Rogers. GL: Phillips. CA: Reichard. JF:
Donnevong. MM: Blot. W. (2020). The Effects of Social, Personal, and Behavioral Risk Factors and
PM2.5 on Cardio-Metabolic Disparities in a Cohort of Community Health Center Patients. Int J Environ
Res Public Health 17. http://dx.doi.org/10.3390/iierphl7103561
Kannel. WB: Abbott. RD: Savage. DP: McNamara. PM. (1983). Coronary heart disease and atrial fibrillation: the
Framingham Study. Am Heart J 106: 389-396.
Kaufman. JP: Adar. SD: Barr. RG: Budoff. M: Burke. GL: Curl. CL: Diez Roux. AY: Gassett. AJ: Jacobs. DR. Jr:
Kronmal. R: Larson. TV: Navas-Aciea A: Olives. C: Sampson. PD: Sheppard. L: Siscovick. PS: Stein. JH:
Szpiro. AA: Watson. KE. (2016). Association between air pollution and coronary artery calcification within
six metropolitan areas in the USA (the Multi-Ethnic Study of Atherosclerosis and Air Pollution): a
longitudinal cohort study. Lancet 388: 696-704. http://dx.doi.org/10.1016/S0140-6736(16)00378-0
Keller. JP: Larson. TV: Austin. E: Barr. RG: Sheppard. L: Vedal. S: Kaufman. JD: Szpiro. AA. (2018). Pollutant
composition modification of the effect of air pollution on progression of coronary artery calcium: the
Multi-Ethnic Study of Atherosclerosis. Environmental Epidemiology 2.
http://dx.doi.org/10.1097/EE9.0000000000000Q24
Keller. JP: Olives. C: Kim. SY: Sheppard. L: Sampson. PD: Szpiro. AA: Oron. AP: Lindstrom. J: Vedal. S:
Kaufman. JD (2015). A unified spatiotemporal modeling approach for predicting concentrations of
multiple air pollutants in the multi-ethnic study of atherosclerosis and air pollution. Environ Health
Perspect 123: 301-309. http://dx.doi.org/10.1289/ehp. 1408145
Kelly. JT: Jang. C: Timin. B: Di. O: Schwartz. J: Liu. Y: van Donkelaar. A: Martin. RV: Berrocal. V: Bell. ML.
(2020). Examining PM2.5 concentrations and exposure using multiple models. Environ Res 196: 110432.
http://dx.doi.Org/10.1016/i.envres.2020.110432
Kim. SY: Bechle. M: Hankev. S: Sheppard. L: Szpiro. AA: Marshall. JD. (2020). Concentrations of criteria
pollutants in the contiguous U.S., 1979 - 2015: Role of prediction model parsimony in integrated empirical
geographic regression. PLoS ONE 15: e0228535. http://dx.doi.org/10.1371/iournal.pone.0228535
Kim. SY: Olives. C: Sheppard. L: Sampson. PD: Larson. TV: Keller. JP: Kaufman. JD. (2017). Historical Prediction
Modeling Approach for Estimating Long-Term Concentrations of PM2.5 in Cohort Studies before the 1999
Implementation of Widespread Monitoring. Environ Health Perspect 125: 38-46.
http://dx.doi.org/10.1289/EHP131
Kleinman. L: Sedlacek. AJ. Ill: Adachi. K: Buseck. PR: Collier. S: Dubev. MK: Hodshire. AL: Lewis. E: Onasch.
TB: Pierce. JR: Shilling. J: Springston. SR: Wang. J: Zhang. O: Zhou. S: Yokelsoa RJ. (2020). Rapid
evolution of aerosol particles and their optical properties downwind of wildfires in the western US. Atmos
Chem Phys 20: 13319-13341. http://dx.doi.org/10.5194/acp-20-13319-2020
Kloog. I: Nordio. F: Zanobetti. A: Coull. BA: Koutrakis. P: Schwartz. JD. (2014). Short term effects of particle
exposure on hospital admissions in the Mid-Atlantic states: a population estimate. PLoS ONE 9: e88578.
http://dx.doi.org/10.1371/iournal.pone.0Q88578
Kloog. I: Ridgwav. B: Koutrakis. P: Coull. BA: Schwartz. JD. (2013). Long- and short-term exposure to PM2.5 and
mortality: Using novel exposure models. Epidemiology 24: 555-561.
http://dx.doi.org/10.1097/EDE.Qb013e318294beaa
Kotoa S: Molshatzki. N: Yuval. N: Myers. V: Brodav. DM: Drorv. Y: Steinberg. DM: Gerber. Y. (2013).
Cumulative exposure to particulate matter air pollution and long-term post-myocardial infarction outcomes.
Prev Med 57: 339-344. http://dx.doi.Org/10.1016/i.ypmed.2013.06.009
September 2021
R-6
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
Krall. JR: Chang. HH: Waller. LA: Mulholland. JA: Winquist. A: Talbott. EQ: Rager. JR: Tolbert. PE: Sarnat. SE.
(2018). A multicity study of air pollution and cardiorespiratory emergency department visits: Comparing
approaches for combining estimates across cities. Environ Int 120: 312-320.
http://dx.doi.Org/10.1016/i.envint.2018.07.033
Ktinzli. N: Jerrett. M: Garcia-Esteban. R: Basagafia. X: Beckermann. B: Gilliland. F: Medina. M: Peters. J: Hodis.
HN: Mack. WJ. (2010). Ambient air pollution and the progression of atherosclerosis in adults. PLoS ONE
5: e9096. http://dx.doi.org/10.1371/iournal.pone.0009Q96
Laing. JR: Jaffe. DA: Hee. JR. (2016). Physical and optical properties of aged biomass burning aerosol from
wildfires in Siberia and the Western USA at the Mt. Bachelor Observatory. Atmos Chem Phys 16: 15185-
15197. http://dx.doi.org/10.5194/acp-16-15185-2016
Lanzinger. S: Schneider. A: Breitner. S: Stafoggia. M: Erzen. I: Postal. M: Pastorkova. A: Bastian. S: Cvrvs. J:
Zscheppang. A: Kolodnitska. T: Peters. A. (2016). Associations between ultrafine and fine particles and
mortality in five central European cities - Results from the UFIREG study. Environ Int 88: 44-52.
http://dx.doi.Org/10.1016/i.envint.2015.12.006
Laupacis. A: Bovsen. G: Connolly. S. (1994). Risk factors for stroke and efficacy of antithrombotic therapy in atrial
fibrillation: Analysis of pooled data from five randomized controlled trials. Arch Intern Med 154: 1449-
1457.
Lavigne. E: Burnett. RT: Weichenthal. S. (2018). Association of short-term exposure to fine particulate air pollution
and mortality: effect modification by oxidant gases. Sci Rep 8: 16097. http://dx.doi.org/10.1038/s41598-
018-34599-x
Lee. HJ. (2019). Benefits of High Resolution PM2.5 Prediction using Satellite MAIAC AOD and Land Use
Regression for Exposure Assessment: California Examples. Environ Sci Technol 53: 12774-12783.
http://dx.doi.org/10.1021/acs.est.9b03799
Lee. HJ: Park. HY. (2020). Prioritizing the control of emission sources to mitigate PM2.5 disparity in California.
Atmos Environ 224. http://dx.doi.org/10.1016/i.atmosenv.2020.117316
Lee. M: Koutrakis. P: Coull. B: Kloog. I: Schwartz. J. (2015). Acute effect of fine particulate matter on mortality in
three Southeastern states from 2007-2011. J Expo Sci Environ Epidemiol 26: 173-179.
http://dx.doi.org/10.1038/ies.2Q15.47
Lefler. JS: Higbee. JD: Burnett. RT: Ezzati. M: Coleman. NC: Mann. DP: Marshall. JD: Bechle. M: Wang. Y:
Robinson. AL: Arden Pope. C. (2019). Air pollution and mortality in a large, representative U.S. cohort:
multiple-pollutant analyses, and spatial and temporal decompositions. Environ Health 18: 101.
http://dx.doi.org/10.1186/sl2940-019-0544-9
Leiser. CL: Smith. KR: Vanderslice. JA: Glotzbach. JP: Farrell. TW: Hanson. HA. (2019). Evaluation of the Sex-
and-Age-Specific Effects of PM2.5 on Hospital Readmission in the Presence of the Competing Risk of
Mortality in the Medicare Population of Utah 1999-2009. J Clin Med 8: 2114.
http://dx.doi.org/10.3390/icm8122114
Lepeule. J: Laden. F: Dockerv. D: Schwartz. J. (2012). Chronic exposure to fine particles and mortality: an extended
follow-up of the Harvard Six Cities study from 1974 to 2009. Environ Health Perspect 120: 965-970.
http://dx.doi.org/10.1289/ehp. 1104660
Liang. D: Shi. L: Zhao. J: Liu. P: Sarnat. JA: Gao. S: Schwartz. J: Liu. Y: Ebeit. ST: Scovronick. N: Chang. HH.
(2020). Urban air pollution may enhance COVID-19 case-fatality and mortality rates in the United States.
Innovation (N Y) 1: 100047. http://dx.doi.Org/10.1016/i.xinn.2020.100047
Liao. D: Peuauet. DJ: Duan. Y: Whitsel. EA: Dou. J: Smith. RL: Lin. HM: Chen. JC: Heiss. G. (2006). GIS
approaches for the estimation of residential-level ambient PM concentrations. Environ Health Perspect 114:
1374-1380. http://dx.doi.org/10.1289/ehp.9169
Liao. D: Whitsel. EA: Duan. Y: Lin. HM: Ouibrera. PM: Smith. R: Peuauet. DJ: Prineas. RJ: Zhang. ZM: Anderson.
G. (2009). Ambient particulate air pollution and ectopy-the environmental epidemiology of
arrhythmogenesis in Women's Health Initiative Study, 1999-2004. J Toxicol Environ Health A 72: 30-38.
http://dx.doi.org/10.1080/152873908Q2445483
Lievanos. RS. (2019). Racialized Structural Vulnerability: Neighborhood Racial Composition, Concentrated
Disadvantage, and Fine Particulate Matter in California. Int J Environ Res Public Health 16: 3196.
http://dx.doi.org/10.3390/iierphl6173196
Lim. CC: Haves. RB: Ahn. J: Shao. Y: Silverman. DT: Jones. RR: Garcia. C: Thurston. GD. (2018). Association
between long-term exposure to ambient air pollution and diabetes mortality in the US. Environ Res 165:
330-336. http://dx.doi.Org/10.1016/i.envres.2018.04.011
September 2021
R-7
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
Lipfert. FW: Wvzga. RE. (2020). Environmental predictors of survival in a cohort of U.S. military veterans: A
multi-level spatio-temporal analysis stratified by race. Environ Res 183: 108842.
http://dx.doi.Org/10.1016/i.envres.2019.108842
Lippmann. M: Chen. LC: Gordon. T: Ito. K: Thurston. GD. (2013). National Particle Component Toxicity (NPACT)
Initiative: Integrated epidemiologic and toxicologic studies of the health effects of particulate matter
components: Investigators' Report [HEI] (pp. 5-13). (177). Boston, MA: Health Effects Institute.
Lipsett. MJ: Ostro. BP: Reynolds. P: Goldberg. D: Hertz. A: Jerrett. M: Smith. DF: Garcia. C: Chang. ET:
Bernstein. L. (2011). Long-term exposure to air pollution and cardiorespiratory disease in the California
teachers study cohort. Am J Respir Crit Care Med 184: 828-835. http://dx.doi.org/10.1164/rccm.201Q12-
2082QC
Liu. C: Chen. R: Sera. F: Vicedo-Cabrera. AM: Guo. Y: Tong. S: Coelho. M: Saldiva. PHN: Lavigne. E: Matus. P:
Valdes Ortega. N: Osorio Garcia. S: Pascal. M: Stafoggia. M: Scortichini. M: Hashizume. M: Honda. Y:
Hurtado-Diaz. M: Cruz. J: Nunes. B: Teixeira. JP: Kim. H: Tobias. A: Ifiiguez. C: Forsberg. B: Astrom. C:
Ragettli. MS: Guo. YL: Chen. BY: Bell. ML: Wright. CY: Scovronick. N: Garland. RM: Miloievic. A:
Kvselv. J: Urban. A: Orru. H: Indermitte. E: Jaakkola. JJK: Rvti. NRI: Katsouvanni. K: Analitis. A:
Zanobetti. A: Schwartz. J: Chen. J: Wu. T: Cohen. A: Gasparrini. A: Kan. H. (2019). Ambient Particulate
Air Pollution and Daily Mortality in 652 Cities. N Engl J Med 381: 705-715.
http://dx.doi.org/10.1056/NEJMoal817364
Liu. X: Bertazzon. S: Villeneuve. PJ: Johnsoa M: Stieb. D: Coward. S: Tanvingoh. D: Windsor. JW: Underwood.
F: Hill. MP: Rabi. D: Ghali. WA: Wilton. SB: James. MT: Graham. M: Mcmurtrv. MS: Kaplan. GG.
(2020). Temporal and spatial effect of air pollution on hospital admissions for myocardial infarction: a
case-crossover study. CMAJ Open 8: E619-E626. http://dx.doi.org/10.9778/cmaio.20190160
Loop. MS: McClure. LA: Levitan. EB: Al-Hamdan. MZ: Crosson. WL: Safford. MM. (2018). Fine particulate
matter and incident coronary heart disease in the REGARDS cohort. Am Heart J 197: 94-102.
http://dx.doi.Org/10.1016/i.ahi.2017.ll.007
Luben. TJ: Buckley. BJ: Patel. MM: Stevens. T: Coffman. E: Rappazzo. KM: Owens. EO: Hines. EP: Moore. D:
Painter. K: Jones. R: Datko-Williams. L: Wilkie. AA: Madden. M: Richmond-Bryant. J. (2018). A cross-
disciplinary evaluation of evidence for multipollutant effects on cardiovascular disease [Review]. Environ
Res 161: 144-152. http://dx.doi.Org/10.1016/i.envres.2017.ll.008
Malik. AO: Jones. PG: Chan. PS: Peri-Okonnv. PA: Heiiaii. V: Spertus. JA. (2019). Association of long-term
exposure to particulate matter and ozone with health status and mortality in patients after myocardial
infarction. Circulation Cardiovascular Quality and Outcomes 12: e005598.
http://dx.doi.org/10.1161/CIRCOUTCOMES. 119.005598
Malm. W: Cismoski. S: Prenni. A: Peters. M. (2018). Use of cameras for monitoring visibility impairment. Atmos
Environ 175: 167-183. http://dx.doi.Org/10.1016/i.atmosenv.2017.12.005
Malm. W: Kellev. K: Molenar. J: Daniel. T. (1981). Human perception of visual air quality (uniform haze). Atmos
Environ 15: 1875-1890.
Malm. WC. (2016). Visibility: The seeing of near and distant landscape features (First ed.). Amsterdam, the
Netherlands: Elsevier. http://store.elsevier.com/Visibilitv/William-Malm/isbn-97801280445Q6/
Malm. WC: Schichtel. B: Molenar. J: Prenni. A: Peters. M. (2019). Which visibility indicators best represent a
population's preference for a level of visual air quality? J Air Waste Manag Assoc 69: 145-161.
http://dx.doi.org/10.1080/10962247.2018.150637Q
Malm. WC: Sisler. JF: Huffman. D: Eldred. RA: Cahill. TA. (1994). Spatial and seasonal trends in particle
concentration and optical extinction in the United States. J Geophys Res 99: 1347-1370.
http://dx.doi.org/10.1029/93JDQ2916
McClure. LA: Loop. MS: Crosson. W: Kleindorfer. D: Kissela. B: Al-Hamdaa M. (2017). Fine Particulate Matter
(PM2.5) and the Risk of Stroke in the REGARDS Cohort. J Stroke Cerebrovasc Dis 26: 1739-1744.
http://dx.doi.Org/10.1016/i.istrokecerebrovasdis.2017.03.041
Mendv. A: Wu. X: Keller. JL: Fassler. CS: Apewokin. S: Mersha. TB: Xie. C: Pinnev. SM. (2021). Long-term
exposure to fine particulate matter and hospitalization in COVID-19 patients. Respir Med 178: 106313.
http://dx.doi.Org/10.1016/i.rmed.2021.106313
Meng. J: Li. C: Martin. RV: van Donkelaar. A: Hvstad. P: Brauer. M. (2019). Estimated Long-Term (1981-2016)
Concentrations of Ambient Fine Particulate Matter across North America from Chemical Transport
Modeling, Satellite Remote Sensing, and Ground-Based Measurements. Environ Sci Technol 53: 5071-
5079. http://dx.doi.org/10.1021/acs.est.8b06875
September 2021
R-8
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
Mikati. I: Benson. AF: Luben. TJ: Sacks. JD: Richmond-Bryant. J. (2018). Disparities in distribution of particulate
matter emission sources by race and poverty status. Am J Public Health 2018: el-e6.
http://dx.doi.org/10.2105/AJPH.2017.304297
Miller. KA: Siscovick. PS: Sheppard. L: Shepherd. K: Sullivan. JH: Anderson. GL: Kaufman. JD. (2007). Long-
term exposure to air pollution and incidence of cardiovascular events in women. N Engl J Med 356: 447-
458. http://dx.doi.org/10.1056/NEJMoa054409
NAPCA (National Air Pollution Control Administration). (1969). Air quality criteria for particulate matter.
Washington, DC.
Nasari. MM: Szvszkowicz. M: Chen. H: Crouse. D: Turner. MC: Jerrett. M: Pope. CA: Hubbell. B: Fann. N: Cohen.
A: Gapstur. SM: Diver. WR: Stieb. D: Forouzanfar. MH: Kim. SY: Olives. C: Krewski. D: Burnett. RT.
(2016). A class of non-linear exposure-response models suitable for health impact assessment applicable to
large cohort studies of ambient air pollution. Air Qual Atmos Health 9: 961-972.
http://dx.doi.org/10.1007/sll869-016-Q398-z
NHLBI (National Institutes of Health, National Heart Lung and Blood Institute). (2017). NHLBI fact book, fiscal
year 2012: Disease statistics. Available online at
https://www.nhlbi.nih.gov/about/documents/factbook/2012/chapter4 (accessed August 23, 2017).
Ostro. B: Malig. B: Broadwin. R: Basu. R: Gold. EB: Bromberger. JT: Derby. C: Feinstein. S: Greendale. GA:
Jackson. EA: Kravitz. HM: Matthews. KA: Sternfeld. B: Tomev. K: Green. RR: Green. R. (2014). Chronic
PM2.5 exposure and inflammation: Determining sensitive subgroups in mid-life women. Environ Res 132:
168-175. http://dx.doi.Org/10.1016/i.envres.2014.03.042
Pappin. AJ: Christidis. T: Pinault. LL: Crouse. PL: Brook. JR: Erickson. A: Hvstad. P: Li. C: Martin. RV: Meng. J:
Weichenthal. S: van Donkelaar. A: Tiepkema. M: Brauer. M: Burnett. RT. (2019). Examining the Shape of
the Association between Low Levels of Fine Particulate Matter and Mortality across Three Cycles of the
Canadian Census Health and Environment Cohort. Environ Health Perspect 127: 107008.
http://dx.doi.org/10.1289/EHP5204
Parker. JD: Kravets. N: Vaidvanathan. A. (2018). Particulate matter air pollution exposure and heart disease
mortality risks by race and ethnicity in the United States: 1997-2009 NHIS with mortality followup through
2011. Circulation 137: 1688-1697. http://dx.doi.org/10.1161/CIRCULATIONAHA. 117.029376
Pascal. M: Falq. G: Wagner. V: Chatignoux. E: Corso. M: Blanchard. M: Host. S: Pascal. L: Larrieu. S. (2014).
Short-term impacts of particulate matter (PM10, PM10-2.5, PM2.5) on mortality in nine French cities.
Atmos Environ 95: 175-184. http://dx.doi.Org/10.1016/i.atmosenv.2014.06.030
Peterson. GCL: Hogrefe. C: Corrigan. AE: Neas. LM: Mathur. R: Rappold. AG. (2020). Impact of Reductions in
Emissions from Major Source Sectors on Fine Particulate Matter-Related Cardiovascular Mortality.
Environ Health Perspect 128: 17005. http://dx.doi.org/10.1289/EHP5692
Pinault. L: Brauer. M: Crouse. PL: Weichenthal. S: Erickson. A: van Donkelaar. A: Martin. RV: Charbonneau. S:
Hvstad. P: Brook. JR: Tiepkema. M: Christidis. T: Menard. R: Robichaud. A: Burnett. RT. (2018).
Piabetes Status and Susceptibility to the Effects of PM2.5 Exposure on Cardiovascular Mortality in a
National Canadian Cohort. Epidemiology 29: 784-794. http://dx.doi.org/10.1097/EPE.00000000000009Q8
Pinault. L: Tiepkema. M: Crouse. PL: Weichenthal. S: van Ponkelaar. A: Martin. RV: Brauer. M: Chen. H: Burnett.
RT. (2016). Risk estimates of mortality attributed to low concentrations of ambient fine particulate matter
in the Canadian community health survey cohort. Environ Health 15: 18. http://dx.doi.org/10.1186/sl294Q-
016-0111-6
Pinault. LL: Weichenthal. S: Crouse. PL: Brauer. M: Ericksoa A: Ponkelaar. AV: Martin. RV: Hvstad. P: Chen. H:
Fines. P: Brook. JR: Tiepkema. M: Burnett. RT. (2017). Associations between fine particulate matter and
mortality in the 2001 Canadian Census Health and Environment Cohort. Environ Res 159: 406-415.
http://dx.doi.Org/10.1016/i.envres.2017.08.037
Pitchford. M: Maim. W: Schichtel. B: Kumar. N: Lowenthal. P: Hand. J. (2007). Revised algorithm for estimating
light extinction from IMPROVE particle speciation data. J Air Waste Manag Assoc 57: 1326-1336.
Pope. CA. Ill: Thua MJ: Namboodiri. MM: Pockerv. PW: Evans. JS: Speizer. FE: Heath. CW. Jr. (1995).
Particulate air pollution as a predictor of mortality in a prospective study of U.S. adults. Am J Respir Crit
Care Med 151: 669-674. http://dx.doi.org/10.1164/airccm/15L3 Pt 1.669
Pope. CA: Lefler. JS: Ezzati. M: Higbee. JP: Marshall. JP: Kim. SY: Bechle. M: Gilliat. KS: Vernon. SE:
Robinson. AL: Burnett. RT. (2019). Mortality Risk and Fine Particulate Air Pollution in a Large,
Representative Cohort of U.S. Adults. Environ Health Perspect 127: 77007.
http://dx.doi.org/10.1289/EHP4438
September 2021
R-9
PRAFT: Po Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
Pope. CA: Turner. MC: Burnett. R: Jerrett. M: Gapstur. SM: Diver. WR: Krewski. D: Brook. RD. (2014).
Relationships between fine particulate air pollution, cardiometabolic disorders and cardiovascular
mortality. Circ Res 116: 108-U258. http://dx.doi.org/10.1161/CIRCRESAHA. 116.305060
Prenni. AJ: Hand. JL: Malm. WC: Copeland. S: Luo. G: Yu. F: Taylor. N: Russell. LM: Schichtel. BA. (2019). An
examination of the algorithm for estimating light extinction from IMPROVE particle speciation data.
Atmos Environ 214. http://dx.doi.Org/10.1016/i.atmosenv.2019.116880
Prvor. SC. (1996). Assessing public perception of visibility for standard setting exercises. Atmos Environ 30: 2705-
2716. http://dx.doi.org/10.1016/1352-2310(95)00365-7
Prvstowskv. EN: Benson. DW. Jr: Fuster. V: Hart. RG: Kav. GN: Mverburg. RJ: Naccarelli. GV: Wvse. DG.
(1996). Management of patients with atrial fibrillation. A Statement for Healthcare Professionals. From the
Subcommittee on Electrocardiography and Electrophysiology, American Heart Association. Circulation 93:
1262-1277.
Pun. VC: Kazemiparkouhi. F: Maniourides. J: Suh. HH. (2017). Long-term PM2.5 exposures and respiratory, cancer
and cardiovascular mortality in American older adults. Am J Epidemiol 186: 961-969.
http://dx.doi.org/10.1093/aie/kwxl66
Oiu. X: Wei. Y: Wang. Y: Pi. O: Sofer. T: Awad. YA: Schwartz. J. (2020). Inverse probability weighted distributed
lag effects of short-term exposure to PM2.5 and ozone on CVD hospitalizations in New England Medicare
participants - Exploring the causal effects. Environ Res 182: 109095.
http://dx.doi.Org/10.1016/i.envres.2019.109095
Rappazzo. KM: Joodi. G: Hoffman. SR: Pursell. IW: Mounsev. JP: Cascio. WE: Simpson. RJ. (2019). A case-
crossover analysis of the relationship of air pollution with out-of-hospital sudden unexpected death in
Wake County, North Carolina (2013-2015). Sci Total Environ 694: 133744.
http://dx.doi.Org/10.1016/i.scitotenv.2019.133744
Rhinehart. ZJ: Kinnee. E: Essien. UR: Saul. M: Guhl. E: Cloughertv. JE: Magnani. JW. (2020). Association of Fine
Particulate Matter and Risk of Stroke in Patients With Atrial Fibrillation. JAMA Netw Open 3: e2011760.
http://dx.doi.org/10.1001/iamanetworkopen.2020.1176Q
Richmond-Bryant. J: Mikati. I: Benson. AF: Lubea TJ: Sacks. JD. (2020). Disparities in distribution of particulate
matter emissions from US coal-fired power plants by race and poverty status after accounting for
reductions in operations between 2015 and 2017. Am J Public Health 110: 655-661.
http://dx.doi.org/10.2105/AJPH.2019.305558
Rosofskv. A: Levy. JI: Zanobetti. A: Janulewicz. P: Fabian. MP. (2018). Temporal trends in air pollution exposure
inequality in Massachusetts. Environ Res 161: 76-86. http://dx.doi.Org/10.1016/i.envres.2017.10.028
Roy. D: Talaiic. M: Dubuc. M: Thibault. B: Guerra. P: Made. L: Khairv. P. (2009). Atrial fibrillation and
congestive heart failure. Curr Opin Cardiol 24: 29.
Samoli. E: Stafoggia. M: Rodopoulou. S: Ostro. B: Declerca. C: Alessandrini. E: Diaz. J: Karanasiou. A: Kelessis.
AG: Le Tertre. A: Pandolfi. P: Randi. G: Scarinzi. C: Zauli-Saiani. S: Katsouvanni. K: Forastiere. F.
(2013). Associations between fine and coarse particles and mortality in Mediterranean cities: Results from
the MED-PARTICLES Project. Environ Health Perspect 121: 932-938.
http://dx.doi.org/10.1289/ehp. 1206124
Sanders. NJ: Barreca. AI: Neidell. MJ. (2020). Estimating Causal Effects of Particulate Matter Regulation on
Mortality. Epidemiology 31: 160-167. http://dx.doi.org/10.1097/EDE.00000000000Q1153
Schulz. AJ: Mentz. GB: Sampson. N: Ward. M: Dvonch. JT: de Maio. R: Israel. BA: Reyes. AG: Wilkins. D.
(2018). Independent and Joint Contributions of Fine Particulate Matter Exposure and Population
Vulnerability to Mortality in the Detroit Metropolitan Area. Int J Environ Res Public Health 15.
http://dx.doi.org/10.3390/iierphl50612Q9
Schwartz. J: Coull. B: Laden. F: Ryan. L. (2008). The effect of dose and timing of dose on the association between
airborne particles and survival. Environ Health Perspect 116: 64-69. http://dx.doi.org/10.1289/ehp.9955
Schwartz. J: Fong. K: Zanobetti. A. (2018a). A National Multicity Analysis of the Causal Effect of Local Pollution,
[Formula: see text], and [Formula: see text] on Mortality. Environ Health Perspect 126: 87004.
http://dx.doi.org/10.1289/EHP2732
Schwartz. J: Wei. Y: Yitshak-Sade. M: Pi. O: Dominici. F: Zanobetti. A. (2021). A national difference in
differences analysis of the effect of PM2.5 on annual death rates. Environ Res 194: 110649.
http://dx.doi.Org/10.1016/i.envres.2020.110649
Schwartz. JD: Wang. Y: Kloog. I: Yitshak-Sade. M: Dominici. F: Zanobetti. A. (2018b). Estimating the Effects of
PM2.5 on Life Expectancy Using Causal Modeling Methods. Environ Health Perspect 126: 127002.
http://dx.doi.org/10.1289/EHP3130
September 2021
R-10
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
Shi. L: Zanobetti. A: Kloog. I; Coull. BA: Koutrakis. P: Mellv. SJ: Schwartz. JD. (2015). Low-concentration PM2.5
and mortality: estimating acute and chronic effects in a population-based study. Environ Health Perspect
124: 46-52. http://dx.doi.org/10.1289/ehp. 1409111
Shia HH: Cakmak. S: Brioq O: Villeneuve. P: Turner. MC: Goldberg. MS: Jerrett. M: Chen. H: Crouse. D: Peters.
P: Pope. CA: Burnett. RT. (2014). Indirect adjustment for multiple missing variables applicable to
environmental epidemiology. Environ Res 134: 482-487. http://dx.doi.Org/10.1016/i.envres.2014.05.016
Shia HH: Gogna. P: Maauiling. A: Paraiuli. RP: Hague. L: Burr. B. (2021a). Comparison of hospitalization and
mortality associated with short-term exposure to ambient ozone and PM2.5 in Canada. Chemosphere 265:
128683. http://dx.doi.Org/10.1016/i.chemosphere.2020.128683
Shia HH: Paraiuli. RP: Gogna. P: Maauiling. A: Dehghani. P. (2021b). Pollutant-sex specific differences in
respiratory hospitalization and mortality risk attributable to short-term exposure to ambient air pollution.
Sci Total Environ 755: 143135. http://dx.doi.Org/10.1016/i.scitotenv.2020.143135
Shia S: Burnett. RT: Kwong. JC: Hvstad. P: van Donkelaar. A: Brook. J. R.: Goldberg. MS: Tu. K: Copes. R:
Martia R: Lia. Y: Kopp. A: Chen. H. (2019). Ambient Air Pollution and the Risk of Atrial Fibrillation and
Stroke: A Population-Based Cohort Study. Environ Health Perspect 127: 87009.
http://dx.doi.org/10.1289/EHP4883
Son. JY: Lane. KJ: Miranda. ML: Bell. ML. (2020). Health disparities attributable to air pollutant exposure in North
Carolina: Influence of residential environmental and social factors. Health Place 62: 102287.
http://dx.doi.Org/10.1016/i.healthplace.2020.102287
Stieb. DM: Evans. GJ: To. TM: Brook. JR: Burnett. RT. (2020). An ecological analysis of long-term exposure to
PM2.5 and incidence of COVID-19 in Canadian health regions. Environ Res 191: 110052.
http://dx.doi.Org/10.1016/i.envres.2020.110052
Sun. S: Stewart. JD: Eliot. MN: Yanoskv. JD: Liao. D: Tinker. LF: Eaton. CB: Whitsel. EA: Wellenius. GA. (2019).
Short-term exposure to air pollution and incidence of stroke in the Women's Health Initiative. Environ Int
132: 105065. http://dx.doi.Org/10.1016/i.envint.2019.105065
Tanzer. R: Malings. C: Haurvliuk. A: Subramanian. R: Presto. AA. (2019). Demonstration of a Low-Cost Multi-
Pollutant Network to Quantify Intra-Urban Spatial Variations in Air Pollutant Source Impacts and to
Evaluate Environmental Justice. Int J Environ Res Public Health 16.
http://dx.doi.org/10.3390/iierphl6142523
Tessum. CW: Apte. JS: Goodkind. AL: Muller. NZ: Mullins. KA: Paolella. DA: Polaskv. S: Springer. NP: Thakrar.
SK: Marshall. JD: Hill. JD. (2019). Inequity in consumption of goods and services adds to racial-ethnic
disparities in air pollution exposure. Proc Natl Acad Sci USA 116: 6001-6006.
http://dx.doi.org/10.1073/pnas.1818859116
Tessum. CW: Paolella. DA: Chambliss. SE: Apte. JS: Hill. JD: Marshall. JD. (2021). PM2.5 polluters
disproportionately and systemically affect people of color in the United States. Science Advances 7.
http://dx.doi.org/10.1126/sciadv.abf4491
Tonne. C: Halonea JI: Beevers. SD: Dainak. D: Gulliver. J: Kelly. FJ: Wilkinson. P: Anderson. HR. (2015). Long-
term traffic air and noise pollution in relation to mortality and hospital readmission among myocardial
infarction survivors. Int J Hyg Environ Health 219: 72-78. http://dx.doi.Org/10.1016/i.iiheh.2015.09.003
U.S. EPA (U.S. Environmental Protection Agency). (1982). Air quality criteria for particulate matter and sulfur
oxides (final, 1982) [EPA Report]. (EPA 600/8-82/029a). Washington, DC: Environmental Criteria and
Assessment Office, http://cfpub.epa.gov/ncea/cfm/recordisplav,cfm?deid=46205
U.S. EPA (U.S. Environmental Protection Agency). (1996). Air quality criteria for particulate matter [EPA Report].
(EPA/600/P-95/001aF-cF. 3v). Research Triangle Park, National Center for Environmental Assessment-
RTP Office NC.
U.S. EPA (U.S. Environmental Protection Agency). (2004). Air quality criteria for particulate matter [EPA Report].
(EPA/600/P-99/002aF-bF). Research Triangle Park, NC: U.S. Environmental Protection Agency, Office of
Research and Development, National Center for Environemntal Assessment- RTP Office.
http://cfpub.epa.gov/ncea/cfm/recordisplav.cfm?deid=87903
U.S. EPA (U.S. Environmental Protection Agency). (2009). Integrated science assessment for particulate matter
[EPA Report]. (EPA/600/R-08/139F). Research Triangle Park, NC: U.S. Environmental Protection
Agency, Office of Research and Development, National Center for Environmental Assessment- RTP
Division. http://cfpub.epa.gov/ncea/cfm/recordisplav.cfm?deid=216546
September 2021
R-ll
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
U.S. EPA (U.S. Environmental Protection Agency). (2013). Integrated science assessment for ozone and related
photochemical oxidants [EPA Report]. (EPA/600/R-10/076F). Research Triangle Park, NC: U.S.
Environmental Protection Agency, Office of Research and Development, National Center for
Environmental Assessment-RTP Division, http://cfpub.epa. gov/ncea/isa/recordisplav.cfm?deid=247492
U.S. EPA (U.S. Environmental Protection Agency). (2015). Preamble to the integrated science assessments [EPA
Report]. (EPA/600/R-15/067). Research Triangle Park, NC: U.S. Environmental Protection Agency, Office
of Research and Development, National Center for Environmental Assessment, RTP Division.
https://cfpub.epa.gov/ncea/isa/recordisplav.cfm?deid=310244
U.S. EPA (U.S. Environmental Protection Agency). (2016). Integrated review plan for the national ambient air
quality standards for particulate matter [EPA Report]. (EPA-452/R-16-005). Research Triangle Park, NC.
https://vosemite.epa.gov/sab/sabproduct.nsf/0/eb862b233fbd0cde85257dda004fcb8c!QpenDocument&Tabl
eRow=2.0
U.S. EPA (U.S. Environmental Protection Agency). (2018). Supplemental material: Chapter 6 (Cardiovascular
Effects) Integrated Science Assessment for Particulate Matter.
U.S. EPA (U.S. Environmental Protection Agency). (2019). Integrated Science Assessment (ISA) for particulate
matter (final report, Dec 2019) [EPA Report]. (EPA/600/R-19/188). Washington, DC.
https ://cfpub .epa. gov/ncea/isa/recordisplav .cfm?deid=3 47534
U.S. EPA (U.S. Environmental Protection Agency). (2020a). Integrated science assessment for oxides of nitrogen,
oxides of sulfur, and particulate matter—ecological criteria (final). (EPA/600/R-20/278). Research Triangle
Park, NC. https://cfpub.epa. gov/ncea/isa/recordisplav,cfm?deid=349473
U.S. EPA (U.S. Environmental Protection Agency). (2020b). Policy assessment for the review of the National
Ambient Air Quality Standards for particulate matter. (EPA-452/R-20-002). Research Triangle Park, NC.
U.S. EPA (U.S. Environmental Protection Agency). (2020c). Responses to significant comments on the 2020
proposed rule on the National Ambient Air Quality Standards for particulate matter (April 30, 2020; 85 FR
24094). (EPA-HQ-OAR-2015-0072). https://www.epa.gov/sites/production/files/202Q-
12/documents/pm naaas response to comments final.pdf
U.S. EPA (U.S. Environmental Protection Agency). (2021). Michael Regan administrator message to EPA
employees on commitment to environmental justice, April 07, 2021. Available online
Ueda. K: Nitta. H: Ono. M: Takeuchi. A. (2009). Estimating mortality effects of fine particulate matter in Japan: A
comparison of time-series and case-crossover analyses. J Air Waste Manag Assoc 59: 1212-1218.
http://dx.doi.org/10.3155/1047-3289.59.10.1212
van Donkelaar. A: Martin. RV: Brauer. M: Kahn. R: Lew. R: Verduzco. C: Villeneuve. PJ. (2010). Global estimates
of ambient fine particulate matter concentrations from satellite-based aerosol optical depth: development
and application. Environ Health Perspect 118: 847-855. http://dx.doi.org/10.1289/ehp.0901623
van Donkelaar. A: Martin. RV: Li. C: Burnett. RT. (2019). Regional estimates of chemical composition of fine
particulate matter using a combined geoscience-statistical method with information from satellites, models,
and monitors. Environ Sci Technol 53: 2595-2611. http://dx.doi.org/10.1021/acs.est.8b06392
van Donkelaar. A: Martin. RV: Spurr. RJ: Burnett. RT. (2015). High-resolution satellite-derived PM2.5 from
optimal estimation and geographically weighted regression over North America. Environ Sci Technol 49:
10482-10491. http://dx.doi.org/10.1021/acs.est.5b02Q76
Villeneuve. PJ: Goldberg. MS. (2020). Methodological Considerations for Epidemiological Studies of Air Pollution
and the SARS and COVID-19 Coronavirus Outbreaks [Review]. Environ Health Perspect 128: 95001.
http://dx.doi.org/10.1289/EHP7411
Wang. B: Eum. KD: Kazemiparkouhi. F: Li. C: Maniourides. J: Pavlu. V: Suh. H. (2020). The impact of long-term
PM2.5 exposure on specific causes of death: exposure-response curves and effect modification among 53
million U.S. Medicare beneficiaries. Environ Health 19: 20. http://dx.doi.org/10.1186/sl2940-020-0Q575-0
Wang. M: Hopke. PK: Masiol. M: Thurston. SW: Cameron. S: Ling. F: van Wiingaardea E: Croft. D: Sauizzato. S:
Thevenet-Morrison. K: Chalupa. D: Rich. DO. (2019). Changes in triggering of ST-elevation myocardial
infarction by particulate air pollution in Monroe County, New York over time: a case-crossover study.
Environ Health 18: 82. http://dx.doi.org/10.1186/sl2940-019-0521-3
Ward-Caviness. CK: Weaver. AM: Buranoskv. M: Pfaff. ER: Neas. LM: Devlin. RB: Schwartz. J: Pi. O: Cascio.
WE: Diaz-Sanchez. D. (2020). Associations between long-term fine particulate matter exposure and
mortality in heart failure patients. J Am Heart Assoc 9: e012517.
http://dx.doi.org/10.1161/JAHA.119.012517
September 2021
R-12
DRAFT: Do Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
Weaver. AM: Mcguinn. L: Neas. L: Mirowskv. J: Devlin. RB: Phingra. R: Ward-Caviness. C: Cascio. WE: Kraus.
WE: Hauser. ER: Pi. O: Schwartz. J: Diaz-Sanchez. D. (2019). Neighborhood sociodemographic effects on
the associations between long-term PM2.5 exposure and cardiovascular outcomes and diabetes.
Environmental Epidemiology 3. http://dx.doi.org/10.1097/EE9.0000000000000Q38
Weaver. AM: Wang. Y: Wellenius. GA: Bidulescu. A: Sims. M: Vaidvanathan. A: Hickson. DA: Shimbo. D:
Abdalla. M: Diaz. KM: Seals. SR. (2021). Long-term air pollution and blood pressure in an African
American cohort: The Jackson Heart Study. Am J Prev Med 60: 397-405.
http://dx.doi.Org/10.1016/i.amepre.2020.10.023
Wei. Y: Wang. Y: Pi. O: Choirat. C: Wang. Y: Koutrakis. P: Zanobetti. A: Dominici. F: Schwartz. JD. (2019). Short
term exposure to fine particulate matter and hospital admission risks and costs in the Medicare population:
time stratified, case crossover study. BMJ 367:16258. http://dx.doi.org/10.1136/bmi.16258
Wei. Y: Wang. Y: Wu. X: Pi. O: Shi. L: Koutrakis. P: Zanobetti. A: Dominici. F: Schwartz. JD. (2020). Causal
effects of air pollution on mortality in Massachusetts. Am J Epidemiol 189: 1316-1323.
http://dx.doi.org/10.1093/aie/kwaa098
Wei. Y: Yazdi. MP: Pi. O: Reauia. WJ: Pominici. F: Zanobetti. A: Schwartz. J. (2021a). Emulating causal dose-
response relations between air pollutants and mortality in the Medicare population. Environ Health 20: 53.
http://dx.doi.org/10.1186/sl2940-021-0Q742-x
Wei. YG: Coull. B: Koutrakis. P: Yang. JB: Li. LX: Zanobetti. A: Schwartz. J. (2021b). Assessing additive effects
of air pollutants on mortality rate in Massachusetts. Environ Health 20: 19.
http://dx.doi.org/10.1186/sl2940-021-007Q4-3
WHO (World Health Organization). (1989). Recommendations on stroke prevention, diagnosis, and therapy. Report
of the WHO task force on stroke and other cerebrovascular disorders [Editorial]. Stroke 20: 1407-1431.
http://dx.doi.org/10.1161/01.STR.20.10.14Q7
Wing. JJ: Adar. SP: Sanchez. BN: Morgenstern. LB: Smith. MA: Lisabeth. LP. (2017). Short-term exposures to
ambient air pollution and risk of recurrent ischemic stroke. Environ Res 152: 304-307.
http://dx.doi.Org/10.1016/i.envres.2016.ll.001
Wu. X: Braun. P: Kioumourtzoglou. MA: Choirat. C: Pi. O: Pominici. F. (2019). Causal inference in the context of
an error prone exposure: Air pollution and mortality. Ann Appl Stat 13: 520-547.
http://dx.doi.org/10.1214/18-AQAS1206
Wu. X: Braun. P: Schwartz. J: Kioumourtzoglou. MA: Pominici. F. (2020a). Evaluating the impact of long-term
exposure to fine particulate matter on mortality among the elderly. Science Advances 6: eaba5692.
http://dx.doi.org/10.1126/sciadv.aba5692
Wu. X: Netherlv. RC: Sabath. MB: Braun. P: Pominici. F. (2020b). Air pollution and COVIP-19 mortality in the
United States: Strengths and limitations of an ecological regression analysis. Science Advances 6:
eabd4049. http://dx.doi.org/10.1126/sciadv.abd4049
Wvatt. LH: Pevlin. RB: Rappold. A: Case. MW: Piaz-Sanchez. P. (2020a). Low levels of fine particulate matter
increase vascular damage and reduce pulmonary function in young healthy adults. Part Fibre Toxicol 17:
58. http://dx.doi.org/10.1186/sl2989-020-0Q389-5
Wvatt. LH: Peterson. GCL: Wade. TJ: Neas. LM: Rappold. AG. (2020b). The contribution of improved air quality
to reduced cardiovascular mortality: Peclines in socioeconomic differences over time. Environ Int 136:
105430. http://dx.doi.Org/10.1016/i.envint.2019.105430
Wvatt. LH: Xi. YZ: Kshirsagar. A: Pi. O: Ward-Caviness. C: Wade. TJ: Cascio. WE: Rappold. AG. (2020c).
Association of short-term exposure to ambient PM2.5 with hospital admissions and 30-day readmissions in
end-stage renal disease patients: population-based retrospective cohort study. BMJ Open 10: e041177.
http://dx.doi.org/10.1136/bmiopen-2020-Q41177
Yanoskv. JP: Paciorek. CJ: Laden. F: Hart. JE: Puett. RC: Liao. P: Suh. HH. (2014a). Spatio-temporal modeling of
particulate air pollution in the conterminous United States using geographic and meteorological predictors.
Environ Health 13: 63. http://dx.doi.org/10.1186/1476-069X-13-63
Yanoskv. JP: Paciorek. CJ: Laden. F: Hart. JE: Puett. RC: Liao. P: Suh. HH. (2014b). Spatio-temporal modeling of
particulate air pollution in the conterminous United States using geographic and meteorological predictors -
Supplementary material [Supplemental Pata]. Environ Health 13.
Ye. P: Klein. M: Mulholland. JA: Russell. AG: Weber. R: Edgerton. ES: Chang. HH: Sarnat. JA: Tolbert. PE: Ebelt
Sarnat. S. (2018). Estimating Acute Cardiovascular Effects of Ambient PM2.5 Metals. Environ Health
Perspect 126: 027007. http://dx.doi.org/10.1289/EHP2182
September 2021
R-13
PRAFT: Po Not Cite or Quote
-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
Yitshak-Sade. M: James. P: Kloog. I; Hart. JE: Schwartz. JD: Laden. F: Lane. KJ: Fabian. MP: Fong. KG':
Zanobetti. A. (2019a). Neighborhood Greenness Attenuates the Adverse Effect of PM2.5 on
Cardiovascular Mortality in Neighborhoods of Lower Socioeconomic Status. Int J Environ Res Public
Health 16. http://dx.doi.org/10.3390/iierphl6050814
Yitshak-Sade. M: Kloog. I: Zanobetti. A: Schwartz. JD. (2019b). Estimating the causal effect of annual PM2.5
exposure on mortality rates in the Northeastern and mid-Atlantic states. Environmental Epidemiology 3:
e052. http://dx.doi.org/10.1097/EE9.0000000000000Q52
Yitshak-Sade. M: Lane. KJ: Fabian. MP: Kloog. I: Hart. JE: Davis. B: Fong. KC: Schwartz. JD: Laden. F:
Zanobetti. A. (2020). Race or racial segregation? Modification of the PM2.5 and cardiovascular mortality
association. PLoS ONE 15: e0236479. http://dx.doi.org/10.1371/iournal.pone.0236479
Young. SS: Smith. RL: Lopiano. KK. (2017). Air quality and acute deaths in California, 2000-2012. Regul Toxicol
Pharmacol 88: 173-184. http://dx.doi.Org/10.1016/i.vrtph.2017.06.003
Zanobetti. A: Dominici. F: Wang. Y: Schwartz. JD. (2014). A national case-crossover analysis of the short-term
effect of PM2.5 on hospitalizations and mortality in subjects with diabetes and neurological disorders.
Environ Health 13: 38. http://dx.doi.org/10.1186/1476-069X-13-38
Zanobetti. A: Schwartz. J. (2009). The effect of fine and coarse particulate air pollution on mortality: A national
analysis. Environ Health Perspect 117: 1-40. http://dx.doi.org/10.1289/ehp.08001Q8
Zhang. W: Lia S: Hopke. PK: Thurstoa SW: van Wiingaardea E: Croft. D: Sauizzato. S: Masiol. M: Rich. DO.
(2018). Triggering of cardiovascular hospital admissions by fine particle concentrations in New York state:
Before, during, and after implementation of multiple environmental policies and a recession. Environ Pollut
242: 1404-1416. http://dx.doi.Org/10.1016/i.envpol.2018.08.030
Zhang. Z: Wang. J: Kwong. JC: Burnett. RT: van Donkelaar. A: Hvstad. P: Martin. RV: Bai. L: McLaughlin. J:
Chen. H. (2021). Long-term exposure to air pollution and mortality in a prospective cohort: The Ontario
Health Study. Environ Int 154: 106570. http://dx.doi.Org/10.1016/i.envint.2021.106570
September 2021
R-14
DRAFT: Do Not Cite or Quote
------- |