oEPA
EPA/600/R 22/0281 May 2022 | www.epa.gov/isa
United States
Environmental Protection
Agency
Supplement to the 2019 Integrated Science
Assessment for Particulate Matter
-------
United States
Environmental Protection
»m Agency
EPA/600/R-22/028
May 2022
www. epa. gov/isa
Supplement to the 2019 Integrated
Science Assessment
for
Particulate Matter
May 2022
Center for Public Health and Environmental Assessment
Office of Research and Development
U.S. Environmental Protection Agency
Research Triangle Park, NC
-------
DISCLAIMER
This document has been reviewed in accordance with the U.S. Environmental Protection Agency
policy and approved for publication. Mention of trade names or commercial products does not constitute
endorsement or recommendation for use.
May 2022
ii
-------
CONTENTS
INTEGRATED SCIENCE ASSESSMENT TEAM FOR PARTICULATE MATTER xi
ACRONYMS AND ABBREVIATIONS XIII
EXECUTIVE SUMMARY ES-I
1. INTRODUCTION AND SCOPE 1-1
1.1. Introduction 1-1
1.2. Rationale and Scope 1-2
1.2.1. Rational for Inclusion of Health and Welfare Effects 1-2
1.2.2. Scope 1-4
1.3. Development of the Supplement 1-5
1.4. Organization of the Supplement 1-6
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 ofPIVh.s 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 2-12
2.1.1.4.1. Cancer Associated with Long-Term PM2.5 Exposure 2-12
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 2-30
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 2-37
2.3.1. Visibility Impairment 2-38
2.3.2. Climate Effects 2-38
2.3.3. Materials Effects 2-39
May 2022
iii
-------
3. EVALUATION OF RECENT HEALTH EFFECTS EVIDENCE 3-1
3.1. Cardiovascular Effects 3-2
3.1.1. Short-Term PM2.5 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 3-8
3.1.1.2.1. Ischemic Heart Disease and Myocardial Infarction 3-8
3.1.1.2.2. Cerebrovascular Disease and Stroke 3-13
3.1.1.2.3. Heart Failure 3-16
3.1.1.2.4. Arrhythmia 3-19
3.1.1.2.5. Combinations of Cardiovascular-Related Outcomes 3-22
3.1.1.2.6. Cardiovascular Mortality 3-23
3.1.1.2.7. Consideration of Copollutant Exposures 3-23
3.1.1.2.8. Lag Structure of Associations 3-24
3.1.1.3. Recent Epidemiologic Studies Examining the PM2.s-Cardiovascular Effects
Relationship through Accountability Analyses and Alternative Methods for
Confounder Control 3-24
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 3-28
3.1.2. Long-Term PM2.5 Exposure 3-28
3.1.2.1. Summary and Causality Determination from 2019 Integrated Science Assessment
for Particulate Matter 3-29
3.1.2.2. Recent U.S. and Canadian Epidemiologic Studies 3-34
3.1.2.2.1. Ischemic Heart Disease and Myocardial Infarction 3-35
3.1.2.2.2. Cerebrovascular Disease and Stroke 3-38
3.1.2.2.3. Atherosclerosis 3-40
3.1.2.2.4. Heart Failure and Impaired Heart Function 3-41
3.1.2.2.5. Cardiac Electrophysiology and Arrhythmia 3-42
3.1.2.2.6. Blood Pressure and Hypertension 3-43
3.1.2.2.7. Cardiovascular Mortality 3-44
3.1.2.2.8. Copollutant Confounding 3-44
3.1.2.2.9. Examination of the Concentration-Response (C-R) Relationship between
Long-Term PM2.5 Exposure and Cardiovascular Effects 3-46
3.1.2.3. Recent Epidemiologic Studies Examining the PM2.s-Cardiovascular Effects
Relationship through Accountability Analyses and Alternative Methods for
Confounder Control 3-51
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 Cardiovascular Effects 3-54
3.2. Mortality 3-55
3.2.1. Short-Term PM2.5 Exposure 3-55
3.2.1.1. Summary and Causality Determination from 2019 Integrated Science Assessment
for Particulate Matter 3-56
3.2.1.2. Recent U.S. and Canadian Epidemiologic Studies 3-61
3.2.1.2.1. All-Cause and Total (Nonaccidental) Mortality 3-61
3.2.1.2.2. Cause-specific Mortality 3-64
3.2.1.2.3. Potential Copollutant Confounding of the PM2.s-Mortality Relationship 3-65
3.2.1.2.4. Effect Modification of the PM2.s-Mortality Relationship 3-66
3.2.1.2.5. Lag Structure of Associations 3-70
3.2.1.2.6. Examination of the Concentration-Response (C-R) Relationship between
Short-Term PM2.5 Exposure and Mortality 3-70
3.2.1.3. Recent Epidemiologic Studies Examining the PM2.s-Mortality Relationship through
Accountability Analyses and Alternative Methods for Confounder Control 3-73
May 2022
iv
-------
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 3-75
3.2.2. Long-Term PM2.5 Exposure 3-76
3.2.2.1. Summary and Causality Determination from 2019 Integrated Science Assessment
for Particulate Matter 3-77
3.2.2.2. Recent U.S. and Canadian Cohort Studies 3-82
3.2.2.2.1. All-Cause and Total (Nonaccidental) Mortality 3-83
3.2.2.2.2. Cause-Specific Mortality 3-91
3.2.2.2.3. Long-Term PM2.5 Exposure and Mortality in Populations with Preexisting
Conditions 3-98
3.2.2.2.4. Studies of Life Expectancy 3-99
3.2.2.2.5. Potential Copollutant Confounding of the PM2 5-Mortality Relationship _ 3-100
3.2.2.2.6. Studies that Address the Potential Implications of Unmeasured
Confounders on PM2 5-Mortality Associations 3-102
3.2.2.2.7. Examination of the Concentration-Response (C-R) Relationship between
Long-Term PM2.5 Exposure and Mortality 3-106
3.2.2.3. Recent Epidemiologic Studies Examining the PM2.s-Mortality Relationship through
Accountability Analyses and Alternative Methods for Confounder Control 3-112
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 3-127
3.3. Key Scientific Topics that Further Inform the Health Effects of PM2.5 3-128
3.3.1. Recent Experimental Studies at Near-Ambient Concentrations 3-129
3.3.2. PM2.5 Exposure and COVID-19 Infection and Death 3-130
3.3.2.1. Short-Term PM2.5 Exposure 3-131
3.3.2.2. Long-Term PM2.5 Exposure 3-131
3.3.2.3. Summary of Recent Epidemiologic Studies Examining PM2.5 Exposure and
COVID-19 Infection and Death 3-133
3.3.3. Populations and Lifestages at Potentially Increased Risk of a PM-Related Health Effect_ 3-134
3.3.3.1. Socioeconomic Status 3-135
3.3.3.1.1. Exposure Disparity 3-135
3.3.3.1.2. Health Risk Disparity 3-140
3.3.3.2. Race/Ethnicity 3-147
3.3.3.2.1. Exposure Disparity 3-148
3.3.3.2.2. Health Risk Disparity 3-153
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 3-160
4. 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 Estimates of Light Extinction Trends 4-6
4.2.3. Recent Advancements in Visibility Monitoring and Assessment 4-9
4.3. Summary of Recent Evidence in the Context of the 2019 Integrated Science Assessment for
Particulate Matter Causality Determination for Visibility Effects 4-11
5. SUMMARY AND CONCLUSIONS 5-1
APPENDIX A A-1
A.1 Advances in Estimating Light Extinction A-67
A.2 Quality Assurance Summary A-71
REFERENCES R-1
May 2022
v
-------
LIST OF TABLES
Table 2-1 Causal and likely to be causal causality determinations for short- and long-term PM2.5
exposure.
2-3
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. 2-16
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. 2-40
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. 3-4
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. 3-31
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.
3-46
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.
3-57
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.
3-79
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.
3-107
Table 3-7 Description of methods from epidemiologic studies using accountability analyses or
alternative methods for confounder control to examine long-term exposure to PM2.5 and
mortality. 3-113
May 2022
VI
-------
LIST OF FIGURES
Figure 3-1 Results of studies of short-term PM2.5 exposure and hospital admissions and emergency
department visits for ischemic heart disease. 3-12
Figure 3-2 Results of studies of short-term PM2.5 exposure and hospital admissions and emergency
department visits for stroke. 3-15
Figure 3-3 Results of studies of short-term PM2.5 exposure and hospital admissions and emergency
department visits for heart failure. 3-18
Figure 3-4 Results of studies of short-term PM2.5 exposure and hospital admissions and emergency
department visits for arrhythmia. 3-21
Figure 3-5 Associations between long-term PM2.5 exposure and ischemic heart disease or myocardial
infarction. 3-38
Figure 3-6 Associations between long-term PM2.5 exposure and the incidence of stroke. 3-40
Figure 3-7 Associations between long-term exposure to PM2.5 and cardiovascular morbidity in single
pollutant models and models adjusted for copollutants. 3-45
Figure 3-8 Concentration-response relationship for the association of PM2.5 concentration with acute
myocardial infarction. 3-49
Figure 3-9 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. 3-49
Figure 3-10 Concentration-response relationship for the association of PM2.5 concentration with first
admissions for myocardial infarction. 3-50
Figure 3-11 Predicted log hazard for incident nonfatal myocardial infarction versus previous 1-year
mean ambient PM2.5 concentration. 3-50
Figure 3-12 Analysis steps used by Danesh Yazdi et al. (2021) to examine long-term PM2.5 exposure
and cardiovascular-related hospital admissions. 3-52
Figure 3-13 Summary of associations between short-term PM2.5 exposure and total (nonaccidental)
mortality in multicity studies. 3-62
Figure 3-14 Summary of associations between short-term PM2.5 exposure and cardiovascular and
respiratory mortality in multicity studies. 3-65
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. 3-68
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). 3-69
Figure 3-17 Concentration-response curves for the United States (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. 3-71
May 2022 vii
-------
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. 3-72
Figure 3-19 Associations between long-term PM2.5 exposure and total (nonaccidental) mortality in
recent North American cohorts. 3-84
Figure 3-20 Associations between long-term PM2.5 exposure and all-cause and total (nonaccidental)
mortality in cohort studies in the United States and Canada published since the 2019
Integrated Science Assessment for Particulate Matter. 3-85
Figure 3-21 Hazard ratios for spatially decomposed analyses for an interquartile range increase in
PM2.5 concentrations for all-cause mortality. 3-87
Figure 3-22 Associations between long-term PM2.5 exposure and all cardiovascular disease and all
respiratory disease mortality in recent North American cohorts. 3-92
Figure 3-23 Associations between long-term PM2.5 exposure and cause-specific cardiovascular
mortality in recent North American cohorts. 3-95
Figure 3-24 Associations between long-term PM2.5 exposure and cause-specific respiratory mortality in
recent North American cohorts. 3-97
Figure 3-25 Estimated loss in life expectancy by county for females and males for PM2.5
concentrations in excess of the lowest observed PM2.5 concentration of 2.8 |jg/m3. 3-100
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). 3-104
Figure 3-27 Shape Constrained Health Impact Function predictions by PM2.5 concentration for the
pooled CanCHEC cohort. 3-110
Figure 3-28 Estimated concentration-response associations between PM2.5 and all-cause mortality
with a flexible modeling approach within the NHIS cohort. 3-111
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). 3-132
Figure 3-30a Differences in PM2.5 exposure by socioeconomic status. 3-138
Figure 3-30b Differences in PM2.5 exposure by socioeconomic status (continued). 3-139
Figure 3-31 Risk ratio of association between PM2.5 and mortality, stratified by socioeconomic status. 3-142
Figure 3-32 Hazard ratios for the association between PM2.5 and mortality, by greenspace and
community-level material deprivation. 3-143
Figure 3-33 County-level life expectancy losses due to PM2.5 exceeding 2.8 |jg/m3. 3-144
Figure 3-34 Estimated national and socioeconomic deprivation quintile-specific mortality rates. 3-145
Figure 3-35 Odds ratios for the association between PM2.5 and cardiovascular outcomes and diabetes
by neighborhood cluster. 3-147
Figure 3-36 Racial and ethnic inequities in PM2.5 exposure caused by population-adjusted group
consumption ("caused") and total personal consumption ("exposed"). 3-150
Figure 3-37 Source contributions to racial-ethnic disparity in PM2.5 exposure. 3-151
May 2022
viii
-------
Figure 3-38
Figure 3-39
Figure 3-40
Figure 3-41
Figure 3-42
Figure 4-1
Figure 4-2
Figure 4-3
Difference in PM2.5 exposure by race. 3-152
Percent change in cardiovascular disease mortality by PM2.5 exposure, stratified by
census block group racial composition in Massachusetts (2001-2011). 3-155
Percent change in cardiovascular disease mortality by PM2.5 exposure, stratified by the
Racial Residential Segregation (RSS) metric and Index of Racial Dissimilarity in
Massachusetts (2001-2011). 3-156
Relationship between life expectancy and PM2.5 exposure by income inequality and
percent Black. 3-158
Probability of cardiometabolic disease and PM2.5 exposure, stratified by race, gender, and
hypertension status. 3-160
Percent acceptability levels plotted against light extinction (bext) 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
Percent acceptability levels plotted against apparent contrast of distant landscape features
for each of the images used in studies for Phoenix, AZ (PHX), Chilliwack, BC (CHIL),
Abbotsford, BC (ABBT), Denver, CO (DEN), and the Grand Canyon, AZ (GRCA). 4-6
Annual mean reconstructed light extinction (bext) for the a) East, b) Intermountain
West/Southwest, c) West Coast by species, including ammonium sulfate (AS), ammonium
nitrate (AN), particulate organic matter (POM), fine dust (FD), coarse mass (CM), and sea
salt (SS). Map insets show individual sites aggregated into regional means. 4-8
May 2022
ix
-------
INTEGRATED SCIENCE ASSESSMENT TEAM FOR
PARTICULATE MATTER
Executive Direction
Dr. Steven J. Dutton (Division Director)—Center for Public Health and Environmental
Assessment, Office of Research and Development, U.S. Environmental Protection
Agency, Research Triangle Park, NC
Dr. Emily Snyder (Associate Division Director)—Center for Public Health and
Environmental Assessment, 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
May 2022
x
-------
Dr. Michael Stewart—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
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
Management, Technical, and Quality Assurance (QA) Support Staff
Ms. Christine Alvarez—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
Ms. Andrea Bartolotti—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
Ms. Marieka Boyd—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
Ms. Shannon Cassel—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
Ms. Cheryl Itkin—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Washington, DC
Ms. Maureen Johnson—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Washington, DC
Mr. Ryan Jones—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
Ms. Emma Leath—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. 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
May 2022
xi
-------
Ms. Jenna Strawbridge—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. Wayne Cascio—Center for Public Health and Environmental Assessment, Office of
Research and Development, U.S. Environmental Protection Agency, Research Triangle
Park, NC
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
Dr. Samantha Jones—Center for Public Health and Environmental Assessment, Office of
Research and Development, 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
Mr. Timothy Watkins—Center for Environmental Measurement and Modeling, 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
May 2022
xii
-------
ACRONYMS AND ABBREVIATIONS
Acronym/
Abbreviation
Meaning
AARP
American Association of Retired
Persons
ABI
ankle-brachial index
ACS
American Cancer Society
adj
adjustment
AF
atrial fibrillation
Ag Health
Agricultural Health Study
AHSMOG
Adventist Health Study and
Smog
AIC
Akaike information criterion
AL
Alabama
AMI
acute myocardial infarction
AN
ammonium nitrate
AOD
aerosol optical depth
APPROACH
Alberta Provincial Project for
Outcome Assessment in
Coronary Heart Disease
AQCD
Air Quality Criteria Document
AQS
Air Quality System
AS
ammonium sulfate
ASD
autism spectrum disorder
avg
average
BAD
bronchial artery diameter
BASIC
Brain Attack Surveillance in
Corpus Christi
BC
black carbon
bext
light extinction coefficient
BME
Bayesian maximum entropy
BMI
body mass index
BP
blood pressure
BRFSS
Behavioral Risk Factor
Surveillance System
CA
California
CAC
coronary artery calcification
CAD
coronary artery disease
Cancer Prev
cancer prevention
Acronym/
Abbreviation Meaning
CanCHEC
Canadian Census Health and
Environment Cohort
CAN-Marg
Canadian Marginalization Index
CAPs
concentrated ambient particles
CASAC
Clean Air Scientific Advisory
Committee
CATHGEN
Catheterization Genetics study
CBSA
core-based statistical area
CBVD
cerebrovascular disease
CCHEC
Canadian Census Health and
Environment Cohort
CCHS
Canadian Community Health
Survey
CFR
case fatality rate
CHD
coronary heart disease
CHF
congestive heart failure
CI
confidence interval
cIMT
carotid intima-media thickness
CM
coarse mass
CMA
census metropolitan area size
CMAQ
Community Multiscale Air
Quality model
CMR
cardiovascular mortality rate
CO
carbon monoxide
COPD
chronic obstructive pulmonary
disease
COVID-19
coronavirus disease 2019
C-R
concentration-response
CSN
Chemical Speciation Network
CTM
chemical transport model
CTS
California Teachers Study
C-V
cross-validation
CVD
cardiovascular disease
DBP
diastolic blood pressure
DC
District of Columbia
DE
diesel exhaust
df
degrees of freedom
May 2022
xiii
-------
Acronym/
Abbreviation Meaning
DID
difference-in-difference
DNA
deoxyribonucleic acid
DOW
day of week
DRAM
doubly robust additive model
DVT
deep vein thrombosis
EC
elemental carbon
ECG
electrocardiogram
ED
emergency department
EFFECT
Enhanced Feedback for
Effective Cardiac Treatment
EJ
environmental justice
EPA
Environmental Protection
Agency
ESCAPE
European Study of Cohorts for
Air Pollution Effects
ESRD
end-stage renal disease
exp
exposure
FA
filtered air; fatty acid
FD
find dust
FEVi
forced expiratory volume in 1
second
FMD
flow-mediated dilation
FVC
forced vital capacity
GA
Georgia
GAM
generalized additive model
GLM
generalized linear model
GEOS-Chem
Goddard Earth Observing
System-Chem
GP
general practitioner
GPS
generalized propensity score
GWR
geographically weighted
regression
h
hour(s)
HA
hospital admission
HDL-c
Fligh-density lipoprotein
cholesterol
Health Prof
health professionals
HeartSCORE
Fleart Strategies Concentration
on Risk Evaluation
HEI
Flealth Effects Institute
Acronym/
Abbreviation
Meaning
HEPA
high efficiency particle filter
HF
heart failure; high frequency
HHD
hypertensive heart disease
HISA
Highly Influential Scientific
Assessment
HNR
Heinz Nixdorf Recall (study)
HPFU
Health Professionals Foliow-Up
Study
HR
hazard ratio
HRV
heart rate variability
HS
hemorrhagic stroke
HSC
Harvard Six Cities
HYSLPLIT
HYbrid Single-Particle
Lagrangian Integrated
Trajectory
IAD
inter-adventitial diameter
ICD-10
International Classification of
Disease version 10
ICD-9
International Classification of
Disease version 9
ICU
intensive care unit
IDW
inverse distance weighting
IHD
ischemic heart disease
IL
Illinois
IMPROVE
Interagency Monitoring of
Protected Visual Environments
InMAP
Intervention Model for Air
Pollution
IPTW
inverse probability of treatment
weighting
IPW
inverse probability weighting
IQR
interquartile range
IRD
Index of Racial Dissimilarity
IRP
Integrated Review Plan
IRR
incidence rate ratio
IS
ischemic stroke
IV
instrumental variable
JHS
Jackson Heart Study
km
kilometer(s)
km2
square kilometer(s)
LDH
lactate dehydrogenase
May 2022
xiv
-------
Acronym/
Abbreviation
Meaning
Acronym/
Abbreviation
Meaning
LDL-c low-density lipoprotein
cholesterol
LF low frequency
LUR land use regression
LUR-BME land use regression—Bayesian
maximum entropy
LV left ventricular
m2 Square meter(s)
MAP mean arterial pressure
MAPLE mortality-air pollution
associations in low-exposure
environments
max maximum
MCAPS Medicare Cohort Air Pollution
Study
MCC Multi-City Multi-Country
Collaborative Research Network
mCCHS Canadian Community Health
Survey—mortality cohort
MCM multi-cause multicity
MD Maryland
MESA Multi-Ethnic Study of
Atherosclerosis
mg milligram(s)
MI myocardial infarction
min minimum
MINAP Myocardial Ischemia National
Audit Project
MISR Multiangle Imaging
Spectroradiometer
MISS monotonically increasing
smoothing splines
mm Hg millimeters of mercury
mo month(s)
MO Missouri; month
MR mortality ratio
MRR mortality risk ratio
NAAQS National Ambient Air Quality
Standards
NAPS National Air Pollution
Surveillance System
NC number concentration; North
Carolina
NCHS National Center for Health
Statistics
NEI National Emissions Inventory
NH non-Hispanic
NHIS National Health Interview
Survey
NHS Nurses' Health Study
NIH National Institutes of Health
NIH-AARP National Institutes of Health-
American Association of Retired
Persons (diet and health cohort)
NMMAPS National Morbidity, Mortality,
and Air Pollution Study
NN Normal-to-Normal
NO2 nitrogen dioxide
NO3 nitrate
NOx oxides of nitrogen (NO + NO2)
NPMs neighborhood PM monitors
NR not reported
NSTEMI non-ST segment elevation MI
O3 ozone
OC organic carbon
OHCA out-of-hospital cardiac arrest
OLS ordinary least squares
OM organic matter
OMB Office of Management and
Budget
ONPHEC Ontario Population Health and
Environment Cohort
OR odds of recurrent
Ox Redox weighted average of NO2
and O2
PA Pennsylvania
PAH poly cyclic aromatic
hydrocarbon(s)
PE prediction error
PEF peak expiratory flow
PM particulate matter
May 2022
xv
-------
Acronym/
Acronym/
Abbreviation
Meaning
Abbreviation
Meaning
PMio
particulate matter with a
nominal mean aerodynamic
SARS-CoV-2
severe acute respiratory
syndrome coronavirus 2
diameter less than or equal to
10 um
systolic blood pressure
SBP
PMlO-2.5
particulate matter with a
SC
surface area concentration
nominal mean aerodynamic
SCHIF
Shape Constrained Health
diameter greater than 2.5 um
Impact Function
and less than or equal to 10 um
SD
standard deviation
PM2.5
particulate matter with a
nominal mean aerodynamic
SDI
social deprivation index
diameter less than or equal to
SDNN
standard deviation of NN
2.5 um
SE
standard error
POM
particulate organic matter
PPb
parts per billion
SED
socioeconomic deprivation
parts per million
SEIR
susceptible-exposed-infected-rec
ppm
overed
PQAPP
Program-level Quality
Assurance Project Plan
SES
SHS
socioeconomic status
second-hand smoke
PREMIER
Prospective Registry Evaluating
Myocardial Infraction: Events
SHV
Social and Health Vulnerability
and Recovery
sICAM
soluble intercellular adhesion
QA
quality assurance
molecule 1
QAPP
quality assurance project plans
SO2
sulfur dioxide
QRS
time interval between the
SO4
sulfate
beginning of the Q wave and the
peak of the S wave
SPARCS
New York State Department of
Health Statewide Planning and
r
correlation coefficient
Research Cooperative System
R2
coefficient of determination
SPE
standardized prediction error
RAMP
Real-time Affordable
Multi-Pollutant
ST
beginning of S wave to end of T
wave
RC
regression calibration
STEMI
ST elevated myocardial
infarction
RCS
restricted cubic splines
sVCAM
Soluble vascular cell adhesion
redox
reduction-oxidation
molecule 1
REGARDS
REasons for Geographic and
SHV
Social Health Vulnerability
Study of Women's Health
Racial Differences in Stroke
SWAN
re-HA
Readmission to the hospital
Across the Nation
RF
radiative forcing
TRAP
traffic-related air pollution
RH
relative humidity
TrIPS
Trucking Industry Particle Study
RMSS
root mean square standardized
TRIUMPH
Translational Research
RR
relative risks
Investigating Underlying
Disparities in Acute Myocardial
RRS
racial residential segregation
Infarction
RV
right ventricular
TX
Texas
SARS
severe acute respiratory
syndrome
May 2022
xvi
-------
Acronym/
Abbreviation Meaning
UFIREG Ultrafine Particles—An
Evidence-Based Contribution to
the Development of Regional
and European Environmental
and Health Policy
UFP ultrafine particle
U.S. United States of America
U.S. EPA U.S. Environmental Protection
Agency
USRDS
U.S. Renal Data System
W
west
WHI
Women's Health Initiative
WHO
World Health Organization
WS Fe
water-soluble iron
yr
year(s)
ZIP
Zone Improvement Plan
May 2022
xvii
-------
EXECUTIVE SUMMARY
In June 2021, the U.S. Environmental Protection Agency (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, EPA indicated that it would develop a supplement to
the 2019 Integrated Science Assessment for PM (2019 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, EPA presents an evaluation of recent studies (i.e., published since the
literature cutoff date of the 2019 PM ISA) that potentially are of 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 consist of U.S. and Canadian studies, specifically: (a) epidemiologic studies for health
effect categories for which the 2019 PM ISA concluded a causal relationship (i.e., short- and long-term
PM2 5 exposure1 and cardiovascular effects and mortality); (b) epidemiologic studies that employed
statistical approaches that attempt to more extensively account for confounders and are more robust to
model misspecification (i.e., used alternative methods for confounder control)2 or conducted
accountability analyses; (c) studies that address key scientific topics that have evolved since the literature
cutoff date for the 2019 PM ISA, including experimental studies conducted at near-ambient PM2 5
concentrations, epidemiologic studies that examined the association between PM2 5 exposure and severe
acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection and coronavirus disease 2019
(COVID-19) death, and epidemiologic or exposure studies that examined disparities in PM2 5 exposure or
health risks by race and ethnicity or socioeconomic status; and (d) 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 does not represent a full multidisciplinary
evaluation of evidence that results in the formation of weight-of-evidence conclusions (i.e., causality
determinations), but instead puts the results of recent studies that encompass specific criteria in the
context of the scientific conclusions presented within the 2019 PM ISA. As such, the Supplement
indicates whether recent evidence supports (is consistent with), supports and extends (is consistent with
1 Consistent with the scope of the 2019 PM ISA (SectionP.3.1), short-term exposures are defined as those exposures
occurring over hours up to 1 month, while long-term exposures are defined as those exposures occurring over 1
month to years.
2 In the peer-reviewed literature, these epidemiologic studies are often referred to as causal inference studies or
studies that used causal modeling methods. For the purposes of this Supplement this terminology is not used to
prevent confusion with the main scientific conclusions (i.e., the causality determinations) presented within an ISA.
In addition, as is consistent with the weight-of-evidence framework used within ISAs and discussed in the Preamble
to the Integrated Science Assessments, an individual study on its own cannot provide the evidence needed to make a
causality determination, but instead represents a piece of the overall body of evidence.
May 2022
ES-i
-------
and reduces uncertainties), or does not support (is not consistent with) the causality determinations
detailed in the 2019 PM ISA for the health effects categories evaluated within this Supplement.
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 further supports, and in some
instances extends, the evidence that contributed to the conclusion of a causal relationship detailed
in 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 micrograms per cubic meter (|ig/m3); mean 24-hour avg PM2 5
concentrations ranging from 7.1 to 15.4 (.ig/ni').
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, strong evidence remains
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
with cause-specific mortality outcomes.
o A number of recent long-term PM2 5 exposure and mortality studies conducted in cohorts
consisting of populations with diverse demographic characteristics and encompassing
large geographic areas report consistent, positive associations, with most reporting mean
annual PM25 concentrations ranging from 5.9 to 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 statistical
approaches that attempt to account more extensively for confounders and are
May 2022
ES-ii
-------
more robust to model misspecification (i.e., used alternative methods for
confounder control) 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.
o In response to the global COVID-19 pandemic, numerous studies provide initial
assessments of short- and long-term PM2 5 exposure and SARS-CoV-2 infection and
COVID-19 death. While some of these studies report initial evidence of 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 PIVbs-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 among
minority populations, specifically Black populations. Additionally recent evidence
supports the evidence presented in the 2019 PM ISA that there may be PM2 5 exposure
and health risk disparities by socioeconomic status (SES), specifically among people of
low SES.
• Recent studies continue to support a relationship between PM and visibility impairment and
provide additional insights on the impact of choice of metric on preference study results, impacts
of changing PM composition on the relationship between PM and visibility impairment, and
alternative approaches to estimating light extinction.
May 2022
ES-iii
-------
1.INTRODUCTION AND SCOPE
1.1. Introduction
The U.S. Environmental Protection Agency (EPA) completed the Integrated Science Assessment
for Particulate Matter (PM ISA) in December 2019 (hereafter referred to as the 2019 PM ISA) (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 short- and long-term3 PM exposure and health and PM 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.4
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 and were considered in EPA's final decision in the 2020
review to retain the PM NAAQS (see Section 1.3.5, (U.S. EPA. 2022)).
On June 10, 2021, EPA announced it is reconsidering the December 2020 decision to retain the
PM NAAQS "because available scientific evidence and technical information indicate that the current
standards may not be adequate to protect public health and welfare, as required by the Clean Air Act.
EPA explained that as part of the reconsideration process " the agency will develop a supplement to the
2019 [PM ISA] that will take into account the most up-to-date science" (EPA Press Office. 2021).5 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, provide the scientific basis to support a
robust and thorough reconsideration of the 2020 PM NAAQS.
3 Consistent with the scope of the 2019 PM ISA (SectionP.3.1), short-term exposures are defined as those exposures
occurring over hours up to 1 month, whereas long-term exposures are defined as those exposures occurring over 1
month to years.
4Hereafter 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).
5 See Section 1.4 of the Policy Assessment for the Reconsideration of the National Ambient Air Quality Standards
for Particulate Matter for additional details (U.S. EPA. 2022).
May 2022
1-1
-------
1.2. Rationale and Scope
In completing the review of the PM NAAQS in December 2020, 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, EPA
"concluded that none of the studies materially change any of the broad scientific conclusions of 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, 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. Additionally, the evaluation of recent
studies identified would occur in the form of a supplement and EPA would rely on the Supplement to the
2019 PM ISA and the 2019 PM ISA as the scientific foundation for the reconsideration, rather than
revising the 2019 PM ISA or developing anew PM ISA. To facilitate the identification and evaluation of
recent studies that warrant review, the 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 EPA in support of the reconsideration of the primary and secondary PM NAAQS. This
targeted approach to developing the Supplement to the 2019 PM ISA for the purpose of reconsidering the
2020 PM NAAQS decision does not reflect a change to EPA's approach for developing ISAs for NAAQS
reviews.
1.2.1. Rational for Inclusion of Health and Welfare Effects
The causality determinations presented within the 2019 PM ISA (discussed in Section 2). 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 this Supplement.
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, [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
May 2022
1-2
-------
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 for which 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 for
which the 2019 PM ISA concluded a causal relationship.
In addition, this Supplement also considers recent health effects evidence that addresses key
scientific topics for which 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
statistical approaches that attempt to more extensively account for confounders and are more robust to
model misspecification (i.e., used alternative methods for confounder control)6 or conducted
accountability analyses, studies that assess the relationship between PM2 5 exposure and severe acute
respiratory syndrome coronavirus 2 (SARS-CoV-2) infection and coronavirus disease 2019 (COVID-19)
death; and in accordance with recent EPA goals 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 by race/ethnicity
and socioeconomic status (SES).
In identifying the studies to consider for inclusion within this Supplement, the focus was on those
studies conducted in locations that were most informative to the reconsideration of the 2020 PM NAAQS.
This criterion resulted in an assessment of the scientific literature that is more refined compared with the
2019 PM ISA. While the 2019 PM ISA considered and included studies conducted globally when
evaluating the evidence and forming causality determinations, the rationale for the scope of this
Supplement is directly informed by policy considerations surrounding the types of scientific information
included in the 2020 PM PA. In addition to focusing on studies for health effect categories for which the
2019 PM ISA concluded causal or a likely to be causal relationship, as noted above, the 2020 PM PA
also focused on a narrower set of studies conducted in locations that are most relevant to informing the
level, form, averaging time, and indicator of the NAAQS for PM. Specifically, the 2020 PM PA states
that the emphasis is on "multicity studies that examine health effect associations in the U.S. or Canada, as
such studies examine potential associations over large geographic areas with diverse atmospheric
conditions and population demographics (e.g., U.S. EPA (2019). Sections 11.1 and 11.2). Additionally,
studies examining associations outside the U.S. or Canada reflect air quality and exposure patterns that
may be less typical of the U.S., and thus less likely to be informative for purposes of reviewing the
6 In the peer-reviewed literature, these epidemiologic studies are often referred to as causal inference studies or
studies that used causal modeling methods. For the purposes of this Supplement, this terminology is not used to
prevent confusion with the main scientific conclusions (i.e., the causality determinations) presented within an ISA.
In addition, as is consistent with the weight-of-evidence framework used within ISAs and discussed in the Preamble
to the Integrated Science Assessments, an individual study on its own cannot inform causality, but instead represents
a piece of the overall body of evidence.
May 2022
1-3
-------
NAAQS" (U.S. EPA. 2020b).7 Therefore, within this Supplement the studies considered for inclusion are
limited to those studies conducted in the U.S. and Canada. However, it is the combination of the scientific
evidence detailed in the 2019 PM ISA and this Supplement that forms the complete scientific record
informing the reconsideration of the 2020 PM NAAQS.
Consistent with the rationale for the health effects, the selection of welfare effects to evaluate
within this Supplement is 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, and
consistent with the health effects rationale, is limited to studies conducted in the U.S. and Canada.
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 inclusion within the Supplement. Specifically,
studies must be peer reviewed and published between approximately January 2018 and March 2021, and
satisfy the following criteria:
Health Effects
• 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)
o U.S. and Canadian epidemiologic studies that employed alternative methods for
confounder control or conducted accountability analyses (i.e., examined the effect of a
policy on reducing PM2 5 concentrations)8
Key Scientific Topics
• Experimental studies (i.e., controlled human exposure and animal toxicological) conducted at
near-ambient PM2 5 concentrations experienced in the U.S.
7 This emphasis on studies conducted in the U.S. or Canada is consistent with the approach in previous reviews of
the PM NAAQS (U.S. EPA (2011). section 2.1.3).
8These 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 filters (HEPA) or indoor air cleaners.
May 2022
1-4
-------
• U.S.- and Canadian-based epidemiologic studies that examined the relationship between PM2 5
exposures and SARS-CoV-2 infection and COVID-19 death
• At-risk populations
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 SES
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 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). Additionally, this scope does not allow for the evaluation of
recent studies for health effect categories from the 2019 PM ISA for which a likely to be causal
relationship was concluded nor an assessment as to whether recent evidence may strengthen the causality
determination to a causal relationship 9 Therefore, this Supplement critically evaluates and provides key
study-specific information for only 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 described in the 2019 PM ISA.
1.3. Development of the Supplement
The process used in developing this Supplement is consistent with the 2019 PM ISA as captured
in the Preface and Appendix of the 2019 PM ISA. Because this Supplement builds on the 2019 PM ISA,
that process is not reiterated but instead is cross referenced. Within the 2019 PM ISA, the Preface
provides a detailed description of the process for developing ISAs (Section P.3.), including a discussion
of the scope of the ISA (Section P.3.1.) and how evidence is evaluated (Section P.3.2.). A more detailed
description of the process of evaluating evidence in ISAs is described in the Preamble to the Integrated
Science Assessments (U.S. EPA. 2015) with information specific to the PM ISA in the Appendix of the
2019 PM ISA. Specifically, the Appendix describes in detail the various steps that encompassed the
development of the PM ISA. These steps include the literature search and the evaluation of individual
study quality, which details scientific considerations for evaluating the strength of inference from studies
9 The narrow scope also does not allow for the evaluation of recent studies for health effect categories from the 2019
PM ISA where suggestive of, but not sufficient to infer, a causal relationship and inadequate to infer the presence or
absence of a causal relationship was concluded.
May 2022
1-5
-------
that examined the health effects of PM (2019 PM ISA, Section A.3.2., Table A-l). The information
presented in Table A-l in the 2019 PM ISA, which includes the identification and rationale behind
advantageous study characteristics (e.g., study design, study population, exposure assessment) as well as
information on the selection of results to present from individual studies, was relied upon in the process
of considering and identifying recent studies evaluated within this Supplement.
1.4. Organization of the Supplement
The Supplement to the 2019 PM ISA is not intended to be a stand-alone 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 that 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. The
organization of Section 3 is consistent with the overall organization of the health effects discussion in the
2019 PM ISA, which includes separate discussions of the evidence organized in relation to exposure
duration (i.e., short- or long-term exposure) and then within the exposure duration sections discussions
organized around specific health effects (e.g., myocardial infarction, nonaccidental mortality) and specific
issues of importance (e.g., copollutant confounding, concentration-response relationship). In addition,
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, Section 3 evaluates recent studies that assess key science topics that have evolved
since the completion of the 2019 PM ISA. Study-specific details for the epidemiologic studies evaluated
in Section 3, such as information on study population, exposure assessment, PM2 5 concentrations, and
confounder control (e.g., copollutants) are detailed in tables presented in Section 3. Section 4 consists of
an evaluation of recent studies that inform visibility effects and is organized similar to the health effects
chapter. Therefore, Section 4 first presents the summary and causality determination from the 2019 PM
ISA, then evaluates recent studies, and concludes by assessing new evidence in the context of the
conclusions for visibility impairment presented in the 2019 PM ISA. Finally, Section 5 provides a
summary and presents overarching conclusions based on the evaluation of recent studies within this
Supplement.
May 2022
1-6
-------
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 !s idenee spaiiuum scientific disciplines (i e.. atmospheric chcniistrs. exposure science, dosimetry.
cpideniiolous. controlled human exposure. ;ind animal toxicolous I hmll upon es idcncc detailed in the
2<>()1) |»\| IS \ ;md re; i I'll lined ;i c ,ius,ilrel,iii"iislii/< between short- ;md loim-terni I'M exposure ;nid
cardios ascular effects ;ind total (uouaccideiilal) niorlahts. ;ind ;i likely i" he i ,ius,il relationship for
ivspir;iun\ effects
• I Aperinieiilal ;ind epidemiologic es idenee supported ;i UL Ir i" he c ausul rcl.iii<>iiship between
loim-tcrni I'M exposure ;nid uers oils ss stem ell eels
• I !s idcncc. priiiKiri l\ from studies of hum e;nieer ineidenee ;ind niorlahts . in combination w illi I lie
dee;ides ii|" research on llie niulaueuicils ;ind carcuioueuicils of I'M supported ;i lilelv1<< he > ,ius,il
relationship between loim-terni I'M exposure ;ind cancer
• Rcniaiuum uncertainties ;ind limitations in llie seiciiiifie es idenee contributed to ;i surest ive ni. hiu nut
sufficient to infer, a causal relationship and inadequate to infer the presence or absence of a causal
relationship fiir all other exposure. size fraction. and lieallli effcels cateuors combinations
• I !s idenee hmll upon and reaffirmed dial there is a c .utsal relationship helween I'M and llie
uoiiecolomcal welfare elleels \ isibilits inipairnieni. eliniale elTeels. and nialerials effects
• l lie assessment of I'M sources and components confirmed and continued to support the conclusion
from the 2009 PM ISA:lany PM; s components and sources are associated with many health effects,
and the evidence does not indicate thai any one source or component is more strongly related with
health effects than PM; 5 mass.
• Mau\ populations ie.u . healths, diseased) and lilestaues ie.u . children, older adults) ha\e heeu shown
to he at risk of a health effect mi response to short- or loim-ierni I'M exposure. particularls I'M
I lowe\ er. of the populations and hfestaues examined. scientific e\ idenee indicated that ouls some
populations mas he at disfrofortion.iieh- in< re.isi d risk of a I'M -related health effect, uicludiim
nniiorits populations (often defined as non-White populations within uidis idual studies), children,
people w illi specific ueuetic \ ariauts mi ueues mi the uluialhioue transferase pathwas. people w ho are
o\erwemht orohese. peoplewith preexistum cardios ascular and respirators diseases, people of low
socioeconomic status i SI !Si. and people w ho smoke or were former smokers Inadequate e\ idenee
exists to deterniiue w hclher has iiiu diabetes, heiuu iu au older hfestaue (i.e . older adults), residential
location (iiicludiim proxumis to source and urban residence). se\. or diet increase the risk of
I'M -related health effects
2.1. Health Effects
The 2019 Integrated Science Assessment for Particulate Matter (2019 PM ISA) evaluated
relationships between short-term and long-term exposures to PM (i.e., PM2 5, PM10-2.5, and UFPs) and an
array of health effects described in epidemiologic, controlled human exposure, and animal toxicological
studies. In the assessment of the overall evidence, the strengths and limitations of individual studies were
evaluated based on scientific considerations detailed in the Appendix to the 2019 PM ISA. Short-term
exposures are defined as those with durations of hours up to 1 month, with most studies examining effects
related to exposures in the range of 24 hours to 1 week. Long-term exposures are defined as those with
May 2022
2-1
-------
durations of more than 1 month to years. As detailed in the 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
PM10 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 for which 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 for which 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 for which the 2019 PM ISA 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.
10Composite measures of PM may include mass, volume, surface area, or number concentration.
May 2022
2-2
-------
Table 2-1 Causal and likely 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
May 2022
2-3
-------
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.
May 2022
2-4
-------
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
May 2022
2-5
-------
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
May 2022
2-6
-------
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 a key 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 concentrated
ambient particles (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 PM2 5
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 cardiac contractility and changes in left ventricular
May 2022
2-7
-------
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, heart rate variability (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 PM25
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 micrograms per cubic meter (|ig/m3) (2019 PM ISA, Section 6.2.10).
May 2022
2-8
-------
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
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) 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 [NHS], California Teachers Study [CTS]) 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 cardiovascular disease (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 PM25 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,
May 2022
2-9
-------
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 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 affect 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
May 2022
2-10
-------
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.
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.
May 2022
2-11
-------
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
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 relationship11 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 polycyclic aromatic hydrocarbons are
known human carcinogens. Extensive analyses of PM2 5 and PM2 5 extracts in the Ames
,S'fl/«?(-w7t'//fl/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,
"Since 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.
May 2022
2-12
-------
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.
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 in 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% to 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,
May 2022
2-13
-------
Figure 11-1; Table 11-1).12 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 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); 2019 PM ISA, 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 in 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) provided further evidence of positive
associations between long-term PM2 5 exposure and total mortality, particularly in areas with annual mean
12Throughout 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.
May 2022
2-14
-------
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 provided by
the coherence of effects across scientific disciplines for cardiovascular morbidity, particularly for CHD,
stroke, and atherosclerosis, and for respiratory morbidity, particularly for the development of COPD.
Recent studies extensively examined the C-R relationship between long-term PM2 5 exposure and total
mortality, specifically in several U.S. and Canadian cohorts, and collectively continued to support a
linear, no-threshold C-R relationship (2019 PM ISA, Section 11.2.4; Table 11-7).
A series of studies evaluated in the 2019 PM ISA, examined the relationship between long-term
exposure to PM2 5 and mortality by examining the temporal trends in PM2 5 concentrations to test the
hypothesis that decreases in PM2 5 concentrations are associated with increases in life expectancy (2019
PM ISA, Section 11.2.2.5). These studies reported that decreases in long-term PM2 5 concentrations were
associated with an increase in life expectancy across the U.S. for the multiple time periods examined.
May 2022
2-15
-------
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
May 2022
2-16
-------
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
May 2022
2-17
-------
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
May 2022
2-18
-------
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)
May 2022
2-19
-------
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
May 2022
2-20
-------
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
May 2022
2-21
-------
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
May 2022
2-22
-------
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
May 2022
2-23
-------
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 Integrated Review Plan for the National Ambient Air Quality Standards for Particulate
Matter (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 exists
between PM exposure and various health outcomes (e.g., mortality, hospital admissions),
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 2.2.5V
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
May 2022
2-24
-------
(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.
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., r < 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 examined, followed by NO2 and PM2 5 Within these studies 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
May 2022
2-25
-------
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.
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 nor
coherently indicated a stronger or weaker effect of combined exposure to PM2 5 and another pollutant
compared with 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., NO2, NOx, and CO) and O3, with some studies also examining PM10-2.5. 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.
May 2022
2-26
-------
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 NO2, SO2, PM10-2.5, and benzene. Studies that examined
O3 reported correlations that were generally moderate (ranging from r = 0.49 to 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
May 2022
2-27
-------
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
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 to 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.
May 2022
2-28
-------
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
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; 2019 PM ISA, 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 a threshold concentration exists 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
May 2022
2-29
-------
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
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 PM10. 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.
May 2022
2-30
-------
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.
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), but the ones that 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 PM2 5 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
May 2022
2-31
-------
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) and PM components; (2) applied some approach
to assess the particle effect (e.g., particle trap) of a mixture; or (3) conducted formal statistical 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
May 2022
2-32
-------
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,
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
May 2022
2-33
-------
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.
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 PM2 5 component. Although
the respiratory and cardiovascular effects studies focused mainly on EC/BC, the studies of mortality did
not examine any one component disproportionately over 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). 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:
May 2022
2-34
-------
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
Preamble to the IS As (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).
The causality determinations briefly summarized within this section, which are more fully
detailed in the health effects chapters of the 2019 PM ISA, suggest that 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 versus 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
PM25 exposure. Specifically, epidemiologic studies evaluated in the 2019 PM ISA of long-term PM25
May 2022
2-35
-------
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 minority populations (often defined as non-White populations within
individual studies) 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 minority populations are at greater risk for both
PM2 5-related health effects and PM2 5 exposure than are Whites.
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 who are overweight or obese
(2019 PM ISA, Section 12.3.3), those with particular genetic variants (2019 PM ISA, Section 12.4), those
who 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
May 2022
2-36
-------
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) in assessing whether populations are at increased risk of a PM2 5-related health effect. In
studies examining both differential exposure and increased risk of health effects, there was some 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.
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
May 2022
2-37
-------
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 to 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 a major contributor to light extinction, 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
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.
May 2022
2-38
-------
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 America and Europe because this greenhouse-gas warming was unmasked (2019 PM
ISA, Section 13.3.6). In developing countries in Asia, by contrast, PM concentrations have increased 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).
May 2022
2-39
-------
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.
May 2022
2-40
-------
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 Integrated
Science Assessment for Particulate Matter (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 (i.e., hours up to 1 month) and long-term (i.e., over 1 month to years)
PM2 5 exposure and cardiovascular effects (Section 3.1) and mortality (Section 3.2).13 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 2020 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 for which 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 PM2 5 exposure on severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection and
coronavirus disease 2019 (COVID-19) 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 in 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). that is, U.S. and Canadian epidemiologic studies and other studies that address key
scientific topics. Therefore, the scientific information presented in 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
13Throughout 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.
May 2022 3-1
-------
reduces uncertainties), or does not support (is not consistent with) the causality determinations in the
2019 PM ISA.
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 in 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).14 In addition, with the expansion of epidemiologic studies that
used statistical approaches that attempt to more extensively account for confounders and are more robust
to model misspecification (i.e., used alternative methods for confounder control), 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. Finally, a summary of the results of
recent studies evaluated in the section is presented in the context of the scientific conclusions detailed in
the 2019 PM ISA (Section 3. l.b4). 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 addition to evaluating evidence across scientific disciplines that examined the relationship
between short-term PM2 5 exposure and cardiovascular effects, discussed below, the 2019 PM ISA
14 Throughout this section, 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, unless otherwise noted.
May 2022
3-2
-------
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 ED visits and hospital admissions for IHD and heart failure, and
ultimately mortality (2019 PM ISA, Figure 6-1).
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 support to the causality determination and to biological plausibility.
Finally, epidemiologic studies of cardiovascular-related mortality provided additional evidence that
demonstrated a continuum of effects from biomarkers of inflammation and coagulation, subclinical
endpoints (e.g., HRV, BP, endothelial dysfunction), ED visits and hospital admissions, and eventually
death. The evidence evaluated in the 2019 PM ISA also reduced uncertainties from the previous review
related to potential copollutant confounding and limited biological plausibility for cardiovascular effects
following short-term PM2 5 exposure. Evidence supporting the causality determination for short-term
PM2 5 exposure and cardiovascular effects reached in the 2019 PM ISA is discussed below and
summarized in Table 3-1. using the framework for causality determinations described in the Preamble to
the ISAs (U.S. EPA. 2015).
May 2022
3-3
-------
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
Increases in ED visits and hospital
Section 6.1
.2.1
5.8-18.6
epidemiologic evidence
from multiple studies at
relevant PM2.5
admissions for IHD and heart failure in
multicity studies conducted in the U.S.,
Canada, Europe, and Asia
Section 6.1
Section 6.1
.3.1
.9
5.8-18.0
concentrations
Increases in cardiovascular mortality in
multicity studies conducted in the U.S.,
Canada, Europe, and Asia.
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
Epidemiologic evidence
The magnitude of PM2.5 associations Section 6.1.14.1
from copollutant
remain positive, but in some cases are
models provides some
reduced with larger confidence intervals
support for an
in copollutant models with gaseous
independent PM2.5
pollutants. Further support from
association
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).
May 2022
3-4
-------
Rationale for Key References „ PM2.5 Concentrations
Causality and Sections in Associated with Effects'
Determination3 Key Evidence13 the 2019 PM ISAb (|jg/m3)
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
Klooq 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
Note: This table corresponds to Table 6-34 in the 2019 PM ISA.
AOD = aerosol optical depth; CMAQ = Community Multiscale Air Quality model; CVD = cardiovascular disease; 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. 2015).
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.
May 2022
3-5
-------
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
cardiovascular disease (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 acontrolled human exposure study and animal
toxicological study showing decreased cardiac function following short-term PM2 5 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 brachial
artery diameter (BAD) or flow-mediated dilatation (FMD). More specifically, and in contrast to the 2009
PM ISA for which 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 PM25 on measures of blood flow (2019 PM ISA, Section 6.1.13.2) relative to filtered air (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 versus
endothelial-dependent mechanisms). In addition to endothelial dysfunction, controlled human exposure
studies evaluated in the 2019 PM ISA that used CAPs, but not filtered diesel exhaust (DE), generally
reported evidence for small increases in blood pressure, although there were inconsistencies across studies
with respect to changes in systolic blood pressure (SBP) and diastolic blood pressure (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
May 2022
3-6
-------
BP usually changed as well, but the change was not found to be statistically significant (2019 PM ISA,
Section 6.1.6.3). That 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
May 2022
3-7
-------
PM2 5 exposure is associated with some of the same cardiovascular endpoints reported in controlled
human exposure and animal toxicological studies. The number of studies is limited that evaluate 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, the evidence continues to be sufficient 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
Recent epidemiologic studies conducted in the U.S. and Canada build on the strong
epidemiologic evidence base evaluated in the 2019 PM ISA, as well as in previous assessments, which
provided the scientific rationale supporting a causal relationship between short-term PM2 5 exposure and
cardiovascular effects (Section 3.1.1.IV 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
confounding (Section 3.1.1.2.7) and the lag structure of associations (Section 3.1.1.2.8). The following
sections present an evaluation of 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. Study-specific details (e.g., study population, exposure assessment
approach, confounders considered) for the epidemiologic studies evaluated in this section are presented in
Appendix A (Table A-l).
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 International
May 2022
3-8
-------
Classification of Diseases (ICD) codes recorded when a patient is admitted or discharged from the
hospital or ED (ICD-Ninth Revision [ICD-9]: 410-414 or ICD-Tenth Revision [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 Cohort Air Pollution Study (MCAPS) (Dominici et al.. 2006). a four-city study in
Australia (Barnctt 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 with 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
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
May 2022
3-9
-------
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 (RR) 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). No association between 0-2-day average PM2 5
concentration with hospital admissions for MI among the entire population was observed [OR: 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 the association strengthened in terms of magnitude and precision 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. (2017) performed a
case-crossover analysis to examine the relationship between short-term PM2 5 concentration and STEMI
in acute coronary syndrome or unstable angina patients (n = 362) in Monroe County, NY (2007-2012).
The association between previous 1-hour PM2 5 concentration and STEMI reported by these authors (OR:
1.25 [95% CI: 0.99, 1.59]) was virtually identical to the association (OR: 1.26 [95% CI: 1.01, 1.57])
reported in a previous analysis of this population conducted by Gardner et al. (2014) that reported fewer
patients (n = 338) and a shorter follow-up time (2007-2010).
Results of studies of IHD and MI 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 3-1.
May 2022
3-10
-------
Overall, recent studies support and extend the findings of the 2019 PM ISA with additional studies
reporting positive associations between short-term PM2 5 exposure and both IHD and MI hospital
admissions and ED visits.
May 2022
3-11
-------
Study
Dominici et al. (2006)
Bamett et al. (2006)
Hostetal. (2007)
tBell et al. (2015)
tKlooa et al. (2014)
Zanobetti et al (2006)
fBell et al. (2015)
tZanobetti et al. '(2009)
*DeSouza et al. (2019)
tHalev et al. (2O09)
*Krall et al. (201S)
*Hsu et al. (2017)
tTalbott et al. (2014)
tOstro et al. (2016)
tMiloievic et al. (2014)
»Samat et al. (20l5)
tMilojevic et aL (2014)
tStieb et al. (2009)
tSzyskowicz et al. (20
tWeichenthal et al. (21
*Liu etal. (2020)
tTalbott etal. (2014)
tOstro et al. (2016)
tRich et al. (2010)
fPope etal. (2015)
*Leiser et aL (2019)
*Evans et aL (2017)
tGardner et al. (2014)
Outcome
Location
Mean (ug/m3)
Lag
IHD
204 U.S. Counties
13.4
2
IHD
4 Australian Cities
S. 1-9.7
0-1
IHD
6 French Cities
13.S-1S.6
0-1
IHD
213 U.S. Counties
12.3
0
IHD
7 Mid-Atlantic States
11.9
0-1
MI
Boston, MA
11.1
0
MI
213 U.S. Counties
12.3
0
MI
26 U.S. Cities
15.3
0-1
MI
U.S. (National)
11.5
0-1
S New York Cities
5.8
0
IHD
5 Cities. U.S.
10.8-15.4
0
IHD
4 NY Regions
NR
0
0
IHD
7 U.S. States
6.5-12.8
0
0-2
0-2
0-2
0-2
0-2
0-2
0-2
IHD
8 CA counties
16.5
0
IHD
15 Conurbations. UK.
10
0-4
IHD
St. Louis. MO
IS
0-2
MI
15 Conurbations, UK
10
0-4
0-4
0-4
Angina MI
6 Canadian Cities
6.7-9.8
0
Angina
6 Canadian Cities
8.3
0
Mr
16 Cities Ontario
6.9
0-2
MI
Alberta, Canada
9.79
0-2
MI
7 U.S. States
6.5-12.8
0-2
MI
MI
IHD
MI
MI
mi
S CA counties
New Jersey state
4 counties, Utah. U.S.
Rochester. NY
Rochester. NY
16.5
7.6-12.3
10.96
17.1
0-2
0-2
0-2
0-2
0-2
0-2
0
0-23h
0-23h
0-23h
0
0
0
0-2
1 h
0-1
0-1
Notes
Ages 65+
Ages 65+
Ages 65+
Ages 65+
Aaes 65+
Ages 65+
Ages 65+
Ages 65+
Medicaid
Traditional Multicitv
MCM Multicity
NYC, Lonp Island & Hudson
Adirondack & North
Mohawk Valley & Binghamton
Central & Western NT
Florida
Massachusetts
New Jersey
New Hampshire
New Mexico —
New York
Washington
STEMI
NSTEMI
Overall
High N02
Florida
Massachusetts
New Jersey
New Hampshire
New Mexico
New York
Washington
Transmural MI
Non-Transmural MI
STEMI
NSTEMI
Unstable Angina
30 dReadmission
30 dReadmission
STEMI
STEMI M
NSTEMI -4
4
i •
i—
0.8
0.9
1
1.1
1.2
1.3
Relative Risk (95% CI)
Source: Update of Figure 6-2, 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. IHD = ischemic heart
disease, MCM = multi-cause multicity; Ml = myocardial infarction, NR = not reported; NSTEMI = non-ST segment elevation Ml, STEMI = ST- elevation Ml. Risk estimates are
standardized to a 10 pg/m3 increase in PM2.s concentrations.
Figure 3-1 Results of studies of short-term PM2.5 exposure and hospital admissions and emergency
department visits for ischemic heart disease.
May 2022
3-12
-------
3.1.1.2.2. Cerebrovascular Disease and Stroke
Cerebrovascular disease (CBVD) typically includes conditions classified under ICD-10 codes
160-169 (ICD-9: 430-438) such as hemorrhagic stroke (HS), cerebral infarction (i.e., ischemic stroke
[IS]) and occlusion of the precerebral and cerebral arteries. IS results from an obstruction within a blood
vessel that supplies oxygen to the brain, potentially leading to infarction, and accounts for 87% of all
strokes (Goldberger et al.. 2008). Hemorrhagic stroke is less common but results in a disproportionate
number of fatalities. The HS subtype results from a brain aneurysm or leaking vessel in the brain and can
be further categorized by brain region (e.g., intracerebral, or subarachnoid). 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 nonfatal 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
May 2022
3-13
-------
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/ni1 versus 12 to 150.4 |ig/m3) and the odds of
stroke in the higher category was compared with 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 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 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 modeled 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. The epidemiologic evidence for an association
between short-term PM2 5 and various stroke subtypes assessed in the 2019 PM ISA was characterized as
inconsistent and limited. Some recent studies report evidence of a positive association with stroke while
others report null or inverse associations. Therefore, the evidence pertaining to the effect of short-term
PM2 5 exposure and stroke remains inconsistent overall.
May 2022
3-14
-------
Study
*deSouza et al. (2019)
Outcome
Ischemic Stroke
Location
Medicaid, U.S.
Mean
(Hg.'m3)
11.5
Lag
0-1 dav
*Fisher et al. (2019)
Stroke
Ischemic Stroke
Hemorrhagic Stroke
Undetermined
non-fatal
HPFU, U.S.
12.9
0-3 dav
*Sun et aL (2019)
Stroke
Ischemic Stroke
Hemorrhagic Stroke
win, u.s.
10.8 (Med)
3 d avg
*McLure et al. (2017)
Stroke
REGARDS, U.S.
NR
fKloog et al. (2012)
Stroke
6 New England States
9.6
0-1
tKloog et al. (2014)
Stroke
7 Mid-Atlantic States
11.9
0-1
tMilojevic et al. (2014)
Stroke
15 Conurbations. UK
10 (Med.)
0-4
*Krall et al. (201S)
Stroke
5 Cities. U.S.
10.8-15.4
Lag d 1
tVilleneuve etal. (2012)
Stroke
Hemorrhagic Stroke
Ischemic Stroke
Transient Ischemic Attacks
Denver. CO
8.1
0
tWellenius et al. (2012)
Acute Ischemic Stroke
Boston. MA. +
102
0-24h
tLisabethetal. (2008)
Ischemic Stroke and
Transient Ischemic Attacks
Nueces County. TX
7
1
tWing et aL (2015)
Ischemic Stroke
Nueces County. TX
7.7 (Med)
0
*Wing et aL (2017)
Ischemic Stroke
BASIC-Nueces Co TX
7.7 (Med.)
Lag d 1
tODonnell etaL(2011)
Acute Ischemic Stroke
S Cities. Canada
6.9
0-47h
High v Low (<12 jig m3)
Age 65+
Age 65-!-
Traditional Multicity
MCM Multicity
Age 20-j-
Age 45+
Recurrent
Subjects < 20km
from monitor
4
i
+
i
•l
i
#¦
0.5 0.7 0.8 1 1.2
Relative Risk (95% CI)
1.4
Source: Update of Figure 6-5, 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. NR = not reported. Risk
estimates are standardized to a 10 |jg/m3 increase in PM2.5 concentrations.
Figure 3-2 Results of studies of short-term PM2.5 exposure and hospital admissions and emergency
department visits for stroke.
May 2022
3-15
-------
3.1.1.2.3. Heart Failure
Heart failure (HF) refers to a set of conditions in which pumping action of the heart is weakened.
In congestive heart failure (CHF), the flow of blood from the heart slows, failing to meet 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). The effect of short-term PM2 5 exposure on people with
CHF—which is a chronic condition—is generally evaluated using ICD codes recorded when a patient is
admitted or discharged from the hospital or ED. The relevant diagnostic codes for heart failure are ICD-9
428 and ICD-10 150. These codes encompass left, systolic, diastolic, and combined heart failure. Similar
to the other cardiovascular outcomes, the majority of the evidence in the 2009 PM ISA was from
epidemiologic studies of hospital admissions and ED visits [i.e., multicity studies in the U.S.(Dominici et
al.. 2006) and Australia (Barnctt 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 modeled 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
May 2022
3-16
-------
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 RR 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.
May 2022
3-17
-------
Study
Dominici et al. (2006)
Outcome
HF
204 U.S. Counties
Mean
(jig'nri)
13.4
Lag
0
Notes
Ages 65+
1
I
I •
Bamettetal. (2006)
HF
4 Australian Cities
8.1-9.7
0-1
0-1
Ages 14-64
Ages 65+
1 •
I •
fBell et al. (2015)
HF
213 U.S. Counties
12.3
0
Ages 65+
l«
tZanobettiet al. (2009)
HF
26 U.S. Cities
15.3
0-1
Ages 65+
l ~
•DeSouza et al. (2019)
HF
Medicaid, U.S.
11.5
0-1
I •
tTalbottetal. (2014)
HF
7 U.S. States
6.5-12.8
0-2
0-2
0-2
0-2
0-2
0-2
0-2
Florida
Massachusetts
New Jersey
New Hampshire
New Mexico
New York
Washington
-•-t
f-
T-
1 •
K®
1 ~
tHsuetal. (2017)
HF
4 NY Regions
NR
0
0
0
0
NYC. Long Island & Hudson
Adirondack & North —
Mohawk Valley & Bingham ton
Central & Western NY
1
• H
tHaley et al. (2009)
HF
8 New Yoric Cities
5.8
0
1
*Krall et al. (201S)
HF
5 Cities. U.S.
10.S-15.4
0
Traditional Multieitv
MCMMulticity
—4*-
4#-
tOstroetal. (2016)
HF
8 CA counties
16.5
0
tStieb etaL (2009)
HF
6 Canadian Cities
8.2
0
1 »
fMilojevic etal. (2014)
HF
15 Conurbations. UK
10 (Med.)
0-4
tRodopoulouet al. (2015)
HF,HHD
Little Rock. AR
12.4
1
*-l
fSarnat etal. (2015)
HF
St Louis. MO
18
0-2
—i-«
*Leiser et aL (2019)
HF
Wasatch Front. Utah
11
0-2
30 day readmit
1 »
•Wyattetal. (2020)
HF
ESRD Patient
9.3
0
0-7 d readmission
8-30 d readmission
1 m
1 »
0.8 0.9 1 1.1 1.2 1.3 1.4 1.5
Relative Risk (95% CI)
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 pg/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.
May 2022
3-18
-------
3.1.1.2.4. Arrhythmia
In epidemiologic studies, the association between short-term PM2 5 exposure and arrhythmia is
generally evaluated using ICD codes (ICD-9 427 or ICD-10 149.9) for hospital admissions and ED visits.
Out-of-hospital cardiac arrests (OHCA) that typically result from ventricular arrhythmia were evaluated
with the body of evidence pertaining to arrhythmia. Overall, the evidence evaluated in the 2009 PM ISA
and the 2019 PM ISA was limited. However, in the 2019 PM ISA, some evidence from epidemiologic
panel studies indicated an association between short-term PM2 5 exposure and potential indicators of
arrhythmia (e.g., ectopic beats and tachycardia). The small number of recent studies support a positive
association of short-term PM2 5 exposure with arrythmias.
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 among Medicare
beneficiaries who survived a cardiovascular event, and examined differences across sex and age in
models that adjusted for the competing risk of readmission due to a non-cardiovascular cause or death.
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. Krall et 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 modeled 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 RR 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 extend the limited
evidence evaluated in the 2019 PM ISA as they report positive associations between short-term PM2 5
May 2022
3-19
-------
exposure and arrhythmia in most studies. However, an analysis of Medicare recipients in Utah that
adjusted for the competing risk of readmission for a non-cardiovascular cause or death reported an inverse
association.
May 2022
3-20
-------
Study
Dominici et aL (2006)
tBell et al. (2015)
tTalbottet al. (2014)
tHsuet al. (2017)
tMilojevic et al. (2014)
tStieb etal. (2009)
tHaley etal. (2009)
tOstroetal. (2016)
~Krall etal. (2018)
tSarnat etal. (2015)
tRodopoulouet al. (2015)
•Wyattetal. (2020)
Outcome Location
Arrhythmia 204 U.S. Counties
Arrhythmia 213 U.S. Counties
Arrhythmia 7 U.S. States
Mean
(fig/tn3)
13.4
12.3
6.5-12.8
Arrhythmia 15 Conurbations. UK
Atrial Fibrillation
Arrhythmia 6 Canadian Cities
Arrhythmia 8 New York Cities
8 CA counties
Dysrhythmia ED Visits 5 Cities
Dysrhythmia
Arrhythmia St. Louis. MO
Arrhythmia Little Rock. AR
Dysrhythmia ESRD Patients
6.7-9.8
5.8
16.5
Lag Notes
0 Age 65+
0 Age 65+
10 (Med.) 0-4
0-4
Massachusetts
New Jersey
New York
NYC, Long Island & Hudson
Central & Western NY
10.8-15.4 Lag d 1 Traditional Multicity
Lag d 1 MCM Multicity
IS
12.4
Cold Season
Warm Season
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 PM2.5 concentrations.
Figure 3-4 Results of studies of short-term PM2.5 exposure and hospital admissions and emergency
department visits for arrhythmia.
May 2022
3-21
-------
3.1.1.2.5. Combinations of Cardiovascular-Related Outcomes
In addition to analyses of individual CVDs (e.g., MI, stroke, and HF), epidemiologic studies
examined CVDs in aggregate (i.e., specific combination of cardiovascular diseases). The 2009 PM ISA
and the 2019 PM ISA reviewed multicity studies of adults ages 65 years and older that provided strong
evidence of an association lYBell et al.. 2008; Host et al.. 2008; Barnett et al.. 2006); Table 6-19 of the
2019 PM ISA], Studies of aggregate CVD have larger case counts than studies of specific CVDs,
potentially providing statistical power needed to detect associations. Several recent studies examine the
association between short-term exposure to PM2 5 and CVD hospital admissions and ED visits, and report
results that are generally consistent with studies evaluated in the 2019 PM ISA.
In a study of low-income and/or disabled Americans enrolled in Medicaid deSouza et al. (2021)
estimated the association ofPM25 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). In addition, the authors reported that the association with all CVD hospital
admissions was larger in magnitude when restricting the analysis to PM2 5 concentrations less than
25 micrograms per cubic meter (|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 PM25 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.
Evidence assessed in the 2019 PM ISA from multicity studies reported consistent positive
associations between short-term PM2 5 exposure and cardiovascular-related ED visits and hospital
admissions. Recent studies, including one in renal disease patients and another in the Medicaid
population, support the conclusion of the 2019 PM ISA and extend the evidence base.
May 2022
3-22
-------
3.1.1.2.6. Cardiovascular Mortality
As noted in the 2019 PM ISA, "studies that examine the association between short-term PM2 5
exposure and cause-specific mortality outcomes, such as cardiovascular mortality, provide additional
evidence for PM2 5-related cardiovascular effects, specifically whether there is evidence of an overall
continuum of effects" (2019 PM ISA, Section 6.1.9). Epidemiologic studies evaluated in the 2019 PM
ISA, expanded upon the evidence presented in the 2009 PM ISA indicating consistent positive
associations between short-term PM2 5 exposure and cardiovascular mortality (2019 PM ISA, Section
6.1.9). Experimental evidence (i.e., both animal toxicological and controlled human exposure studies)
presented within both the 2009 PM ISA and 2019 PM ISA provided coherence and biological plausibility
for the PM2 5-related cardiovascular mortality associations reported in epidemiologic studies. A recent
multicity study conducted by Lavigne et al. (2018) in addition to examining short-term PM2 5 exposure
and total (nonaccidental) mortality also examined cause-specific mortality and provided evidence that
continues to support a relationship between short-term PM2 5 exposure and cardiovascular mortality
(Section 3.2.1.2.2).
3.1.1.2.7. Consideration of Copollutant Exposures
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. As noted in the Appendix (Table A-l) to the 2019
PM ISA, copollutant models are not without their limitations, such as instances for which correlations are
high between pollutants resulting in greater bias in results. However, a change in the PM2 5 risk estimate,
after adjustment for a copollutant may indicate the potential for confounding. The evidence reviewed in
the 2019 PM ISA represented an expanded set of studies that performed analyses using two-pollutant,
also referred to as copollutant, models. These studies addressed a data gap, generally supporting an
association of PM25 with cardiovascular-related health effects that persisted after adjustment for
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. Overall, the evidence and the available statistical
methods were limited with respect to characterizing the multipollutant effects of air pollution on
cardiovascular disease. This limited evidence neither consistently nor coherently indicated a stronger or
weaker effect of combined exposure to PM2 5 and another pollutant compared with exposure to a single
pollutant alone (Lubcn et al.. 2018).
Recent studies that examine the potential confounding of the relationship between short-term
PM2 5 exposure and cardiovascular effects by copollutants are limited; however, the results of available
studies are consistent with the evidence evaluated in the 2019 PM ISA. deSouza et al. (2021) found that
the positive single-pollutant association (OR: 1.09 [95% CI: 1.06, 1.11]) between all CVD and short-term
PM2 5 observed among Medicaid recipients persisted in a two-pollutant model adjusted for ozone (OR:
1.10 [95% CI: 1.07, 1.12]). In another recent study, Wing et al. (2017) reported no association between
May 2022
3-23
-------
short-term PM2 5 exposure and recurrent stroke in both single- and two-pollutant model that were adjusted
for ozone. This study does not alter the conclusion of the 2019 PM ISA with respect to copollutant
confounding because associations between PM2 5 exposure and stroke were not consistently reported.
3.1.1.2.8. Lag Structure of Associations
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 (e.g., lag 0-1 days), delayed (e.g.,
lag 2-5 days), or prolonged (e.g., lag 0-5 days) 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 with immediate single-day lag periods (i.e., 0, 1).
Several recent studies conducted analyses to determine whether results were sensitive to the
choice of exposure lag. Overall, the available studies continue to support an immediate effect of
short-term PM2 5 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 et al.. 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 Epidemiologic Studies Examining the PIVh.s-Cardiovascular
Effects Relationship through Accountability Analyses and Alternative
Methods for Confounder Control
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
May 2022
3-24
-------
epidemiologic studies that conducted accountability analyses or employed alternative methods for
confounder control because no such studies were published prior to the literature cutoff date for the 2019
PM ISA. Studies that conduct accountability analyses can provide insight on whether the implementation
of environmental policies or air quality interventions result in changes/reductions in air pollution
concentrations and the corresponding effect on health outcomes. Additionally, accountability studies can
reduce uncertainties related to residual confounding of temporal and spatial factors. Alternative methods
for confounder control seek to mimic randomized experiments through the use of study design and
advanced statistical methods to reduce the potential bias of effects due to confounding more than
traditional regression model approaches. Examples of alternative methods for confounder control are
general propensity scores and inverse probability weighting models. Since the literature cutoff date for the
2019 PM ISA, several studies that conducted accountability analyses or implemented alternative methods
for confounder control 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 conducted accountability analyses that evaluated
whether associations between short-term PM2 5 exposures and cardiovascular hospital admissions differed
before, during, and/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, examining associations withPM2 5 at lag 0 or averaged over the previous 1-7 days (lag 0-1,
0-2, 0-3, 0-4, 0-5, 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" implementation period and weaker associations observed
in the "after" implementation period for the 0-6 day lag average. Similar results were reported for
cerebrovascular disease, ischemic stroke, chronic rheumatic heart disease, hypertension, ischemic heart
disease, and myocardial infarction. The incidence rates of all disease categories decreased across the
study period. However, there was no difference in the excess rate of most cardiovascular disease
subgroups associated with each interquartile range increase in PM2 5 concentration "after" the
implementation of environmental policies and actions (2014-2016) compared with "before" (2005-2007)
or "during" (2008-2013) implementation. Conversely, there were increases in the excess rate of hospital
admissions for cardiac arrhythmia and congestive heart failure in the "after" period compared with the
"before" and "during" periods. Although the change in the excess rates for cause-specific cardiovascular
hospital admissions was relatively small in magnitude and varied by lag period and location, overall,
May 2022
3-25
-------
short-term increases in ambient PM2 5 concentrations were associated with increased rates of hospital
admissions for total cardiovascular disease, cardiac arrhythmias, heart failure ischemic stroke, ischemic
heart disease, and myocardial infarction.
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 with 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 an 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. Although the results of this accountability analysis are small in magnitude or null across
the time periods, this study provides support that implementation of air quality policies can lead to
reductions in PM2 5 concentrations and subsequently may affect health effects associated with PM2 5
exposures.
The use of alternative methods for confounder control can further inform the causal nature of the
relationship between short-term PM2 5 exposure and cardiovascular effects through the use of advanced
statistical methods to reduce uncertainties with respect to confounding. Recent epidemiologic studies that
use these alternative methods have primarily focused on examining cardiovascular hospital admission
rates. Inverse probability weighting (IPW) is an alternative method for confounder control 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 (Oiu 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
May 2022
3-26
-------
infarction (AMI), CHF, and IS hospital admissions among 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, for each of the six lags
the outcome was regressed against PM2 5 at each individual lag with weights specific to each individual
lag. After using the weights generated from propensity score models to predict the exposure at each lag
of interest, copollutant exposure and meteorological variables were used to create a pseudo-randomized
population. The pseudo-randomized population was subsequently used in conditional logistic regression
models to regress cardiovascular hospital admissions against each exposure lag, estimating the marginal
effect of each lag of exposure independent of covariates.
Using the IPW method, Qiu 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 Qiu et al. (2020)
reported associations using alternative methods for confounder control 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 exchangeability, meaning that there was no unmeasured confounding, with the caveat
that the authors did not have the resources to obtain all the potential unmeasured confounders. This
assumption was tested through a series sensitivity analyses, testing the most critical confounder of
temperature by including more lags of temperature and spline adjustments. Because the results from the
sensitivity analyses involving temperature did not deviate from the estimates in the main analysis, it can
be inferred that the most important confounders with available data were adjusted for and that the time-
invariant variables are not potential confounders due to the case-crossover study design. The second
assumption was positivity, which was guaranteed in the analysis through the positivity exclusion. Qiu et
al. (2020) note that the positivity assumption means 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; however,
this assumption is difficult to prove. Overall, the inverse probability weighted distributed lag model
employed by Qiu et al. (2020) provides unconstrained, less conditional effect estimates that are less
influenced by highly correlated covariates and reduces uncertainties regarding unmeasured confounders.
The recent studies that utilized accountability approaches and alternative methods for confounder
control 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 reduce uncertainties related to potential confounder bias, and further supports
the conclusions of the 2019 PM ISA.
May 2022
3-27
-------
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
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.
Multiple studies included in the 2019 PM ISA applied hybrid exposure assessment techniques
that combined land use regression with satellite AOD measurements and PM2 5 concentrations measured
at fixed site monitors. Most 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. 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. Finally, recent studies that employed
alternative methods for confounder control 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 while reducing uncertainties
related to potential confounder bias.
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
May 2022
3-28
-------
ISA (Section S.l^.l) 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).15 In addition, with the expansion of epidemiologic
studies that used statistical approaches that attempt to more extensively account for confounders and are
more robust to model misspecification (i.e., used alternative methods for confounder control), 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. Finally, 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 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.
In addition to evaluating evidence across scientific disciplines 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 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 biological 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
15 Throughout this section, as detailed in the Preface of the 2019 PM ISA (Section P.3.2.2), risk estimates from
epidemiologic studies examining long-term exposures are for a 5 |ig/m3 increase in annual concentrations, unless
otherwise noted.
May 2022
3-29
-------
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).
The evidence for the relationship between long-term exposure to PM2 5 and cardiovascular effects
as characterized in the 2019 PM ISA is described below and summarized in Table 3-2. using the
framework for causality determinations described in the Preamble to the ISAs (U.S. EPA. 2015).
May 2022
3-30
-------
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.
PM2.5 Concentrations
Rationale for Key References and Associated with
Causality Sections in the 2019 PM Effects
Determination3 Key Evidence13 ISAb (|jg/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 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
evidence from
epidemiologic
studies of CHD or
stroke
Association with coronary events, CHD,
and stroke (mortality and morbidity
combined) that persist after adjustment for
SES reported in the WHI study.
Association with stroke but not CHD in the
CA Teachers cohort.
No association with CHD or stroke in the
NHS or HPFU.
Section 6.2.2
Section 6.2.3
Range: 13.4-17.1
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.
Association among women with diabetes
in NHS.
Tonne et al. (2015)
Koton et al. (2013)
Hart et al. (2015b)
Mean: 15.5
Mean: 14.6
Mean: 23.9
Mean: 13.4
May 2022
3-31
-------
PM2.5 Concentrations
Rationale for Key References and Associated with
Causality Sections in the 2019 PM Effects
Determination3 Key Evidence13 ISAb (|jg/m3)
Some, but not all, Longitudinal change in CAC observed in Section 6.2.4 Mean: 14.2
epidemiologic MESA but not in Framingham Heart Kaufman et al. (2016) Median: 9.8
studies provide Offspring study. ^ x
evidence for effect Dorans et al. (2016)
of long-term PM2.5
on CAC
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
Note: This table corresponds to Table 6-54 in the 2019 PM ISA.
CAC = coronary artery calcification; CCHS = Canadian Community Health Survey; C-R = concentration-response;
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.
a Based 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).
b Describes 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.
0 Describes the PM25 concentrations with which the evidence is substantiated.
May 2022
3-32
-------
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 to 17.9 (ig/m3. Generally, most of the PM25 effect estimates relating long-term PM25
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 etal.. 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 Nurses' Health Study (NHS) and California Teachers Study (CTS), both of which are
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
coronary artery calcification (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 carotid intima media thickness (cIMT) were not consistently observed across cohorts or
between analyses of the same cohort with variable methods. Relationships between PM2 5 and cIMT at
May 2022
3-33
-------
younger ages were not observed. However, a toxicological study supported 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. (Lippmann et al.. 2013a). 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
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 in previous assessments, which
provided the scientific rationale supporting a causal relationship between long-term PM2 5 exposure and
May 2022
3-34
-------
cardiovascular effects (Section 3.1.1.IV In addition to examining the relationship between long-term
PM2 5 exposure and specific cardiovascular outcomes (i.e., IHD and myocardial infarction
rSection 3.1.2.2.11. cerebrovascular disease and stroke rSection 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
[Section 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 3.1.2.2.9). The following sections present an evaluation of 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. Study-specific details (e.g., study
population, exposure assessment approach, confounders considered) for the epidemiologic studies
evaluated in this section are presented in Appendix A (Table A-3).
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. Most 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 with 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.
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, in terms of magnitude, with acute MI was observed
May 2022
3-35
-------
in the highest tertile of (> 38.97 ppm) Ox concentrations (HR: 1.12 [95% CI: 1.09, 1.15]) compared with
the lowest (HR: 1.04 [95%CI: 1.01, 1.06]) and middle tertiles (HR: 1.06 [95% CI: 1.00, 1.12]).
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 nonlinear 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., sulfate, 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 with
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 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 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.
May 2022
3-36
-------
In another study, Loop et al. (2018) conducted an analysis of the 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 nonfatal 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 nonfatal 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 MI, 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
diseases or patient populations that are followed after a cardiac event or procedure such as catheterization.
May 2022
3-37
-------
Study
Cohort
Outcome
Years
Mean (jig/m3)
Miller et al. 2007
WHI-Women (post-menopause),36
Urban sites, U.S.
CHD
1994-1998
13.4
fHart et al. 2015
NHS-Women, 48 States, U.S.
CHD
1989-2006
13.4
fLipsett etal. 2011
CTS -Women. Los Angeles.
California. U.S.
MI
1999-2005
15.6
tPuett etal. 2011
HPFU. Men, 13 States. US.
Nonfatal MI
1988-2002
17.8
tMadrigano etal. 2013
Worcester Heart Attack. MA U.S.
Confirmed MI
1995-2003
9.4
tHartiala et al. 2016
Cardiac Patients, Ohio, U.S.
MI
1998-2010
15.5
fHoffman et al. 2015
HNR study, Ruhr region Germany
Coronary Event
2008-2009
18.4 M—
f Atkinson et al. 2013
GP Database, U.K.
MI
2003-2007
12.9
tCesaroni et al. 2014
ESCAPE- 11 Cohorts Europe
IHD
2008-2011
7.3-31
*Yazdi et al. 2019
Medicare southeastern U.S.
MI
2000-2012
NR
~Loop et al. 2018
REGARDS U.S.
MI
2003-2012
13.6 —
*Bai etal. 2019
ONPHEC Ontario Canada
MI
1998-2012
9.6
* Elliot et al. 2020
NHS-Women, 48 States, U.S.
MI
1988-2007
13.7
*Weaver etal. 2019
CATHGEN, RTPNC
MI
2000-2008
12.7
tTonne et al. 2015
MINAP, London, U.K.
Recurrent ML death
2003-2010
14.6
tKoton et al. 2013
8 Treatment Centers, Israel
Recurrent MI
2003-2005
23.9
I •
I
I
I
I
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 PM2.5 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
Cerebrovascular disease typically includes the conditions hemorrhagic stroke, cerebral infarction
(i.e., ischemic stroke), and occlusion of the precerebral and cerebral arteries. The 2009 PM ISA identified
one study that indicated a positive association between PM2 5 and cerebrovascular morbidity and mortality
(HR: 1.16 [95%CI: 1.04, 1.30]) in post-menopausal women (Miller et al.. 2007). Although the results
were not entirely consistent across studies or stroke subtype, some studies reviewed in the 2019 PM ISA
provided evidence to support a positive association between long-term exposure to PM2 5 and stroke.
Several recent studies that observe positive associations add to this evidence base (Figure 3-6).
May 2022
3-38
-------
Several studies examined the association between long-term PM2 5 concentration and stroke as
discussed below. 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, PA, 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.
May 2022
3-39
-------
Study
Miller et al.2007
tHartet al. 2015
•Elliot etal. 2020
tLipsett et al. 2011
WHI-Women (post-menopause) 36 Urban „ ,
Sites, US btf0ke
NHS- Women. 4S States. U.S.
NHS- Women. 4S States. U.S.
Stroke
Stroke
CTS -Women. Los Angeles. California.
Years
1994-1998
19S9-2006
19SS-2007
1999-2005
Mean (fig/m3) i
tPuett et al. 2011
HPFU. Men. 13 States, U.S.
IS
19SS-2002
tPuettetal. 2011
HPFU. Men. 13 States, U.S.
HS
1988-2002
*Yazdi etal. 2019
Medicare U.S.
1st Admission Stroke
2000-2012
•Shin et al. 2019
ONPHEC, Ontario Canada
1st Admission Stroke
1998-2012
•Rhinehart et al. 2020
Atrial Fibrillation Patients, U.S.
IS
2007-20017
tHartiala et al. 2016
Cardiac Patients, Ohio. U.S.
Stroke
1998-2010
tStafoggia et al. 2014
ESCAPE- 11 Cohorts Europe
Stroke
2008-2011
tHoffmann et al. 2015
HNR study, Ruhr area, Germany
Stroke
2008-2009
tAtkinsonet al. 2013
GP Database, U.K.
Stroke
2003-2007
tKotonet al. 2013
S Centers, Israel
Post MI Stroke
2003-2005
13.4
13.4
13.7
15.6
17.8
17.8
NR
9.8
10.6
15.5
7.3-31
18.4
12.9
23.9
i
f
i
0.5 1
3 4 5
Relative Risk (95% CI)
Source: Update of Figure 6-18, 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 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 |jg/m3 increase in PM25 concentrations. CU.S. EPA. 20181. 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
Atherosclerosis is the process of plaque buildup that forms lesions on the walls of the coronary
arteries, which can lead to narrowing of the vessel, reduced blood flow to the heart and IHD.
Atherosclerosis can be assessed within large arterial vascular beds in distinct regions of the body
i.e., carotid intima-media thickness (cIMT), coronary artery calcification (CAC), ankle-brachial index
(ABI), and the presence of plaques. Findings from studies reviewed in the 2009 PM ISA were
inconsistent, reporting null or positive, but imprecise associations with cIMT, CAC, and ABI. Similarly,
findings from studies reviewed in the 2019 PM ISA were not entirely consistent across populations,
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 (|3: -0.90 |_im per year [95%CI: -3.00,
May 2022
3-40
-------
1.30]) 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
HF refers to a set of conditions including CHF in which the heart's pumping action is weakened.
With CHF the blood flow from the heart slows, failing to meet 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
May 2022
3-41
-------
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 (Aaron et al.. 2016).
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 increased RV mass, reduced flow to the left ventricle, and
reduced left ventricular (LV) mass.
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]). The hazard ratios in the lowest
and middle Oxtertiles were 1.04 (95% CI: 1.03, 1.06) and 1.06 (95% CI: 1.03, 1.07), respectively. This
study supports the evidence in the 2019 PM ISA that indicates a positive association between long-term
PM2 5 and HF; however, the evidence remains limited overall.
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 and quality of life and
are associated with downstream consequences such as ischemic stroke (Prvstowskv et al.. 1996; Laupacis
etal.. 1994) and CHF (Roy et al.. 2009). contributing to both 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 AF and reported a
positive association (HR: 1.03 [95% CI: 1.01, 1.04]). Overall, the evidence pertaining to the association
between long-term PM2 5 exposure and various types of arrythmias remains limited.
May 2022
3-42
-------
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 in the 2009 PM ISA with longitudinal
analyses generally showing small magnitude increases in SBP, pulse pressure (PP), and mean arterial
pressure (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. (HR: 1.17 [95% CI 1.10, 1.22]). The association with PM2 5 concentration
increased among minority participants (i.e., Black, Asian/Pacific Islander, Hispanic/Latino race/ethnicity,
which were characterized as non-White in the study) 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 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 etal.. 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).
The literature assessed in the 2019 PM ISA provided evidence of associations between long-term
PM2 5 exposure and hypertension. Recent studies are generally consistent with this assessment, reporting
positive associations among post-menopausal women enrolled in the WHI study and in cardiac
catheterization patients. However, no association between long-term PM2 5 exposure and hypertension
was observed among Black women enrolled in the JHS.
May 2022
3-43
-------
3.1.2.2.7. Cardiovascular Mortality
Multiple epidemiologic studies (Section 3.2.2.2.2) reviewed in the 2009 PM ISA and in the 2019
PM ISA reported consistent 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
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 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 etal.. 2019).
3.1.2.2.8. Copollutant Confounding
One approach to assessing the independence of the association between exposure to PM2 5 and a
health effect, such as long-term exposure to PM2 5 and cardiovascular health effects is through the use of
copollutant models. As noted in the Appendix (Table A-3) to the 2019 PM ISA, copollutant models are
not without their limitations, such as instances when correlations are high between pollutants resulting in
greater bias in results. However, in assessing the results from copollutant models, a change in the PM2 5
risk estimates, after adjusting 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 (HR: 1.03 [95%CI: 1.01, 1.04]) persisted after adjustment for NO2 but was attenuated in the
May 2022
3-44
-------
models with O3 and oxidant gases (Ox) represented by the redox weighted average of NO2 and O3 (HR:
1.05 [95% CI: 1.03, 1.06] andHR: 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] andHR: 1.02 [95% CI: 1.01, 1.03], andHR: 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 (mi [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 models adjusted for ozone (OR: 2.29 [95% CI:
0.70, 7.59]) for plaque presence and (OR 3.05 [95% CI: 0.86, 10.82] for plaque index progression).
Overall, the limited evidence indicates that associations between PM2 5 and cardiovascular health effects
persist, but may be slightly attenuated, in models that are adjusted for copollutants.
Study
Cohort
Outcome
CopolUutant (r)
tPuett etal. 2011
HPFU Study
MI
Single
PM10-2.5 (NR)
fMadrigano et al. 2013
Worcester Heart Attack
MI
Single
1
# ~
PM2.5 local (NR)
1
1
0—~
tPuett etal. 2011
HPFU Study
Ischemic Stroke
Single
PM10-2.5 (NR)
«0—|
Hemmoragic Stroke
Single
j-#
PM10-2.5 (NR)
—1 O
*Shin et al. 2019
ONPHEC
Stroke
Single
k
N02 (0.65)
D
|
03 (0.28)
•
Ox (0.67)
*Shin et al. 2019
ONPHEC
Atrial Fibrillation
Single
•
NO2(0.65)
03 (0.28)
*
Ox (0.67)
<£>
tFuks et al. 2014
ESCAPE
Hypertension
Single
N02 (0.19-0.88)
K3-
BPLM
Single
V
N02 (0.19-0.88)
I0"
*Duan et al. 2019
SWAN
Plaque Presence
Single
1
—• ~
03 (NR)
1
1
© ~
Plaque Index
Single
1
—• ~
03 (NR)
1
0 ~
1 1 1 1 1 1 1 1
0.5 1 1.5 2 2.5 3 3.5 4
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.
May 2022
3-45
-------
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 whether there are concentration ranges that exhibit 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
Location—Cohort
(Table/Figure from Reference)
Baietal. (2019)
Figure 3-8
Ontario, Canada
ONPHEC
Exposure
PM2.5 Mean:
Outcome (Range) in |jg/m3
Acute Ml Mean (IQR):9.6
incidence (3.5)
Statistical Analysis
Summary
Identified the shape of the C-R
function for fully adjusted Cox models
using SCHIF** (Nasari etal.. 2016).
A linear concentration-response
relationship between acute Ml and
PM2 5 concentration was observed.
Chen et al. (2020) Acute Ml Mean: 8.61 Identified the shape of the C-R
Fiqure 3-9 Incidence function for fully adjusted Cox models
—7 usina SCHIF** (Nasari etal.. 2016).
Ontario, Canada
Restricted cubic splines with 4 df to
ONPHFP
assess linearity used in sensitivity
analysis.
Approximately linear relationship
observed with both methods.
Danesh Yazdi et al. (2019) First hospital NR
Fiaure 3-10 admission for
V7Z- Ml
Medicare
Southeastern, U.S.
May 2022
3-46
Penalized spline to estimate the
shape of the C-R relationship, with
degrees of freedom chosen based on
corrected AIC values.
-------
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 data set 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
Nonfatal Ml
incidence
Median (IQR): 13.6
(2.7)
Predicted log hazard modeled as a
linear function (nonlinear 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. (2019)
Medicare
Southeastern, U.S.
First 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.
May 2022
3-47
-------
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)
Baietal. (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; df = degrees of freedom; ESCAPE = European Study of Cohorts for Air Pollution Effects, HR = hazard ratio,
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.
Several studies evaluated the shape of the C-R function for the relationship between long-term
PM2 5 exposure and MI, including two analyses of the ONPHEC study (Figure 3-8 and Figure 3-9). an
analysis of the U.S. Medicare population (Figure 3-10). and an analysis of the REGARDS cohort (Figure
3-11). Approximately linear relationships were observed in the ONPHEC analyses (Chen et al.. 2020; Bai
et al.. 2019) using Shape Constrained Health Impact Function (SCHIF) method (Nasari et al.. 2016).
which is described as a new class of variable coefficient risk functions that can capture potentially
nonlinear associations, and in the Medicare analysis using penalized splines, which is described in
Section 3.1.2.1 (Danesh Yazdi etal.. 2019). Both methods allow for deviations from linearity. By
contrast, Loop et al. (2018) found an inverse relationship between annual average PM2 5 exposure and
nonfatal MI.
May 2022
3-48
-------
PM2.5 (ug/m3)
Source: Bai et al. (2019)
Note: The gray shaded area represents the 95% confidence interval.
Figure 3-8 Concentration-response relationship for the association of PM2.5
concentration with acute myocardial infarction.
7 * C J 10 1! 14 16
PM: a Mass Concentration (ugJm )
(A) PM2.5 mass and incidence of AMI
ijpf
15 3.0 45 6.0 75 90 10.5 12.0 13.5 15.0 165 18.0
PM2^5 (ug;m3)
(A) PM2.5 mass and incidence of AMI
Source: Chen et al. (2020)
Note: The gray shaded area represents the 95% confidence interval.
Figure 3-9 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.
May 2022
3-49
-------
PM2.5 vs. First Admission lor Ml
o> _
©
•2
|
«
IT)
O
t
o
Source: Danesh Yazdi et al. C2019)
Note: The gray 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.
"O
!_
03
N
CO
X
-t—'
m
0
a:
CO
o
Source: Loop et al. (2018'i
Note: Gray bands are 95% prediction intervals.
Figure 3-11 Predicted log hazard for incident nonfatal myocardial infarction
versus previous 1-year mean ambient PM2.5 concentration.
Annual PM2.0 (inca/in3)
1.5-
1.0-
0.5-
0.0-
¦0.5-
¦1.0-.
Nonfatal MI
10 15 20
PM25(Lig/m )
May 2022
3-50
-------
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 PM2 5 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 at approximately 14 |ig/m3 (Bai et al.. 2019).
3.1.2.3. Recent Epidemiologic Studies Examining the PIVh.s-Cardiovascular
Effects Relationship through Accountability Analyses and Alternative
Methods for Confounder Control
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 alternative methods for confounder control because no
such studies were prior to the literature cutoff date for the 2019 PM ISA. Since the literature cutoff date of
the 2019 PM ISA, a few recent studies conducted accountability analyses or employed alternative
methods for confounder control 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
and 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 over time 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
May 2022
3-51
-------
(DRAM). The steps for the approach used by Danesh Yazdi et al. (2021) is depicted in Figure 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.
Exposure modeling for
PM, c, NO,, O,
• Data derived from
satellite
measurements,
chemical transport
models,
meteorological and
land-use data
• Ensemble GAM
model using
predictions from RF,
GBM, NN
• Linear propensity
score model for
each pollutant
• Control for
individual,
socioeconomic, and
access to care
variables as well as
other pollutants
• 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
Source: Danesh Yazdi et al. (2021)
Figure 3-12 Analysis steps used by Danesh Yazdi et al. (2021) to examine
long-term PM2.5 exposure and cardiovascular-related hospital
admissions.
Zigler et al. (2018) used a hybrid approach of integrating an accountability analysis with an
alternative method for confounder control to examine whether attainment status for the 1997 NAAQS led
to an improvement in PM2 5 concentrations and subsequently health. By focusing on nonattainment
designations, the authors are able to examine the role of local control strategies in reducing PM2 5
concentrations that occurred above and beyond reductions due to regional strategies. Within this study,
Zigler et al. (2018) employed propensity scores, within a spatial hierarchical regression model to examine
whether designation of nonattainment in 2005 for the 1997 PM NAAQS, for either the annual standard of
15 (ig/m3 or the daily standard of 65 (ig/m3, led to a corresponding reduction in ambient PM2 5
May 2022
3-52
-------
concentrations and hospitalization admission rates for cardiovascular-related outcomes (i.e.,
cardiovascular stroke, heart failure, heart rhythm disorders, ischemic heart disease, and peripheral
vascular disease) among Medicare beneficiaries in the eastern U.S. from 2009 to 2012. Using publicly
available date sources for the analysis, Zigler et al. (2018) compared average annual PM2 5 concentrations
and selected cardiovascular hospital admission rates in nonattainment areas against those in attainment
areas using a two-step approach and adjusting for confounding factors that differed between the areas.
In the first step, propensity scores were used to adjust for confounders by grouping attainment
and nonattainment areas based on similarities in baseline characteristics, which are detailed in Appendix
A (Table A-4). Under the assumption that these baseline factors comprise all factors that differ between
locations in attainment and nonattainment areas and that the factors are correlated with both the exposure
(ambient PM2 5 concentration) and each outcome (cardiovascular-related hospital admission rates), there
should be no unmeasured confounders. To ensure that nonattainment areas are compared only with
attainment areas with similar baseline factors, (1) propensity scores were estimated based on the
probability that a monitoring location is in a nonattainment area, conditional on the baseline factors; (2)
areas with features that are not comparable to other areas in the comparison group were identified and
omitted (propensity score pruning); and (3) the remaining locations were grouped into quartiles, where
attainment and nonattainment areas have similar baseline factors (e.g., population, PM2 5 concentrations,
demographics) within each subgroup.
In the second step, a spatial hierarchical regression model was used to predict the potential annual
ambient PM2 5 concentration in 2010-2012 that would have occurred in nonattainment areas if the
designations had never occurred. For this part of the analysis, the spatial hierarchical model is estimated
jointly with a log-linear model using the same confounding adjustment for propensity score group and
additional covariates for each type of cardiovascular hospital admission. In addition to estimating the
effect estimates for the overall average effects, a principal stratification approach was used to estimate
"associative effects" and "dissociative effects." Within this study, Zigler etal. (2018) define effect
estimates for the "associative effects" as the effects of the nonattainment designations on cardiovascular-
related hospital admissions among areas where the nonattainment designations are estimated to reduce
ambient PM2 5 concentrations by at least 1 |ig/m3. and "dissociative effects" as the effects of the
nonattainment designations estimated to not affect PM2 5 concentrations by more than ±1 (ig/m3.
Zigler et al. (2018) reported a slight reduction in the overall average effect for hospital admission
rates for cardiovascular stroke, heart failure, and ischemic heart disease and an increase in the overall
effect for hospital admission rates for peripheral vascular disease; however, 95% CIs were wide and
included zero. There was no evidence of a reduction in hospital admission rates for heart rhythm disorders
or peripheral vascular disease. When examining the average "associative effects," the authors reported an
average reduction of-2.38 (95% CI: -4.35, -0.44) and -2.60 (95% CI: -4.24, -1.14) for only heart
failure and ischemic heart disease hospital admissions per 1,000 person-years, respectively. The authors
reported a similar pattern of associations for the average "dissociative effects," that is, slight reductions in
May 2022
3-53
-------
hospital admission rates for only cardiovascular stroke and heart failure with wide 95% CIs that included
the null. Overall, the results of Zigler et al. (2018) provide evidence that reductions in ambient PM2 5
concentrations and the selected cardiovascular hospital admissions could not be conclusively attributed to
nonattainment designations against the backdrop of other regional strategies that impacted the eastern
U.S.
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 use 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.
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 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 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 (Rhinehart et al.. 2020; Ward-Caviness et al.. 2020; Malik et
May 2022
3-54
-------
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. Although 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,
there is some evidence for a sublinear or supralinear C-R relationship for specific outcome (i.e., CHF and
AF). Finally, a few recent epidemiologic studies that employed alternative methods for confounder
control 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).16 In addition, with the expansion of epidemiologic studies that
used statistical approaches that attempt to more extensively account for confounders and are more robust
to model misspecification (i.e., used alternative methods for confounder control), 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. Finally, 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
16 Throughout this section, 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, unless otherwise noted.
May 2022
3-55
-------
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 land use regression [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.
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 (2019 PM ISA, Section
11.1.1). 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 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. In summary, although there was coherence of effects across the scientific disciplines (i.e., animal
toxicological, controlled human exposure studies, and epidemiologic) and biological plausibility for
PM2 5-related cardiovascular (2019 PM ISA, Chapter 6) and respiratory (2019 PM ISA, Chapter 5)
morbidity, there was strong evidence indicating biological plausibility for PM2 5-related cardiovascular
mortality with more limited evidence for respiratory mortality.
May 2022
3-56
-------
This section describes the evaluation of evidence for total (nonaccidental) mortality conducted in
the 2019 PM ISA, with respect to the causality determination for short-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 3-4.
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.
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 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 > 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)
May 2022
3-57
-------
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
Note: This table corresponds to Table 11-4 in the 2019 PM ISA.
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
May 2022
3-58
-------
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.. 2013b) 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 hybrid modeling methods, which use additional sources of
PM2 5 data (i.e., monitor, satellite, and LUR) to estimate PM2 5 concentrations and assign exposure,
allowing for the inclusion of less urban and rural locations in 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.. 2013b) 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 evaluated in the
2019 PM ISA 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
May 2022
3-59
-------
provide evidence that 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/nr1) provided initial evidence
May 2022
3-60
-------
indicating that PM2 5-mortality associations persist and may be stronger (i.e., a steeper slope) at lower
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
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 in previous assessments, which
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,
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 3.1.1.2.6V The following sections present an evaluation of 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. Study-specific details (e.g., study population, exposure assessment approach,
confounders considered) for the epidemiologic studies evaluated in this section are presented in Appendix
A (Table A-5V
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 globally, specifically in locations where mean 24-hour
concentrations were generally < 20 (ig/m3 (2019 PM ISA, Section P.3.1). Taken together, these studies
May 2022
3-61
-------
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).
Study
Burnett and Goldberg (2003)
Klemm and Mason (2003)
Burnett et al. (2004)
Zanobettiand Schwartz (2009)
Dominici et al. (2007)
Franklin etal. (2007)
Franklin et al. (2008)
Ostroetal. (2006)
tLippmann et al. (2013)
tBaxter et al. (2017)
tDaietal. (2014)
fKrall etal. (2013)
tKloog et al. (2013)
tLee et al. (2015)a
tJanssenetal. (2013)
f Samoli et al (2013)
tStafoggia et al. (2017)
fLanzinger etal. (2016)b
fPascal etal. (2014)
tLee et al. (2015)
*Liu etal. (2019)c
*Liu etal. (2019)c
*Lavigne etal. (201S)
*Shin etal. (2021)
tDiet al. (2017)c
fZanobettiet al. (2014)c
tShi et al. (2015)c
tYoung etal. (2017)
f Ueda et al. (2009)f
t Atkinson et al (2014)
tAdaret al.(2014)
Location Lag
8 Canadian cities 1
6 U.S. cities 0-1
12 Canadian cities 1
112 U.S. cities 0-1
96 US. cities (NMMAPS) 1
27 U.S. cities 1
25 US. 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 US. 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
All Ages
0.5 1.0 1.5 2.0 2.5 3.0
°o 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 |jm;
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 |jg/m3
increase in PM25 concentrations.
aResults are from modeled PM2.5 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. (20151. Zanobetti et al. (20141. and Liu et al. (20191 only had data for all-cause mortality including accidental mortalities.
dMain model used in Young et al. (20171 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. (20171 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. (20091 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. (20141 primarily focused on single-day lag results.
hAdar et al. (20141 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.
May 2022
3-62
-------
Recent studies that conducted multicity analyses in the U.S. and Canada include a large
international study that performed a global multicity analysis (Liu et al.. 2019) and a few studies in
Canada that relied on data from over 20 cities (Shin et al.. 2021a; Lavigne et al.. 2018). Using the
Multi-City Multi-Country (MCC) Collaborative Research Network, Liu et al. (2019) was able to collect
data globally, resulting in a data set consisting of air pollution and mortality data from 652 urban areas in
24 countries from 1986 to 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 model to pool city-specific estimates into a country-specific estimate. All analyses relied
on PM2 5 data for which the highest and lowest 5% of data was trimmed to remove outliers. In analyses of
25 Canadian cities from 1986 to 2011 and 107 U.S. cities from 1987 to 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. (2021a) 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 3.2.1.2.4). 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).
Sudden 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
May 2022
3-63
-------
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
previously evaluated studies focused on a cardiovascular outcome (i.e., OHCA) and as a result were
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 with 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).
May 2022
3-64
-------
Study
Location
Lag
Zanobettiet al. (2009)
112 U.S. cities
0-1
Ostroetal. (2006)
9 CA counties
0-1
Franklin et al. (2008)
25 U.S. cities
0-1
Franklin etal. (2007)
27 U.S. cities
1
tLippmannet al. (2013)
148 U.S. cities
1
tDai et al. (2014)
75 U.S. cities
0-1
tLee et al. (2015)
3 Southeast states. U.S.
0-1
*Lavigtie et al. (2018)
24 Canadian cities
0-2
f Samoli et al. (2013)
10 European Med cities
0-1
tPascal etal. (2014)
9 French cities
0-1
tLanzinger etal. (2016)a
5 Central European cities (UFIREG)
0-1
f Janssen et al. (2013)
Netherlands
0
fLee et al. (2015)
11 Asian cities
0-1
tChen et al. (2011)
3 Chinese cities (CAPES)
0
t Atkinson et al. (2014)
Meta-analysis
—a
tAdaret al. (2014)
Meta-analysis
—b
Zanobetti et al. (2009)
112 U.S. cities
0-1
Ostroetal. (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 et al. (2013)
148 U.S. cities
1
tDai et al. (2014)
75 U.S. cities
0-1
fLee et al. (2015)
3 Southeast states. U.S.
0-1
*Shin et al. (2021)
24 Canadian cities
1
"Lavigne et al. (2018)
22 Canadian cities
0-2
f Samoli et al. (2013)
10 European Med cities
0-5
tPascal et al. (2014)
9 French cities
0-1
f Lanzinger et al. (2016)a
5 Central European cities (UFIREG)
2-5
t Janssen et al. (2013)
Netherlands
3
fLee et al. (2015)
11 Asian cities
0-1
fChenet al. (2011)
3 Chinese cities (CAPES)
0
t Atkinson et al. (2014)
Meta-analysis
—b
tAdaret al. (2014)
Meta-analysis
—c
Cardiovascular
-• Respiratory
-4.0 -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 4.0
% Increase (959 b Confidence Interval)
Source: Update of Figure 11-2, 2019 PM ISA.
avg = average; |jg/m3 = microgram per cubic meter; PM = particulate matter; PM2.5 = particulate matter with a nominal mean
aerodynamic diameter less than or equal to 2.5 |jm; 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 |jg/m3 increase in PM25 concentrations. All ages examined for all studies except Lanzinger et al. (20161 and
Shin et al. (2021 bl which focused on ages > 1 year old.
aOnly four of the five cities measured PM2 5.
bAtkinson et al. (20141 primarily focused on single-day lag results.
cAdaret al. (20141 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
As discussed in Section 3.1.2.2.8, one approach to assessing the independence of the association
between exposure to PM2 5 and a health effect, such as short-term PM2 5 exposure and mortality, can be
examined is through the use of copollutant models. Appendix (Table A-l) to the 2019 PM ISA notes that
copollutant models are not without their limitations, such as instances where correlations are high
between pollutants resulting in greater bias in results. However, in assessing the results from copollutant
May 2022
3-65
-------
models a change in the PM2 5 risk estimate, after adjustment for a copollutant, may indicate the potential
for confounding.
At the time of the 2009 ISA, only a few studies had assessed the potential for confounding of the
PM2 5-mortality association by co-occurring pollutants. In contrast, the 2019 ISA included a number of
multicity studies that used copollutant models to evaluate this issue, including studies that examined both
gaseous pollutants and other particle size fractions. These studies reported that associations were
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 with 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), 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 (i.e.,
residential infiltration factors and commuting 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 (Zanobctti 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.. 2013b). 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
May 2022
3-66
-------
PM ISA. Both studies assessed associations by season through stratified analyses in which the warm
season is defined as April-September and the cold season as October-March. In Shin et al. (2021a). 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 with the cold season
(0.6% [95% CI: -2.2, 4.1]) at lag 1, the main lag examined for PM25 and mortality within the study.
Across 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, defined as the redox-weighted average of O3 and NO2,
modify the PM2 5-mortality association. The authors examined the role of oxidant gases on the PM2 5-
mortality relationship because previous studies have shown that oxidant gases can deplete antioxidants in
the lung and increase permeability of the lung epithelium, and that oxidant gases may accelerate
photochemical aging of PM2 5, potentially changing its toxicity (Lavigne et al.. 2018). To assess whether
there is evidence of effect modification of the PM2 5-mortality relationship by oxidant gases, 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
third 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
second and third 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.
May 2022
3-67
-------
Ox lag 0
PM2.5 lag 0 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: 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.
As discussed in the 2019 PM ISA (Section 3.4, 2019 PM ISA), examining these exposure factors is
important in the context of interpreting health effects associations reported in epidemiologic studies
because they can affect the relationship between indoor and outdoor ambient PM concentrations and
between personal exposure to ambient PM and ambient PM concentrations.
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
May 2022
3-68
-------
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
2.25)11 increase in mortality across all CBS As, but as depicted in Figure 3-16 there is extensive
city-to-city variability in associations across the U.S.
a/in r,I^K ¦<-1-5% 0-1.5 to -1.0% ~ -1 to -0.5% D-0.5to0%
Effect Estimate (10 Mg/m ). n Q tQ a50/o n 0 5 to Oo/o H 1.0 to 1.5% ¦ > 1.5%
Source: Baxter et al. (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).
In the second-stage analysis, the authors conducted both a univariate and multivariate
meta-regression. In the univariate regression, mortality associations larger in magnitude were observed
for CBSAs with larger homes, more heating degree days, and a higher percentage of homes heating with
oil, while cities with more gas heating had smaller associations. Across all univariate analyses, no
1795% CIs were not presented in this study.
May 2022
3-69
-------
individual factor explained much of the heterogeneity as reflected by R2 < 1%. For the multivariate
model, a backward selection approach was used to develop the final model that included variables for gas
heating use, heating degree days, cooling degree days, and variables for home size and age. Compared
with the univariate models, the multivariate models explained a larger amount of the heterogeneity in
mortality associations across the CBSAs examined, ranging from 11% to 13%. Overall, the results of
Baxter etal. (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
May 2022
3-70
-------
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,
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 global 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.
United States
Canada
0
1
£ "
0 10 20 30 40 50
PM;. £ CMcenifabom lujs/m'l
PMj j Ccncentratans njc^m )
Source: Adapted from Liu et al. (20191..
Note: As noted in Liu et al. (20191. the "y-axis can be interpreted as the relative change from the mean effect of PM25 on mortality;
the fraction of the curve below zero denotes a smaller estimate compared with the mean effect."
Figure 3-17 Concentration-response curves for the United States (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
May 2022
3-71
-------
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 examined in Lavigne et al. (2018). C-R curves support a linear relationship at PM2.5
concentrations often experienced in the U.S. and Canada, with less certainty in the shape of the curve for
nonaccidental mortality at concentrations below approximately 5 (ig/m3 as reflected by the lower bound
of the 95% confidence interval (CI) going below the null and some evidence of nonlinearity in the
respiratory mortality C-R relationship as reflected by the inflection point occurring around 7 (ig/m3
(Figure 3-18). However, compared with nonaccidental mortality, for both cardiovascular and respiratory
mortality, the lower bound of the 95% confidence interval was wider and below the null resulting in less
confidence in the overall shape of the C-R curve for both mortality outcomes.
Source: 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
May 2022
3-72
-------
evaluated in previous assessments, neither study conducted systematic evaluations of alternatives to
linearity.
3.2.1.3. Recent Epidemiologic Studies Examining the PIVh.s-Mortality
Relationship through Accountability Analyses and Alternative Methods
for Confounder Control
Within the 2019 PM ISA, in assessing the relationship between short-term PM2 5 exposure and
mortality several epidemiologic studies were evaluated that employed alternative methods for confounder
control (referred to as causal modeling methods in the 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 alternative methods for confounder control, which further inform the relationship
between short-term PM2 5 exposure and mortality (Table A-6).
Epidemiologic studies that use alternative methods for confounder control 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. 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
May 2022
3-73
-------
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
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 nonlinear 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.. 202 lb).
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 study that employed alternative methods for confounder control, using data
from the National Center for Health Statistics in 135 U.S. cities, Schwartz et al. (2018a) utilized three
May 2022
3-74
-------
statistical 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 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. If no such confounders exist, then control for the negative exposure would be expected to have
no change in the association between the exposure and outcome, which indicates no confounding by any
measured or unmeasured variables. 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/m3. 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 alternative methods for confounder control 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 alternative
methods for confounder control 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 PM25 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
May 2022
3-75
-------
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
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 alternative methods
for confounder control 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 3.2.2.2).18 In addition, with the expansion of in epidemiologic studies
that used statistical approaches that attempt to more extensively account for confounders and are more
robust to model misspecification (i.e., used alternative methods for confounder control), 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. Finally, 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 PM25 exposure and
18 Throughout this section, as detailed in the Preface of the 2019 PM ISA (Section P.3.2.2), risk estimates from
epidemiologic studies examining long-term exposures are for a 5 |ig/m3 increase in annual concentrations, unless
otherwise noted.
May 2022
3-76
-------
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
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.
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 (2019 PM ISA, Section
11.2.1). 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 PM2 5 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. In summary, 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
May 2022
3-77
-------
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.
May 2022
3-78
-------
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
Section 10.2.5.1
Mean across
studies:
4.1-17.9
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).
May 2022
3-79
-------
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
Note: This table corresponds to Table 11-8 in the 2019 PM ISA.
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.
May 2022 3-80
-------
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 (Pi 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.
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., 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
May 2022
3-81
-------
long-term PM2 5 exposure on mortality that is not overtly influenced by or is 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 with
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.
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
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 in assessments, which 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
May 2022
3-82
-------
long-term PM2 5 exposure on life expectancy (Section 3.2.2.2.4). examined potential copollutant
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 3.2.2.2.7). The following sections present an evaluation of 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. Study-specific details (e.g., study population, exposure assessment approach, confounders
considered) for the epidemiologic studies evaluated in this section are presented in Appendix A (Table
AiT).
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.
May 2022
3-83
-------
Reference
tPopeet al. (2014)
tLepeule et al. (2012)
tThurston et al. (2015)
Zeger et al. (200S)
Zeger et al. (2O0S)
Zeeer et al. (2008)
Eftim et al. (200S)
tDiet al. (2017)
tDi et al. (2017)
tDiet al. (2017)
fKioumourtzoglou et al. (2016)
tPun etal. (2017)
tShi et al. (2015)
tShi et al. (2015)
tShi et al. (2015)
tShi et al. (2015)
fWang etal. (2017)
tWang etal. (2017)
Lipfert et al. (2006)
Goss et al. (2004)
tCrouse et al. (2012)
tCrouse et al. (2012)
tChen et al. (2016)
tCrouse et al. (2015)
fWeichenthal etal. (2014)
tWeichenthal etal. (2014)
tPinault et al. (2016)
tLipsettetal. (2011)
fOstroetal. (2010)
tOstroetal. (2010)
tOstroetal. (2015)
tPuettet al. (2009)
tHartet al. (2015)
tHartet al. (2015)
tPuettet al. (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
N1H-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
Ag 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
expel 2 fig m3
nearest monitor
mutual adj
exp <10 ^g m3. mutual adj
no mutual adj
exp <10 |ig m3. no mutual a
exp<12 ug m3
satellite data
monitor data
more precise exp
within 30 km
within S km
nearest monitor
spatio-temp. model
foil model
C\TH-Resp
krisjns
IDW
closest monitor
Follow-up
1982-2004
1974-2009
2000-2009
2000-2005
2000-2005
2000-2005
2000-2002
2000-2012
2000-2012
2000-2012
2000-2010
2000-200S
2003-2008
2003-200S
2003-2008
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.S)
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
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
~
k
0.8
1.0 U 1.4 1.6
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; PM2.s = 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 pg/m3 increase in pollutant concentration.
Circles represent point estimates; horizontal lines represent 95% confidence intervals for PM25. Due to precise confidence intervals
for estimates from some studies, the lines representing the confidence intervals cannot be viewed behind the point representing the
effect estimate. 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. (2014) 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.
May 2022
3-84
-------
Study
Cohort
Follow-up
Mean
Exposure
*Eum et al. (201S)
Medicare
2000-2012
11.65
1-year avg -
base model
•
1-year avg
residual model
•
*Pope et al. (2019)
NHIS - Full Cohort
1986-2015
10.7
17-year avg
~
NHIS - Sub-Cohort
19S6-2015
10.7
17-year avg
-~
*Lefler et al. (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
~
*Wu et al. (2020)
Medicare
2000 - 2016
9.8
1-year avg -
Cox
•
*Wu et al. (2020)
Medicare
2000 - 2016
9.8
1-year avg -
Poisson
•
~Elliot etal. (2020)
NHS
1988 -2008
13.7
2-year avg
*Wang etal. (2020)
Medicare
2000 - 2008
10.55
1 -year avg
•
*Pappin et al. (2019)
CanCHEC
1991-2016
6.6S - 7.95
3-year avg.
-year lag
•
* Zhang etal. (2021)
Ontario Health Study
2009-2017
7.8
5-year avg.
-year lag
*Pinault et al. (2017)
CanCHEC
1991-2011
7.4
3-year avg.
-year lag
~
*Crouse et al. (2020)
CanCHEC
2001-2011
7.98
8-year avg.
-year lag; 1 km
~-
121
8-year avg,
-year lag; 5 km
~
6.9
8-year avg,
-year lag; 10 km
~
7.43
3-year avg.
-year lag; 1 km
~
6.79
3-year avg.
-year lag; 5 km
~
6.44
3-year avg.
-year lag; 10 km
7.21
1-year avg.
-year lag; 1 km
-•
6.59
1-year avg,
-year lag; 5 km
~
6.24
1-year avg,
-year lag; 10 km
~
* Chris tidis etal. (2019)
mCCHS
2000-2012
5.9
3-year avg,
-year lag
~
Xonaccidental
1.00 1.20 1.40 1.60
Hazard Ratio (959 6 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. Due to precise confidence intervals for estimates from some studies, the lines representing the
confidence intervals cannot be viewed behind the point representing the effect estimate. 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 United
States 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 to 2014 with
mortality data through 2015 resulting in approximately 1.5 million participants. The authors also formed a
subcohort comprising 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
May 2022
3-85
-------
a sensitivity analysis, back-casted PM25 concentrations were imputed from 1988 to 1998, allowing for the
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 effect 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, Lefleretal.
(2019) subtracted the minimum PM2 5 concentration identified within the defined circular buffers around
each census tract for each year from 2000 to 2015. When focusing on results presented for an IQR
increase in PM2 5 concentrations, positive associations were reported for each exposure indicator;
however, there was some variability across each of the exposure indicators examined with the regional
indicator being closer in magnitude to the primary PM2 5 exposure indicator (Figure 3-21). Overall, these
results indicate that the mortality risk attributed to PM2 5 is not due solely to regionally or locally derived
PM25.
May 2022
3-86
-------
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: Lefleret 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 to 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 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
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 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).
1.00 1.02 1.04 1.06 1.08 1.10 1.12
HR per IQR
May 2022
3-87
-------
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 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 Harvard 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) (Braucr 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 Harvard
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 (Pi
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 Harvard Medicare study primarily focus on examining alternative methods for confounder
control and are evaluated in Section 3.1.2.3. However, a recent study by Wu et al. (2020a) builds on 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 alternative methods for confounder control (Section 3.1.2.3). and the sensitivity of
associations to different confounder adjustment (Section 3.1.2.2.5V 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 Harvard Medicare study (Dominici et al.. 2019). MAPLE studies
rely 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
May 2022
3-88
-------
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 are 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). Finally, building on 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 in which 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
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
May 2022
3-89
-------
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
statistical 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 with 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
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
May 2022
3-90
-------
with 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 to 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 on 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.
May 2022
3-91
-------
Studv
tLepeuleetal. (2012)
tOsWtal. (2015)
tLipsett et al. £2011)
tHart et al. (2011)
'Hayes et al. (2020)
' a etal. (2015)
t Garcia e
112)
tPopeet al. (2014)
tChenet al. (201.,
•Popeet al. (20195
*Wang etal. (2020)
tWeichenthal etal. (2014)
*Chenetal. (2020)
'Zhang etal. (2021)
*Crouse et al. (2020)
*Pinault et al. (2017)
•Pinault et al. (2018)
*Pinault et al. (2018)
fPinault et al. (2016)
tLipsett etal. (2011)
tHart etal. (2011)
tThurstonetal. (2015)
tTurner et al. (2016)
tPunetal. (2017)
~Wang etal. (2020)
tCrouseet al. (2013)
*Zhane etal. (1021)
*Crouse et al. (2020)
fPinault et al. (2QJ7)
rPinault et al. (201o)
Cohort
Harvard Six Cities
CA Teachers
CA . sachsrs
TrIPS
TrIPS
NIH-AARP
CA cohort
\CS
i&m
NHIS
Medicare
Ag Health
Ontario Health Study
CanCHEC
CanCHEC
CanCHEC
Medicare
-'•-edlliT:
CanCHEC
Ov.:ir:: .-faith s:udv
CanCHEC
c¥®c
2001-2007
2000 - 2005
19S5 - 2000
-:...
1995 -2011
2006
1982 - 2004
mm
1999 - 2011
1986-2015
2000 - 2008
1993 - 2009
2000-2008
2000 - 2008
1991-2006
2009 - 2017
2001-2011
Mean
11.4-23 6
17.9
15.5
14.1
3.06
2.94
2.68
2-S
::
.0.7
0.7
10.55
9.44
7.S
7.43
17.9
Hf
15.6
14.1
10.55
8.9
M
Ml Cardiovascular
srs
Nearest monitor
^lore j>r|cise exposure
Linear model
Nonlinear model
LUR-BME
Near-Source
i'.rrijr.il
All Respiratory
1.20 1.40 1.60 1.80
Hazard Ratio (95% Confidence Interval)
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. Due to precise confidence intervals for estimates from some studies, the lines representing the confidence
intervals cannot be viewed behind the point representing the effect estimate. 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 ofthe 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
Studies investigating cardiovascular mortality provided some of the strongest evidence for a
cardiovascular effect related to long-term PM2 5 exposure in the 2009 PM ISA, which was further
supported by studies evaluated in the 2019 PM ISA (Figure 3-22). Generally, across the cohort studies
evaluated in the 2019 PM ISA, most of the PM2 5 effect estimates relating long-term PM2 5 exposure and
cardiovascular mortality remained relatively unchanged or increased in magnitude in copollutant models
adjusted for ozone, NO2, PM10-2.5, or SO2. The results of recent cohort studies provide additional evidence
for associations with cardiovascular mortality outcomes across the distribution of PM2 5 concentrations,
the potential implications of comorbidity on the PM2 5-cardiovascular mortality relationship, and
associations with individual cardiovascular mortality outcomes.
May 2022
3-92
-------
In a study of the NIH-AARP cohort, which consists of participants ranging from 50 to 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 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 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 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., sulfate, 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 with 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 for whom, in addition to having cardiovascular disease as the primary cause of death,
diabetes was mentioned on the death certificate, whereas for the mCCHS, which consisted of
individual-level data, participants self-reported diabetes status by noting insulin or medication use to
May 2022
3-93
-------
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 with
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 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 (PM25: HR= 1.12 [95% CI: 1.10, 1.14]; PM25 + ozone: HR= 1.06 [95% CI: 1.04, 1.09]).
May 2022
3-94
-------
SS&8813)
BIW1
Popeet al. (2014)
Wane etal. (2Q2Q1
Croitee et al. (20l5)
"Grouse et al. (2020)
let aT(2f>l<
"hen et al. (2005)
uettetal. (2011)
ram
Popeet al. (2014)
Crouse et al. (2012)
>settetaL£2Qll)
SlU: 8813
&kW>
Monitor within 30 km
Monitor within S km
^xcSjling^{ong-haul drivers
Nearest monitor
HF. cardiac arrest, related
AMI
Extended
8-year avg; 1 km
3-year avg; 1 km
1-year avg; 1 km
Mer
Extended
S -year avg;
3-year avg
1-year avg;
cmc
Hypertension
Circulatory
CB\D
1.00 1.20 1.40 1.60 1.80 2.00
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. Due to precise confidence intervals for estimates from
some studies, the lines representing the confidence intervals cannot be viewed behind the point representing the effect estimate.
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. (20141 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.
Few studies to date have examined associations between long-term PM2 5 exposure and
cerebrovascular disease or stroke mortality, but studies by Wang et al. (2020). Pinault et al. (2017).
Crouse et al. (2020). and Haves et al. (2020) along with a study within the NHIS cohort by Pope et al.
(2019) add to the growing body of evidence indicating a positive relationship. Although Pope et al.
(2014) in a study of the ACS cohort, evaluated in the 2019 PM ISA indicated a positive association with
CHF mortality, in a recent study of the Medicare cohort Wang et al. (2020) reported a null association.
Finally, a few studies using the CanCHEC cohort (Crouse et al.. 2020; Pinault et al.. 2017) examined the
combination of cardiovascular mortality with either metabolic-related or diabetes mortality, and found
that associations were similar in magnitude to cardiovascular mortality alone [e.g., within Pinault et al.
May 2022
3-95
-------
(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
with 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]).
May 2022
3-96
-------
Study
tLepeule et al. 2012
tHartet al. 2010
Cohort
Hansard Six Cities
TrIPS
Follow-up Mean
1974 - 2009 11.4-23.6
tTurneretal. 2016
ACS
1982 - 2004
12.6
12
*Pope et al. (2019)
NHIS
1999 - 2015
0.5
10.7
*Wang etal. (2020)
Medicare
2000 - 200S
10.55
f Crouse et al. 2015
CanCHEC
1991-2006
8.9
•Pinault et al. (2017)
CanCHEC
1998-2010
7.4
*Cakmak et al. (201S)
CanCHEC
1991-2011
3.S-7.
3 S - 7
tPinault et al. 2016
CCHS
199S-2011
6.3
tGan et al. 2013
Metro Vancouver
1999 - 2002
4.1
Whole cohort
Excluding long-haul drivers
LUR-BME
Near-Source
Regional
Chronic lower respiratory
Movers
Never movers
tTurner 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
Influenza pneumonia
*Wang etal. (2020)
Medicare
2000 - 200S
10.55
Pneumonia
*Pinault etal. (2017)
CanCHEC
1998-2010
7.4
Pneumonia
Respiratory Infection
0.80
1.00 1.20 1.40
Hazard Ratio (95% Confidence Interval)
1.60
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; |jg/m3 = micrograms per cubic meter; NHIS = National Health Interview Survey; PM25 = particulate matter with a
nominal mean aerodynamic diameter less than or equal to 2.5 |jm; 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 |jg/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. (20141 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.
The examination of respiratory infection mortality is more limited, with recent cohort studies
examining either pneumonia alone (Wang et al.. 2020; Pinault et al.. 2017) or the combination of
influenza and pneumonia (Pope et al.. 2019). Across the studies, which employed different approaches to
assign PM2.5 exposures including a 1-year average in the Medicare cohort (Wang et al.. 2020). a 3-year
average with a 1-year lag in the CanCHEC cohort (Pinault et al.. 2017). and a 17-year average in the
NHIS cohort (Pope et al.. 2019). each reported positive associations with the magnitude of the association
increasing as the length of the exposure window increased.
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
May 2022
3-97
-------
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 HF ("Ward-Caviness et al.. 2020) and previous 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
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
May 2022
3-98
-------
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.
May 2022
3-99
-------
Female
Male
Life expectancy loss (years)
Source: Bennett et al. (2019)
Figure 3-25 Estimated loss in life expectancy by county for females and males
for PM2.5 concentrations in excess of the lowest observed PM2.5
concentration of 2.8 pg/m3.
3.2.2.2.5. Potential Copollutant Confounding of the PM2.5-Mortality Relationship
As discussed in Section 3, J .2.2.8. one approach to assessing the independence of the association
between exposure to PM2.5 and a health effect, such as long-term PM2.5 exposure and mortality, can be
examined is through the use of copollutant models. The Appendix (Table A-l) to the 2019 PM ISA notes
that copollutant models are not without their limitations, such as instances where correlations are high
between pollutants resulting in greater bias in results. However, in assessing the results from copollutant
models a change in the PM •3 risk estimate, after adjustment for a copollutant, may indicate the potential
for confounding.
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, PSfis-ttft S() % 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, PM^s, SO2, or benzene. Recent North American cohort studies conducted additional
analyses that further inform whether the relationship between long-term PMaj exposure and mortality is
confounded by gaseous pollutants or other particle size fractions (i.e., PM10-2 5).
May 2022
3-100
-------
In an analysis of the NHIS subcohort, Lefler et al. (2019) reported that the PM2 5-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 etal. (2020) regressed 12-month PM2 5 on NO2 and used the residuals as the
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.
May 2022
3-101
-------
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.
3.2.2.2.6. Studies That Address the Potential Implications of Unmeasured
Confounders on PM2.5-Mortality Associations
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 unmeasured confounders 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 Grevenetal. (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 to 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, which is analogous to "national" and "local" respectively 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
May 2022
3-102
-------
magnitude of this bias. It is important to note that the "temporal" and "spatiotemporal" coefficients are
not directly comparable to the results of other epidemiologic studies 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 to 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
there was evidence of unmeasured confounding. Consistent with the previous studies larger associations
were reported for the "temporal" compared with 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 that used data for the entire 13 year period and shorter periods ranging from 3 to 12 years
(e.g., 2001-2012, 2009-2012) 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), for which 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,
May 2022
3-103
-------
HRs were relatively unchanged compared with 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.
Length of study period (years)
Source: Eum et al. (2018)
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
May 2022
3-104
-------
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
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 with 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. Finally, 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 nonaccidental, cardiovascular, IHD, and lung cancer mortality, the authors reported an
overall reduction (i.e., bias) in the HRs compared with 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.
May 2022
3-105
-------
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.
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 with 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], Although analyses of the Harvard Six Cities study (Lepeule et al..
2012) and the U.S. Medicare cohort (Di et al.. 2017b; Shi et al.. 2015) reported linear, no-threshold C-R
relationships down to 8, 6 and 5 (ig/m3, respectively, 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
(e.g., analyses of the CanCHEC (Pinault et al.. 2016)).
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
May 2022
3-106
-------
for total (nonaccidental) mortality, though there is some evidence for a sublinear (shallower slope at lower
concentrations and steeper slope at higher concentrations) (Zhang et al.. 2021; Pope et al.. 2019) or
supralinear (steeper slope at lower concentrations and shallower slope at higher concentrations)
(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 PM25 NAAQS of 12 (ig/m3.
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 et al. (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
5.9 (0.4-17.2)
Statistical Analysis
Summary
C-R: SCHIF fits a class of flexible,
but monotonically nondecreasing
functions to select best fitting model
Supralinear at lower concentrations
(< 5 pg/rn3)
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, three-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 SCHIF to the MISS predictions
Supralinear at lower concentrations
(< 5 pg/rn3)
May 2022
3-107
-------
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.
Statistical Analysis
Study Location—Cohort Exposure; PM2.5 Mean;
Table/Figure from Reference Range in |jg/m3 Summary
Pinault et al. (2017)
CanCHEC
(Figure 2; Table S4)
Supralinear at lower concentrations
(< 5 |jg/m3); HRs remained positive
and statistically significant in the two
lowest cutpoint categories with
highest HRs for the 0-5 |jg/m3
category, consistent with the
supralinear C-R function
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: SCHIF fits a class of flexible,
but monotonically nondecreasing
functions to select best fitting model
(counterfactual is 0 |jg/m3); cutpoint
analyses: 0-5, 5—10, > 10 |jg/m3
Pope et al. (2019) Population-weighted annual PM2.5 C-R: Integrated model that fit a class
NHIS Cohort concentrations averaged for of flexible, but monotonically
(Fiqure 4) census-tract centroids. nondecreasing functions to select
10.7; (2.5-19.2) best fitting model.
Generally linear, though some
evidence of a shallower slope at
lower concentrations (< 8 |jg/m3)
Wang et al. (2020)
Medicare Cohort
(Figure 1; Table S4)
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
Daily PM2.5 was estimated on a 6-km C-R: RCS model with 3 knots;
grid using a spatiotemporal model Threshold: PM2.5 < 8, < 10, <
10.3 (NR) 12 pg/m3
Ward-Caviness et al. (2020) Nearest monitor (Threshold model) or Threshold model (PM2.s< 12 |jg/m3)
HF Patient Cohort Harvard's 1 km * 1 km modeled C-R: limited to PM2.5 concentration
(Table 2' Fiqure 3) PM2.5 surface (C-R figure); within inner 95% of distribution
10.3; (8-14) (8-14 |jg/m3)
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
May 2022
3-108
-------
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
Wu et al. (2020a)
Medicare Cohort
(Figure 3)
Exposure; PM2.5 Mean;
Range in |jg/m3
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
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
Zhang et al. (2021)
Ontario Health Study
(Figure 2; Tables S3, S8)
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)
C-R: 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 = Canadian Community Health Survey—mortality cohort; 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 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
May 2022
3-109
-------
concentration-response relationship at relatively low PM^s concentrations (< 5 Lig/nr ) (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/nr. Analyses of exposure categories (i.e., quartiles) by Zhang et
al. (2021) provides additional support for a sublinear concentration-response relationship. A similar
sublinear relationship was reported by Pope et al. (2019) for a U.S. cohort (Figure 3-28).
Pooled SCHIF
0 5 10 15
PMjr 5 - MQ/m3
Source: Pappin et al, (2019)
Note; Uncertainty bounds are displayed as gray shaded area. The uncertainty bounds are anchored at zero because the logarithm
of the hazard ratio is fixed at zero and its associated standard error is also set at zero.
Figure 3-27 Shape Constrained Health Impact Function predictions by PM2.5
concentration for the pooled CanCHEC cohort.
May 2022
3-110
-------
5
-*r-
10
15
~f
20
Source: 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
May 2022
3-111
-------
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 (i.e., the
confidence intervals become relatively wide and in some instances the lower 95% CI crosses the null at
this concentration). Generally, the evidence remains consistent in supporting a no-threshold relationship,
and in supporting a linear relationship for PM2 5 concentrations > 8 (ig/m3. However, uncertainties remain
about the shape of the C-R curve at PM2 5 concentrations < 8 (ig/m3, with some recent studies providing
evidence for either a sublinear, linear, or supralinear relationship at these lower concentrations.
3.2.2.3. Recent Epidemiologic Studies Examining the PM2.5-Mortality
Relationship through Accountability Analyses and Alternative Methods
for Confounder Control
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 alternative
methods for confounder control (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 alternative methods for
confounder control 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.
May 2022
3-112
-------
Table 3-7 Description of methods from epidemiologic studies using accountability analyses or alternative
methods for confounder control to examine long-term exposure to PM2.5 and mortality.
Study/Location/Population
(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 35.4 (95% CI: 33.4, 37.6)
exposed to the observed concentration level, accounting for all measured potential confounders excess deaths per 10 million
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.
person-days for each 1 |jg/m3
increase in annual PM2.5
concentrations
May 2022
3-113
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied alternative methods for
confounder control in examining long-term PM2.5 exposure and mortality.
Study/Location/Population
(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
MoonRF: 1,058 (95% CI: 988,
1,127) annual early deaths for
a 1 |jg/m3 increase in annual
PM2.5 concentrations
MoonRF: 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.
May 2022
3-114
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied alternative methods for
confounder control in examining long-term PM2.5 exposure and mortality.
Study/Location/Population
(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 increase 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.
May 2022
3-115
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied alternative methods for
confounder control in examining long-term PM2.5 exposure and mortality.
Study/Location/Population
(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 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.
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)
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: 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 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 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).
May 2022
3-116
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied alternative methods for
confounder control in examining long-term PM2.5 exposure and mortality.
Study/Location/Population
(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)
less 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
Hiabee 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 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.
For a 10 |jg/m3 increase in
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)
May 2022
3-117
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied alternative methods for
confounder control in examining long-term PM2.5 exposure and mortality.
Study/Location/Population
(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 to 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).
May 2022
3-118
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied alternative methods for
confounder control in examining long-term PM2.5 exposure and mortality.
Study/Location/Population
(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 effect 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
May 2022
3-119
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied alternative methods for
confounder control in examining long-term PM2.5 exposure and mortality.
Study/Location/Population
(Cohort)/Years Statistical Method Results
DID: Used a Poisson survival analysis using the Anderson-Gill formulation with time-varying 4.04% (95% CI: 3.49, 4.59)
covariates. The data were randomly split into subsets due to computational limitations. The effect increase in mortality rates for
estimates were then pooled using a fixed-effect meta-analysis. an IQR (3 |jg/m3) increase in
annual PM2.5 concentrations
Medicare
2000-2013
Schwartz et al. (2021)
U.S.
Medicare
2000-2016
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: Applied the standard approach for continuous predictors. The mortality rate in a ZIP code
given 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 nonlinear 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.
Probability of dying in each
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
May 2022
3-120
-------
Table 3-7 (Continued): Description of methods from epidemiologic studies that applied alternative methods for
confounder control in examining long-term PM2.5 exposure and mortality.
Study/Location/Population
(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.
Second, 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
Ziqler et al. (2018)
Eastern U.S.
Medicare
2000-2012
Accountability and Propensity Score: To examine whether attainment status for the 1997
NAAQS led to an improvement in PM2.5 concentrations and subsequently health, the authors
employed propensity scores, within a spatial hierarchical regression model to examine whether
designation of nonattainment, for either the annual standard of 15 |jg/m3 or the daily standard of
65 |jg/m3, made in 2005 for the 1997 PM NAAQS led to a corresponding reduction in ambient
PM2.5 concentrations and all-cause mortality. In the first step, propensity scores were used to
adjust for confounders by grouping attainment and nonattainment locations based on similarities
of baseline characteristics, including air pollution monitoring data, population demographics,
meteorological data, baseline Medicare characteristics, and PM2.5 and mortality. In the second
step, a spatial hierarchical regression model was used to predict the potential ambient PM2.5
concentration in 2010-2012 that would have occurred in nonattainment areas if the designations
had never occurred. The model used in this step specifies linear adjustment terms for propensity
score group indicators and several specific covariates, each including an interaction with the
indicator of attainment status, while including a spatial random effect that accounts for the
similarity of ambient air quality at nearby locations. For the analysis, the spatial hierarchical
model is estimated jointly with a log-linear model for all-cause mortality, with the same
confounding adjustment for propensity score group and additional covariates. In addition to
estimating the effect estimates for the overall average effects, a principal stratification approach
was used to estimate "associative effects" and "dissociative effects."
Overall average effect: a
reduction in the all-cause
mortality rate by 1.25 (95%
CI: -2.63, 0.11) deaths per
1000 beneficiaries
Average associative effects: a
reduction in the all-cause
mortality rate of 3.16 (95% CI:
-5.19, -1.21) deaths per 1000
beneficiaries
Average disassociative
effects: Quantitative results
not presented
DID:
IRR :
difference-in-difference; GPS = generalized propensity score; HR = hazard ratio; IPTW =
incidence rate ratio; IV = instrument variable; moonRF = m-out-of-n random forests.
inverse probability treatment weighting; IPW = inverse probability weighting;
May 2022
3-121
-------
Wei et al. (2020). Wei et al. (2021b). and Wu et al. (2020a) implemented different 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 nonlinear 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
that the three GPS approaches yielded similar results with a 10 |ig/m3 increase in annual PM2 5
May 2022
3-122
-------
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) forthe 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 (8 < PM2 5 <10 |ig/m3) to
low concentration (PM2 5 < 8 |ig/m3). When comparing high levels of PM2 5 concentrations (PM2 5 >
10 |ig/m3) to low concentrations (PM25< 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 alternative methods for confounder control 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 (Higbee 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 and 2013 by applying an IPW survival
model. The estimated mean age at death for a population with an annual average PM2 5 concentration of
12 |ig/m3 was 0.89 (95% CI: 0.88, 0.91) years less than estimated for a counterfactual PM2 5
May 2022
3-123
-------
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 and 2016. This newer IPW approach used decile
binning, which divided PM2 5 concentrations into 10 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 NO2) and other covariates (Table A-8). The lowest decile group is treated as the reference
with effects estimated for the other decile groups compared with the reference. The RR 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 with the adjustment of copollutants and confounders, 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.. 2021a). 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
(Schwartz et al.. 2021; Yitshak-Sade et al.. 2019b). The predictors of the outcome, such as socioeconomic
May 2022
3-124
-------
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 and 2013 (Table A-8). After the release of the first annual PM25 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
and 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 to 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
decreased both monthly PM2 5 concentrations by 2.1 |ig/m3. and the monthly age-adjusted mortality by
May 2022
3-125
-------
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. (2019b) 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. (2019b) 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-8). 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.
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.
Zigler et al. (2018) used a hybrid approach of integrating an accountability analysis with an
alternative method for confounder control to examine whether attainment status for the 1997 NAAQS led
to an improvement in PM2 5 concentrations and subsequently health, as previously discussed in
Section 3.1.2.3. By focusing on nonattainment designations, the authors are able to examine the role of
local control strategies in reducing PM2 5 concentrations that occurred above and beyond reductions due
to regional strategies. Within this study, Zigler et al. (2018) employed propensity scores, within a spatial
hierarchical regression model to examine whether designation of nonattainment in 2005 for the 1997 PM
May 2022
3-126
-------
NAAQS (see Table 3-7). for either the annual standard of 15 (ig/m3 or the daily standard of 65 (ig/m3, led
to a corresponding reduction in ambient PM2 5 concentrations and all-cause mortality rates among
Medicare beneficiaries in the eastern U.S. from 2009 to 2012. Zigleretal. (2018) reported a reduction in
the overall average effect for all-cause mortality across the nonattainment areas of 1.25 (95% CI: -2.63,
0.11) deaths per 1000 beneficiaries, which is similar in magnitude and precision to the average
"dissociative effect" (i.e., the effects of the nonattainment designations estimated to not affect PM2 5
concentrations by more than ±1 (ig/m3). When examining the average "associative effect" (i.e., the effects
of the nonattainment designations on mortality rates among areas where the nonattainment designations
are estimated to reduce ambient PM2 5 concentrations by at least 1 (.ig/ni3). the authors reported a reduction
in the all-cause mortality rate that was larger than the "dissociative effect" (-3.16 [95% CI: -5.19, -1.21]
deaths per 1000 beneficiaries). Collectively, the results of Zigler et al. (2018) provide evidence that
reductions in ambient PM2 5 concentrations and all-cause mortality could not be conclusively attributed to
nonattainment designations against the backdrop of other regional strategies that impacted the eastern
U.S.
Overall, recent epidemiologic studies employed a variety of alternative methods for confounder
control such as GPS, IPW, and DID and reported consistent results among large study populations across
the U.S. These alternative methods for confounder control in combination with accountability analyses
reduce uncertainties related to confounder bias and further informs the relationship between long-term
PM2 5 exposure and total mortality and supports the conclusions of the 2019 PM ISA.
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
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.
May 2022
3-127
-------
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.
Finally, a number of recent studies employed alternative methods for confounder control 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 for confounder control
in combination with accountability analyses that examined the effect 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 in 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
PM25 exposure and health risk disparities among racial and ethnic groups and SES (Section 3.3.3).
May 2022
3-128
-------
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 (.ig/ni'). As discussed
in Sections 6.1.10 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 that had an average PM2 5
concentration of 24 (ig/m3 or particle-filtered urban street air while at rest. Notably, 27 of these
participants had never smoked, while 33 former smokers had not smoked on average 20 years prior to
study participation. The study was a randomized, repeated measures, single blinded cross-over study.
Decreased vasomotor function was reported immediately (within 1 hour) after exposure when comparing
nonfiltered with particle-filtered air. 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. In the
companion study, Hemmingsen et al. (2015a) (2019 PM ISA, Section 6.1.11) observed no changes in
blood biomarkers of oxidative stress or inflammation. Similarly, blood pressure, blood lipids, and
metabolic biomarkers were unaffected by exposure. With respect to HRV, Hemmingsen et al. (2015b)
found the high frequency domain was statistically significantly decreased and the low frequency domain
was statistically significantly increased when nonfiltered street air was compared with particle-filtered
street air. In addition, the standard deviation of NN intervals (SDNN) was statistically significantly
reduced after first entering the nonfiltered chamber, but this effect did not persist. In addition,
(Hemmingsen et al.. 2015a') (2019 PM ISA, Section 10.2.2.2) found no evidence of oxidative stress or
DNA damage in peripheral blood monocytes of participants exposed to unfiltered street air.
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 (ig/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. When
comparing mean values from the exposed and unexposed groups, changes in lung function were observed
in PM2 5 exposed participants, including a statistically significant decrease in forced expiratory volume in
one second/forced vital capacity (FEV1/FVC) ratio of 1.2% at 1-hour postexposure that returned to
baseline by 20 hours postexposure. Decreases in peak expiratory flow (PEF) (1.8%) and FEV1 (0.8%)
were also observed when comparing mean values from the exposed and unexposed groups at 1-hour
postexposure, but they did not achieve statistical significance. Furthermore, when comparing mean values
from the exposed and unexposed groups, 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
May 2022
3-129
-------
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 when comparing differences in mean values were increases in markers of vascular
inflammation, soluble intercellular adhesion molecule 1 (sICAM) (10.7%) and soluble vascular cell
adhesion molecule 1 (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 time domain measures of HRV in the study population as a whole, but SDNN was statistically
significantly decreased in men and increased in women. No statistically significant results with respect to
frequency domains were reported. 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 with
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 conducted at higher PM2 5 concentrations 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.
3.3.2. PM2.5 Exposure and SARS-CoV-2 Infection and COVID-19 Death
With the advent of the global COVID-19 pandemic, several studies have emerged that evaluate
the relationship between ambient air pollution, specifically PM2 5, on severe acute respiratory syndrome
coronavirus 2 (SARS-CoV-2) (the virus responsible for COVID-19 disease) infections and COVID-19
deaths, including a few studies with locations in the U.S. and Canada. The following sections present an
evaluation of studies that examined the relationship between short-term (Section 3.3.2.1) and long-term
(Section 3.3.2.2) PM25 exposure and outcomes including SARS-CoV-2 infection and replication rate as
May 2022
3-130
-------
well as COVID-19 hospital admissions 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 that 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 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 PM25 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
etal.. 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 SARS-
CoV-2 infections and COVID-19 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 ("NYC. 2022). 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 SARS-CoV-2
infections and COVID-19 deaths, controlling for the lagged outcome (to account for potential
autocorrelation of the time-series and new cases) and a trend for day. Using a 21-day moving average of
PM2 5 exposure, the authors identified a null association between PM2 5 concentrations and increased risk
of SARS-CoV-2 infection (IRR: 0.02 [95% CI: 0.01, 0.02] or COVID-19 death [IRR: 0.32 (95% CI: 0.10,
0.97)]).
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; Wuet
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
May 2022
3-131
-------
(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 zero cases. The 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 ecological 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
and July 17, 2020. At that point in the pandemic, many areas of the U.S. were still reporting zero cases
(USAFacts. 2022). This study also relied on negative binomial regression and 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).
Case-fatalityt
ta
CO
1 g
I "
O 5
O (0
>5
8 a
£ r
W QJ
CD
£ £
JZ O
o C
20
10
-10
Single-pollutant
Air Pollutants
Bi-pollutant ' Tri-pollutant
NO2 Ozone
e a>
o o
2 S
O <0
>5
S 8.
Mortality
40
30
20
.£ 10
£ %
O) O
o —
-10
I
I
Single-pollutant
Bi-pollutant
Tri-pollutant
Source: Liang et al. (2020)
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).
May 2022
3-132
-------
In contrast, a Canadian ecological study by Stieb et al. (2020) evaluated SARS-CoV-2 infections
within 111 Canadian Health Regions through May 18, 2020. These data were used to estimate the IRR of
SARS-CoV-2 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
SARS-CoV-2 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 SARS-CoV-2 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 SARS-CoV-2 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 with those without COPD or asthma (OR: 0.42 [95% CI: 0.12,
1.54]).
In a departure from examining solely SARS-CoV-2 infections, deaths, or COVID-19
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) 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 (AR0: 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 SARS-CoV-2 infections and
COVID-19 deaths reported positive associations, a number of methodological issues could 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. In addition, recent investigations have noted important differences in COVID-19-related
May 2022
3-133
-------
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 that 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 preventive measures are important for slowing the spread of SARS-CoV-2. 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 whether 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
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
documented 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
May 2022
3-134
-------
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 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 health
care in the U.S., therefore indicating that SES may underlie differential risk for PM2 5-related health
effects. Measures of SES can be examined on the individual level (e.g., personal income, education,
occupation, etc.), at a community-level (e.g., median household income per census tract, percent of
population with bachelor's degrees), or as a "composite" metric incorporating several different SES
measurements into one single score or metric. 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 an indicator 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 with higher SES populations. The following sections present an evaluation of recent studies
pertaining to both PM25 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).
3.3.3.1.1. Exposure Disparity
Several recent studies within the U.S. and Canada evaluated the relationship between PM2 5
exposure and community-level 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 et al.. 2019;
Rosofskv et al.. 2018; Han et al.. 2017). These recent studies add to the initial conclusions drawn within
the 2019 ISA, which include that there is disparate PM2 5 exposure among lower SES communities. When
considered together, on average, these additional studies provide further evidence that lower SES
communities are exposed to higher concentrations of PM2 5 compared with higher SES communities
(Figure 3-30a. Figure 3-30b—SES exposure, Table A-11). Specifically, Figure 3-30a and Figure 3-30b
compare the mean exposure of low SES populations with that of higher SES populations, with a ratio
value > 1 indicative of higher PM2 5 exposure among the group with lower SES.
May 2022
3-135
-------
Educational attainment is a common metric with which to assess the SES of a community. A
study by Lee (2019) in the state of California observed that in 2016, census block-groups with a high
percent (> 75th percentile) of low educational attainment had a higher exposure to PM2 5 (9.90 (.ig/ni3)
compared with those with a low percent (< 25th percentile) of low educational attainment (8.46 (.ig/ni3).
Rosofskv et al. (2018) conducted a study in Massachusetts that examined exposure differences by
community-level 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 a 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 PM25 concentrations ranging from 11.1 to 11.2 (ig/m3
in 2003 and 7.9 to 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
different communities in the study conducted by Rosofskv et al. (2018) noted above. Within this study it
was observed that mutually exclusive census block groups with median household income categories with
ranges < $75,000 ($50,000-$75,000; $35,000-$50,000; $20,000-$35,000; < $20,000) were exposed to
slightly higher PM2 5 concentrations of PM2 5 in 2003 (11.2-11.4 (.ig/ni3) and 2010 (8.0-8.2 (.ig/ni3)
compared with 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 with census tracts above the
75th percentile (11.2 (ig/m3) for median household income (Son et al.. 2020). Instead of examining
specific household income cut points, 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 with 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
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 with 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 versus, 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
PM25.
May 2022
3-136
-------
Several recent studies evaluated more complex measurements of community-level SES. These
measurements, or "composite" characteristics incorporated several different SES measurements into one
single score or metric and were applied 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 with ZIP codes with high SES characteristics (9.6 (.ig/ni3) (Han et al.. 2017).
Similarly, a study by Lee and Park (2020) in the state of California, compared vulnerable with less
vulnerable communities based on the Social and Health Vulnerability (SHV) metric within the
Environmental Justice Screening Method. The SHV score is on a scale between 1 and 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 (SHV score 1-2) (6.8 (ig/m3) had lower PM2 5 exposure, compared with communities with
higher vulnerability (SHV 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 with non-EJ communities
(10.3 (ig/m3). A study by Weaver etal. (2019) evaluated certain neighborhood clusters located within
three counties (Durham, Orange, and Wake) in Central North Carolina. In this study, six clusters were
derived using Ward's hierarchical clustering for 11 sociodemographic factors at the census block group
level including percent of the population: with at least a bachelor's degree, in owner occupied housing,
with income below the poverty level, on public assistance, who identify as Black, who identify as "other"
race (neither Black nor White), unemployed, in nonmanagerial positions, of households with a single
parent, and of vacant housing, and urban environment (Weaver et al.. 2019). 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. Finally,
Cluster 6 was similar to Cluster 5 but was more urban. When compared with 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 with Cluster 3 (11.9-12.8 (.ig/ni3). Overall, areas (clusters) with high relative SES disadvantage
were exposed to higher concentrations of PM2 5, compared with areas with high relative SES advantage.
May 2022
3-137
-------
Study
Population
Comparison
Group
Reference
Value
(|ig/m3)
fMikati etal (2018)
US: American Community Survey,
Not in poverty
,, , US: Census (tracts with PM Component . T , .
tBelletal (2012) Monitors) ^000 Not m poverty
fBravo et al (2016) US: Census (Eastern two-thirds), 2000 Not in poverty
All
Urban
Suburban
Rural
11.5
10.2
*Lee et al (2019)
US: American Community Survey
(California), 2016
<25th percentile % in poverty
Statewide 8.66
SF Bay Area 7.32
San Joaquin Valley 11.21
South Coast 10.16
tn , ,US: North Carolina State Center for
•Son et al (2020) ^ 20(p__20u
>75th percentile
50-75 tli percentile
25-50th percentile
<25th percentile
11.16
11.16
11.16
*Wang et al 2020 US: Medicare, 2000-2008
High
Medium
Low
10.71
10.71
*Rosofsky et al (2018),
US: Census and American Community
Survey (Massachusetts), 2003-2010
>$75,000
2003
-------
Study
Population
Comparison
Group
fBravo etal (2016) US: Census (Eastern two-thirds), 2000 College
< High School
All
Urban
Suburban
Rural
High School
All
Urban
Suburban
Rural
Reference
Value
(pg/m3)
12.7
13
11.2
9.87
11.2
9.87
EDUCATIONAL ATTAINMENT
*Lee et al (2019)
US: American Community Survey
(California), 2016
% Low Education
Statewide
SF Bay Area
San Joaquin Valley
South Coast
8.5
7.3
11.2
10.1
*Rosofsky et al (2018)
US: Census and American Community
Survey (Massachusetts), 2003-2010
Masters degree
2003
-------
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 individual and community-level 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 nonaccidental
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
Individual-Level SES
A recent time-stratified case-crossover analysis in North Carolina conducted by Son et al. (2020)
evaluated the association between total mortality and PM2 5 exposure, stratified by individual median
household income. The authors reported null associations for total mortality (excluding external causes)
when stratified at the median household income (< $41,500) as well as by educational attainment.
Community-Level SES
Another time-stratified case-crossover by Yitshak-Sade et al. (2019a) examined the intersection
of greenspace, cardiovascular mortality, PM2 5 exposure, and community-level educational attainment.
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.
May 2022
3-140
-------
Long-Term PM2.5 Exposure
Individual-Level SES
Individual-level SES characteristics were evaluated within some recent studies. Specifically, a
Canadian study evaluating individual-level household income and PM2 5-related nonaccidental mortality
at the six-character postal code (equivalent to a street block or single large building in urban areas) level,
reported the greatest magnitude of association in the lowest income category (< $25,000CAD HR: 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. Additionally, this study also showed that risk of PM2 5-related
cardiovascular and respiratory mortality were influenced by household income, with the strongest
association noted among those in the lowest individual income category for cardiovascular mortality (<
25,000CAD HR: 4.58 [95% CI: 2.48, 8.47]) compared with the highest income category (>
$100,000CAD HR: 1.46 [95% CI: 0.94, 2.25]). Fewer 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]) individual income categories and respiratory mortality.
Community-Level SES
The majority of recent studies focusing on SES evaluate the relationship between PM2 5 and
health effects, stratified by community-level measures of SES. Specifically, some studies explore whether
SES modifies the relationship between exposure to PM2 5 and total or all-cause 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 nonaccidental mortality by ZIP
code level income (high, medium, low) (Figure 3-31).
May 2022
3-141
-------
Urbanicity
Non-accidental
1
1
1
Cardiovascular
IHD
CBV
CHF
Respiratory
COPD
Pneumonia
Cancer
'
1
1
Lung Cancer
1
1
1
—I—
1.0 1.2 1.0
Risk Ratio (95% Confidence Interval)
—— Nonurban I High-Income
— Urban I Mid-Income
—•- I Low-Income
1.2
Source: Wang et al. (20201
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 community-level SES. another Canadian study
evaluated if a combination of greenspace and social deprivation modified the relationship between
long-term PM2.5 exposure and total nonaccidental mortality (Crouse et al.. 2019). Community-level
deprivation was assessed using the Canadian Marginalization Index, which incorporates measures such as
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 nonaccidental mortality compared with groups
with high deprivation and low greenspace. However, there was little difference in the association
comparing high to low deprivation in areas with high greenspace (Figure 3-32).
May 2022
3-142
-------
no
ICO
Non-accidental causes
Cardiometabolic causes
Cardiovascular causes
1 2
3 4
1 2
3 4
1 2
3 4
Source: Crouse et al. (20191
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
compared with males (Figure 3-33). In another study examining life expectancy, Jorgenson et al. (2020)
evaluated the relationship between PM2 5 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 effect of PM2 5 exposure on life expectancy at birth.
May 2022
3-143
-------
Q5 to Q1 difference:
Q5 to Q1 difference:
--0.041 (-0.044.-0.038)
-0.018 (-0.020, -0.016)
r •
. Vy£-- '•
1
5.0 7.5 10.0 0.0 2.5 5.0
PM2 5 exceedance in 2015 (ng/m3)
Q5 to Q1 difference:
0.046 ( 0.043, 0.048)
Q5 to Q1 difference:
0.019 ( 0.017, 0.020)
(: :•'
jgF'
jp*
5.0 7.5 10.0 0.0 2.5 5.0
PM2 5 exceedance in 2015 (ng/m3)
Per capita income » 1 Lgag1
(quintiles)
Percentage of population whose family income 9 1 Least 2 • 3 • 4 • 5 Most
is below the poverty threshold (quintiles)
Q5 to Q1 difference:
Q5 to Q1 difference:
" -0.045 (-0.048, -0.042)
-0.021 (-0.023, -0.020)
jr
.y.-
jjjpT
Q5 lo OI difference:
0.022 ( 0.019,0.025)
Q5 to Q1 difference:
0.010(0.008, 0.012)
¦J&'
Jw?
j* '
0.0 25 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0
PM2 5 exceedance in 2015 (jig/m3) PM2 5 exceedance in 2015 (jig/m3)
Percentage of population who have graduated • 1 LeasI 2 • 3 • 4 • 5 Most Percentage of population who are of « ^ Least 2 • 3 • 4 • 5 Most
from high-school (quintiles) Black or African American race (quintiles)
Source: Bennett et aL (2019)
Q5 to Q1 difference = estimated difference in life expectancy loss between quintile 5 ($34,200—$114,000) and quintile 1
($17,400-$24,900).
Figure 3-33 County-level life expectancy losses due to PM2.5 exceeding
2.8 |jg/m3.
Numerous recent studies also evaluated whether the association between PM2 5 exposure and certain
causes of death (i.e., cause-specific mortality) is modified by community-level SES. 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). However, 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, 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 PM25
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
May 2022
3-144
-------
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 PM;<. However, as time progressed through to 2010, reductions ill CMR were the largest
in the most deprived counties (Figure 3-34).
A) Annual change in PM2 5
(|jg/m3)
National
0.0
-0.1
8 -0.2
-0.3
Deprivation Quintile
B) Annual change in CMR per unit decrease in PM2.5
(deaths/100,000 person-years/pg/m3)
National Deprivation Quintile
& s#
Year
D) % CMR mediated via PM25
(%)
National
40
30
GC.
e 20
S?
10
0
Deprivation Quintile
&
Year
1
County deprivation
National
1 Lowest deprivation
2 Low deprivation
3 Mid deprivation
4 High deprivation
j 5 Highest deprivation
Source: VWattetal. (2020b)
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 |jg/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 (3
indicates an increase in CMR per unit increase in PM25, while a negative sign indicates a decrease. A negative sign for a(3 indicates
a net reduction in CMR related to PM25 change, whereas 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 PM25 and cardiopulmonary,
cardiovascular, and IHD mortality associated with PM2 5 exposure and increased vulnerability. Both PMa 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
& # n
-------
greenspace protected against PM2 5 attributable cardiovascular and cardiometabolic (defined as the
combination of circulatory and diabetes) 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
CHF or AMI, using the Canadian ONPHEC, by neighborhood-level income categories. The authors
observed the strongest association 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 with 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, MI, hypertension, and diabetes, among cardiac catheterization patients. When
compared with 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 CAD
outcomes was in Cluster 3 (OR= 1.15 [95% CI: 1.00, 1.31], per 5 (ig/m3) (Figure 3-35). Overall, this
study concluded that areas with a relatively greater amount of social disadvantage had stronger
associations in magnitude between PM2 5 and hypertension compared with areas with lower social
disadvantage.
May 2022
3-146
-------
2
1
O
-p
o~-
LO
a:
O
2
1
Source: Weaver et al. (2019^
CAD = Coronary Artery Disease, Ml = Myocardial Infarction. Results resented for 1 Mg/m3 increase in PM2.5 concentrations.
Figure 3-35 Odds ratios for the association between PM2.5 and cardiovascular
outcomes and diabetes by neighborhood cluster.
3.3.3.2. Race/Ethnicity
The 2019 PM ISA provided evidence indicating that people of different racial and ethnic
backgrounds experience disparities in the risk of PM2 s-related health effects. When the 2009 PM ISA was
finalized, there were relatively few studies evaluating whether race/ethnicity modifies the relationship
between PM2 5 exposure and health effects. As a result, the 2009 PM ISA observed little evidence for
increased PM2 5-related risk by race and some evidence of increased risk by Hispanic ethnicity. However,
the 2019 PM ISA demonstrated evidence that there are consistent racial and ethnic disparities in PM ?
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
May 2022
3-147
-------
exposure across the U.S., particularly for Black/African Americans, compared with non-Hispanic White
individuals. Additionally, some studies provided evidence of increased PM2 5-related 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 minority populations, particularly Black populations, are at increased risk for PM2 5-related
health effects, in part due to disparities in exposure. The following sections present an evaluation of
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 conducted 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, which included that there are disparities in PM2 5
exposure by race and ethnicity. When considered together, these additional studies provide further
evidence that minority communities are exposed to higher concentrations of PM2 5, compared with
predominantly White communities (Figure 3-38—Race exposure, Table A-14—Race exposure).
Specifically, Figure 3-38 and compares the mean exposure of minority populations with that of
nonminority populations, with a ratio value > 1 indicative of higher PM2 5 exposure among the minority
group.
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 and 2014, observed that Black participants (16.1 (.ig/ni3) were exposed to
slightly higher concentrations of PM2 5 compared with White participants (15.7 (.ig/ni3) (Erqou ct al..
2018). Similarly, the Veterans Cohort Study, conducted between 1976 and 2001 also showed that Black
participants (15.7 (.ig/ni3) were exposed to a substantially higher concentration of PM2 5 compared with
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 with White race (10.9 |_ig/m3). those of Black race (12.3 (.ig/ni3) experience higher
exposures to PM2 5, 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 PM2 5 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 populations being exposed to elevated concentrations of PM2 5
compared with White populations (Awad et al.. 2019). A study by Parker et al. (2018) using the Health
May 2022
3-148
-------
Interview Survey also showed that a greater number (37.2%) of non-Hispanic Black and Hispanic
(33.3%) individuals lived in the highest PM2 5 quartile, compared with White individuals (21.3%).
Additionally, most White (26.1%) and Hispanic (34.7%) individuals lived in the lowest PM2 5 quartile,
compared with only 10.5% of non-Hispanic Black individuals.
Other recent studies used data from the U.S. Census or American Community Survey 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/m3). populations experiencing higher PM2 5 exposures compared with
non-Hispanic White (2003: 11.1 (ig/m3, 2010: 7.8 (ig/m3) 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 with areas with a
higher percentage of White individuals. Additionally, compared with areas with a predominant population
of 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/m3)
compared with 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 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 PM2 5 (6.0 (.ig/m3). 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/m3). while White individuals have an estimated pollution
May 2022
3-149
-------
inequity of-17% (exposed to 4.6 (.ig/m5, consume 5.5 fig/m') (Figure 3-36). Another similar study by
Tessum et al. (2021) showed that people of color are consistently exposed to PM; < caused by each emitter
type in the U.S. The authors estimated Black (7.9 |ag/m3). Asian (7.7 (.ig/m3), Hispanic (7.2 (.ig/m3)
individuals are exposed to greater proportions of PM2.5 compared with White (5.9 ug/m ) individuals
(Figure 3-37).
Source; Tessum et al. (2019)
Note: In 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 Racial and ethnic inequities in PM2.5 exposure caused by
population-adjusted group consumption ("caused") and total
personal consumption ("exposed").
Pop. Avg.
Caused Exposed Caused Exposed Caused Exposed
' Black ' ' Hispanic ' ' White/Otfier
A) Emitters B) End Uses
Road Dustl I Trans, d
Res. Wood Comb. I I Shelter d
Res. Gas Comb. Services l~
Res. Otherl 1 Info. I
Off-HiahwavB I Goods I
Non-Coal Elec.l 1 Food I
Miscellaneous! II Electricity I
LD Gas Veh.l I
Industrial H 1
HD Diesel Veh.d
Construction I I
Comm. Cooking d
Coal Elec. Util.B
Agriculture!
May 2022
3-150
-------
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)
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.
May 2022
3-151
-------
US: American Community Survey, 2009-
2013
Study Pop u l;i
tMikati etal (2018)
tNachman et al (2012) US: NHIS Coliort, 2002-2005
tBravo et al (2016) US: Census (Eastern two-thirds), 2000
"fBell & Ebisu (2012)
tBasu et al (2004)
*Eroqou et al (2018)
~Kelly et al (2020)
*Leeetal (2019)
*Rosofsky el al (2018)
*Awad etal (2019)
*Lim etal (2018)
*Lipfert et al (2020)
*Wang et al (2020)
US: Census (tracts w/ PM Component
Monitors), 2000
US: Birth records (California), 2000
US: HcartSCORE Study (Pittsburg, PA),
2001-2014
US: Census, 2010
US: American Community Survey
(California), 2016
US: Census and American Community
Survey (Massachusetts), 2003-2010
US: Medicare, 2000-2012
US: NIH-AARP Diet and Heallh Study,
1995-2011
US: Veterans Coliort Study, 1976-2001
US: Medicare, 2000-2008
Black
Hispanic
Black
Hispanic
(|ig/'m3)
9.6
12.2
12.5
Black
Hispanic
Other
n
Black
Hispanic
Other
Black
Hispanic
Other
Black
Hispanic
Asian
Other
Hispanic
Black
CMAO
Black .
Hispanic
Other
CAMX
Black
Downscaler 9.9
Black
Hispanic
Other
DI2016 9.7
Black
Hispanic
Other
DI2019 9.5
Black
Hispanic
Other
VNA 9.6
Black
Hispanic
Other
VD2019 9.0
Black .
Hispanic
Other
HU2017 9.3
Black
Hispanic
Other
EVNA 9.1
Black .
Hispanic
Other
>75th Percentile %POC
Statewide 7.8
SF Bay Area 7.0
San Joaquin Valley 10.6
South Coast 9.9
2003 11.1
NH Black
NH Asian
Hispanic
2010 7.8
NH Black
NH Asian
Hispanic
Black . 10.9
Hispanic
Asian
Black 13.9
Minority Gr 10.4
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 on Figure 12-2 in the 2019 PM ISA.
Figure 3-38 Difference in PM2.5 exposure by race.
May 2022
3-152
-------
3.3.3.2.2. Health Risk Disparity
Since the literature cutoff date of the 2019 ISA, several additional studies evaluated disparities in
the risk of PM2 5-related health effects, stratified by race and ethnicity. A small number of studies were
included in the 2009 PM ISA that summarized racial and ethnic disparities in PM2 5 mortality risk. The
2019 PM ISA further identified several studies which provided evidence for an increased association
between mortality and long-term exposure to PM25 among minority 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
present an evaluation of 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. Additionally, the North
Carolina case-crossover study by Son et al. (2020) also reported very little racial and ethnic differences in
the magnitude of the association between short-term PM2 5 exposure and 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]).
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 with 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 with a 3.35% (95% CI: 1.57, 5.16)
May 2022
3-153
-------
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 Residential 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 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 PM2 5 exposure among any of the groups
(Figure 3-40).
May 2022
3-154
-------
X®
0s-
l/j
0\
«
+-
im
©
£
¦o
a>
w
js
"3
ec
=
«
.=
u
+¦>
s
a>
£
o>
Ch
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)
Note: Results presented for a 10 |jg/m3 increase in PM2.3.
Figure 3-39 Percent change in cardiovascular disease mortality by PM2.5
exposure, stratified by census block group racial composition in
Massachusetts (2001-2011).
May 2022
3-155
-------
r
o
c
¦D
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)
Note: Results presented for a 10 |jg/m3 increase in PM2.5.
RRS = racial residential 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 (RSS)
metric 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 (nonaccidental) mortality and long-term PM2.5 exposure. A study by
Parker et al. (2018) using NHIS 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, Awad 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). By utilizing data from the subpopulation of Medicare enrollees that moved, the
authors were able to examine changes in PM2.5 exposure, thus creating a natural experiment that
essentially 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 with 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 to 2001 evaluated mortality risk among
Black and White veterans. The RR were expressed in terms of the difference in the annual average and
May 2022
3-156
-------
the minimum concentration of PM25, and the results are interpreted as the change in mortality that would
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 with 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
that examined the association between long-term PM2 5 exposure and nonaccidental mortality among
Medicare beneficiaries by Wang et al. (2020) observed positive, but equal, RR among Black (RR: 1.02
[95% CI: 1.02, 1.02]) and White (RR: 1.03 [95% CI: 1.02, 1.03]) beneficiaries.
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 with 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).
May 2022
3-157
-------
90th Percentile of Top 10% Share = 50.8%
% Biack
Source; Joraenson et al. (2020)
Figure 3-41 Relationship between life expectancy and PM2.5 exposure by
income inequality and percent Black.
Recent epidemiologic studies also evaluated specific causes of death attributable to PM25
stratified by race and ethnicity. In an analysis of the NIH-AARP Diet and Health Study (1995-2011),
Lim et al. (2018) evaluated diabetes mortality. Minority participants (i.e., Non-Hispanic Black, Hispanic,
Asian, Pacific Islander, or American Indian/Alaskan Native, or unknown race/ethnicity, referred to as
May 2022
3-158
-------
"Other" in the study) had an association larger in magnitude between annual PM2 5 exposure and diabetes
mortality (HR: 1.27 [95% CI: 1.02, 1.58]), compared with White participants (HR: 1.05 [95% CI: 0.96,
1.14]). However, other studies that examined whether there are disparities in PIVb s-related cause-specific
mortality did not report results consistent with Lim et al. (2018). The study by Parker et al. (2018) that
used data from the NHIS between 1997 and 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]). Finally, the study by Wang et al. (2020) evaluating over 53 million Medicare
beneficiaries showed associations between PM2 5 and cause-specific mortality that were larger in
magnitude among White beneficiaries compared with minority beneficiaries (i.e., Black, Asian, Hispanic
race/ethnicity, referred to as "non-White" in the study) for cardiovascular mortality (White RR: 1.05
[95% CI: 1.05, 1.06], minority RR 1.03 [95% CI: 1.02, 1.03]), heart disease mortality (White RR: 1.07
[95% CI: 1.07, 1.08], minority RR 1.03 [95% CI: 1.02, 1.04]), and vascular disease mortality (White RR:
1.07 [95% CI: 1.06, 1.08], minority RR 1.05 [95% CI: 1.04, 1.06]).
In a study of post-menopausal women enrolled in the WHI, Honda et al. (2017) estimated the
association between long-term exposure to PM2 5 and incident hypertension by individual race/ethnicity,
and also by dichotomizing (White versus minority participants [i.e., Black, Asian/Pacific Islander,
Hispanic/Latino, referred to as "non-White" in the study]). This study indicated that the association
between PM2 5 and incident hypertension was larger in magnitude among Asian/Pacific Islander (HR:
1.34 [95% CI: 1.00, 1.64]), minority (i.e., non-White in the study) (HR: 1.27 [95% CI: 1.17, 1.38]), and
Black participants (HR: 1.26 [95% CI: 1.06, 1.44]) compared with 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 (defined as
cardiovascular disease, stroke, and diabetes) and long-term PM2 5 exposure reported no differences in the
association by race (Juarez et al.. 2020) (Figure 3-42).
May 2022
3-159
-------
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 al. (2020)
Figure 3-42 Probability of cardiometabolic disease and PM2.5 exposure,
stratified by race, gender, and hypertension status.
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 to exposure. When considering indicators 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 with 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. Studies
evaluating composite metrics, including a combination of SES factors and even some measures of
race/ethnicity, generally demonstrated associations larger in magnitude between various health outcomes
and long-term PMis exposure, thus demonstrating the complexity of SES indicators. 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 populations 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 minority populations including
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
Probability of CMD
0.20
O.I5
0.10
PM 25(mhir)
May 2022
3-160
-------
evaluating PM2 5-related health risks by SES add to the growing evidence presented in the 2019 PM ISA.
In addition to the indicator-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 (nonaccidental) 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 with 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 with non-Hispanic White populations.
However, similar to SES, there was less consistency when evaluating PM2 5 exposure and all-cause or
total (nonaccidental) (Section 3.3.3.2).
May 2022
3-161
-------
4.EVALUATION OF RECENT WELFARE EFFECTS
EVIDENCE
The Integrated Science Assessment for Particulate Matter (2019 PM ISA) concluded a causal
relationship for each of the three nonecological 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 in 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 light extinction. The greatest improvements have occurred in the eastern half of the
country, 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 large fraction of PM2 5 mass in the eastern U.S., 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 part of the country has resulted in better visibility.
May 2022
4-1
-------
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 part of
the country than in the east, 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 west. 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 light extinction (scattering + absorption) by
light-absorbing carbon. However, the difference in light extinction between the eastern and western U.S.
also reflects considerably higher PM2 5 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
May 2022
4-2
-------
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 extinction coefficient (bext), which
relates the distance of an observed object to 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 (Section 4.2.2).
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 hard copy 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 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
observers that indicated acceptable visual air quality, labeled acceptable (%) on the y-axis, as a function
of light extinction, labeled bext on the x-axis. Data are based on rankings of photographic images from six
visibility preference studies conducted in Phoenix, AZ (ADEO. 2003); Chilliwack and Abbotsford, BC
(Prvor. 1996). Denver, CO (Ely et al.. 1991). Washington, DC (Abt. 2001). and the Grand Canyon, AZ
(Malm et al.. 2019).
May 2022
4-3
-------
LU
_l
m
£
CL
LU
O
400
Source: Malm et al. (2019)
Figure 4-1 Percent acceptability levels plotted against light extinction (bext)
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).
These results clearly demonstrate a large range in light extinction (bext) across different locations
considered acceptable by 50% of observers, indicating that metrics based on light extinction are not
universal indicators of visibility preference levels. For example, considerably more light extinction was
regarded as acceptable by 50% of observers for the Washington, DC scene (192 Mm-1) than for the Grand
Canyon, AZ scene (23 Mm"') (Malm et al.. 2019). For other locations the amount of light extinction
considered acceptable by 50% of observers was intermediate between Washington, DC and the Grand
Canyon, AZ (Malm et al.. 2019). For context, urban monthly average beKt derived from 2011 to 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).
May 2022
4-4
-------
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 light 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, which integrate the effects of bext along
the sight paths between observers and landscape features, are better predictors of preference levels than
universal metrics like light extinction. The explanation for this is that light 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 light 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 light 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 reach the visual range,
corresponding to a contrast between approximately -0.03 and -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 is less dependent on scene, as indicated by a smaller difference between average acceptability
between studies than when acceptability is expressed as a function of light 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 other studies were closer (Malm et al.. 2019).
May 2022
4-5
-------
-0.24 -0.20 -0.16 -0.12
Contrast
-0.08
-0.04
-0.00
Source: Malm et al. (2019)
Figure 4-2 Percent acceptability levels plotted against apparent contrast of
distant landscape features for each of the images used in studies
for 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 -0.03 to -0.05. Further, an acceptability level of 90% would require contrast levels to
remain above a level of about -0.01.
4.2,2. Recent Estimates of Light Extinction Trends
The 2019 PM ISA reported that light extinction by PM was decreasing in most regions of the
U.S. and that the greatest improvements had occurred in the eastern half of the country, in regions with
May 2022
4-6
-------
the poorest visibility based on the analysis of Interagency Monitoring of Protected Visual Environments
(IMPROVE) network (U.S. EPA. 2019). Analysis of IMPROVE network data since the publication of the
2019 PM ISA show that the trends of decreasing light extinction and decreasing contribution of
ammonium sulfate to total light extinction have continued in the eastern U.S., but that in the western part
of the country, changes in total light extinction were smaller, and the contribution of particulate organic
matter to atmospheric light extinction was increasing due to increasing wildfire emissions (Hand ct al..
2020). On average, light 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 light extinction coefficient decreased by 4.3% per year from
2002 to 2018 and was associated with major reductions of SO2 and NOx emissions (Hand et al.. 2020).
The reduction in light 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).
The trend in light extinction is depicted in Figure 4-3, which shows regional time-series plots of
estimated annual mean bext for major aerosol species aggregated from individual remote and rural
monitors for the eastern U.S. (east of 100° W), the Intermountain West and Southwest (between 100° W
and 116° W), and the West Coast (west of 116° W) from 2002 to 2018 as reported by Hand et al. (2020).
The greatest declines in bext were observed in the eastern U.S., where annual mean total bext declined by
74% from 2002 through 2018 and light extinction by ammonium sulfate, which was estimated to account
for 72% of total light extinction in 2002-2004, decreased by 148% (Hand et al.. 2020). The annual mean
bext was also significantly highly correlated (r = 0.096) with combined SO2 + NOx total emissions in the
eastern U.S. (Hand et al.. 2020). During the period 2016-2018, the average contribution to total
reconstructed light extinction in the eastern U.S. was 55% from ammonium sulfate and ammonium nitrate
combined, 31% from carbonaceous aerosols, and 14% from fine dust, coarse particulate matter, and fine
sea salt combined. Compared with 2016-2018, average contributions during 2002-2004 of 72% from
ammonium sulfate and ammonium nitrate combined, 21% from carbonaceous aerosols, and 5% from fine
dust, coarse particulate matter, and fine sea salt combined. Hand et al. (2020) also observed that the
highest bext value had shifted westward from the Ohio Valley and Appalachians to the agricultural regions
of the central U.S.
May 2022
4-7
-------
Regional Mean Average Aerosol bext
soi-(a) E~ast iqr
(b) Intermountain/Southwest
2000
Source: Hand et al. (2020)
2005
2010
Year
2015
2020
Figure 4-3 Annual mean reconstructed light extinction (bext) for the a) East,
b) Intermountain West/Southwest, c) West Coast by species,
including ammonium sulfate (AS), ammonium nitrate (AN),
particulate organic matter (POM), fine dust (FD), coarse mass
(CM), and sea salt (SS). Map insets show individual sites
aggregated into regional means.
May 2022
4-8
-------
In the western regions of the U.S., the reduction of total reconstructed bext was smaller, decreasing
by 15% in the Intermountain West and Southwest and 25% along the West Coast (Hand et al.. 2020). In
these regions, the contribution of ammonium sulfate and nitrate had been roughly equal to the
contribution of carbonaceous aerosols (i.e., organic mass and elemental carbon) during 2002-2004 in both
regions, but in 2016-2018, light extinction became dominated by carbonaceous aerosols, which
accounted for 45% of reconstructed total light extinction in the Intermountain/Southwest and 51% in the
West Coast region. Although some of this change occurred because ammonium sulfate and nitrate also
decreased in these regions, an increase in wildfire emissions was also a likely contributor (Hand et al..
2020). Therefore, the analysis conducted by Hand et al. (2020) indicates that the composition of haze has
shifted away from being dominated by sulfate to having greater contributions from carbonaceous and
crustal aerosols. Additionally, as emissions of SO2 and NOx continue to decline, contributions to haze
from unregulated sources including oil and gas extraction, agricultural activities, international sources,
wildfires, and windblown dust seem to be increasing.
4.2.3. Recent Advancements in Visibility Monitoring and Assessment
The changing PM concentrations and composition in the U.S. have also resulted in an increasing
bias in bext estimates (Prenni et al.. 2019). An artifact of the method used to estimate light extinction from
national monitoring network data could be contributing up to 1% per year in the eastern U.S. and 0.5%
per year in the western U.S. to apparent long-term trends Hand et al. (2020). Research is progressing to
address these concerns about increasing bias in light extinction estimates, and an alternative approach to
estimating the split of component mass between large and small size modes reduced the bias (Prenni et
al.. 2019). A more detailed explanation of this approach in the context of the evolution of the IMPROVE
algorithm used to estimate light extinction from the IMPROVE network is presented in the Appendix A.
Section A.l.
Other recent research has addressed the effects of relative humidity on light extinction, the effect
of increasing wildfires and atmospheric dust, and development of new instrumentation and measurement
methods. A 25% underprediction was reported when reconstructed light extinction was compared with
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 light extinction estimates from both the original and revised IMPROVE equations (Equations
A-l and A-2 in Appendix A. Section A. 1) was reasonable at average relative humidity, but substantially
lower at both higher and lower humidity
The importance of including relative humidity in estimating light 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 micrograms per cubic
meter (fig/m') and the variability of light extinction was controlled primarily by differences in PM
May 2022
4-9
-------
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 light
extinction was controlled by both PM2 5 concentration and relative humidity (Bcvcrsdorf et al.. 2016).
Recent decreases in SO2 and NOx emissions have coincided with increasing PM emissions from
wildland fires (U.S. EPA. 2019) as well as dust in some regions of the U.S. (Lambert et al.. 2020). 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 light extinction based on constant mass scattering
coefficients, as in Equations A-l through A-4 in Appendix A. Section A. 1.
Recent studies have also introduced new instrumentation and measurement methods that could
help to reduce uncertainties in light 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 to 2017 (Chow et al.. 2021). These results suggest 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).
Capabilities for estimating light extinction from photographic images have also progressed. In the
2019 PM ISA photography is identified as one of several methods of measuring light extinction (U.S.
EPA. 2019). and results from image processing techniques had been shown to be highly correlated with
measured light extinction under hazy conditions. Additionally, a recent study demonstrated that light
extinction could also be estimated quantitatively from photographic images under more pristine
conditions from webcams that are routinely operated at the Grand Canyon and Great Smoky Mountains
National Parks (Malm et al.. 2018).
May 2022
4-10
-------
4.3. Summary of Recent Evidence in the Context of the 2019
Integrated Science Assessment for Particulate Matter Causality
Determination for Visibility Effects
Recent studies published since the 2019 PM ISA have addressed several existing research gaps
and emerging trends identified in the 2019 PM ISA. Analyzing visibility preference study results using
contrast as a metric greatly reduced the variability in acceptability between studies that reported results in
terms of light extinction or related metrics. Evaluation of uncertainties and development of alternative
approaches to estimating light extinction using the IMPROVE algorithm as well as alternative methods
for analyzing both light extinction and PM species have helped to meet new needs introduced by rapidly
decreasing sulfate and increasing fire-related contributions to PM. New measurements of physical and
optical properties of wildfire smoke also provide useful new data for understanding PM sources
responsible for light extinction.
Some recent studies evaluated in this section provide the following new insights:
• The wide range in response to the level of acceptable visibility observed across different settings
was reduced by accounting for the distances of 20-59 km between observer and landscape feature
as a part of a visibility metric. This reduction was demonstrated by observation of a smaller
variation across different settings using apparent contrast than light extinction.
• Impacts of the rapidly decreasing sulfate and increasing fire-related contributions to PM have
been evaluated in recognition that the changing nature of PM composition in the U.S. is changing
the relationship between PM and visibility impairment.
• The changing relationship between PM and visibility impairment has led to increased bias and
spurred alternate approaches that have reduced bias in the IMPROVE algorithm used to estimate
light extinction.
Additional recent studies further support the conclusions in the 2019 PM ISA, specifically:
• In polluted environments most of the light extinction (bext) is due to mainly 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) is generally elevated in urban centers compared with surrounding rural
areas, particularly in the western U.S.
• In practice, 6ext is estimated with reasonable accuracy using routinely available PM species
monitoring data using data on dry mass extinction efficiency and hygroscopicity growth functions
for major species. Mass extinction efficiencies can vary by a factor of 10 or more between
particulate species, which vary by region and season as well as by urban versus rural settings.
• Mass extinction efficiencies for sulfates, nitrates, and organics in rural areas tend to increase with
increasing concentrations due to shifts in the size distributions and more recent studies have
shown a similar dependency on concentration in urban and polluted environments.
May 2022
4-11
-------
5.SUMMARY AND CONCLUSIONS
The 2019 Integrated Science Assessment for Particulate Matter (2019 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
minority populations, with more limited evidence for people of low socioeconomic status (SES). Finally,
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 for which 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, both of which 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 minority populations, specifically Black individuals, and low SES
individuals, experience disparities in both PM2 5-related health risks and exposures compared with
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 evidence base that informed the scientific
conclusions of the 2019 PM ISA.
Overall, this Supplement to the 2019 PM ISA found the following:
May 2022
5-1
-------
Cardiovascular Effects
Short-Term PM2.5 Exposure
• Recent U.S. and Canadian multicity studies conducted within populations with diverse
demographic characteristics 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 to 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 alternative methods for
confounder control also report positive associations across a number of statistical approaches,
which further supports a relationship between short-term PM2 5 exposure and cardiovascular
effects. Overall, 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.
Long-Term PM2S 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 among the general population, 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 nonlinearity at lower concentrations for some outcomes. Finally, a few recent
epidemiologic studies that employed alternative methods for confounder control, reduce some
uncertainties related to potential confounding bias and further support a relationship between
long-term PM2 5 exposure and cardiovascular effects. Overall, 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.
Mortality
Short-Term PM2.5 Exposure
• Since the literature cutoff 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 base
evaluated in the 2019 PM ISA and in previous assessments that reported consistent positive
May 2022
5-2
-------
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
unchanged in copollutant models. Additionally, evidence continues to build that the heterogeneity
in city-to-city PM2 5-mortality risk estimates can be attributed partly 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 alternative methods for confounder control and report
consistent, positive associations. Overall, recent epidemiologic studies published since 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.
Long-Term PM2.5 Exposure
• Recent epidemiologic studies conducted in the U.S. and Canada consisting of cohorts with mean
annual PM25 concentrations mostly 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 support a linear, no-threshold relationship at PM2 5 concentrations > 8 (ig/m3.
However, uncertainties remain about the shape of the C-R curve at lower PM2 5 concentrations (<
8 (ig/m3), with some recent studies providing evidence for a sublinear, linear, or supralinear
relationship at these lower concentrations. Finally, an extensive number of epidemiologic studies
that conducted accountability analyses or employed alternative methods for confounder control
have been conducted since the literature cutoff 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. Overall,
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
mortality.
Additional Considerations Regarding the Health Effects of PM2.5
Experimental Studies at Near-Ambient PM2.5 Concentrations
o 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
May 2022
5-3
-------
the 2019 PM ISA, which could be attributed to the higher ventilation rate and longer
exposure duration used compared with other studies.
SARS-Co V-2 Infection and COVID-19 Death
o 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 SARS-CoV-2
infection and COVID-19 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 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 outcomes,
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). While there is initial evidence of positive associations with SARS-CoV-R
infection and COVID-19 death, uncertainties remain due to methodological issues.
Populations at Potentially Increased Risk of a PM-Related Health Effect
Socioeconomic Status
o 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 with higher SES groups.
Race and Ethnicity
o 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 minority populations, specifically Black populations. Black populations or
individuals that live in predominantly Black neighborhoods experience higher PM2 5
exposures, in comparison with non-Hispanic White populations. Additionally, there is
evidence of health risk disparities for both Hispanic and non-Hispanic Black populations
compared with 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 affecting estimates of light extinction. In response, alternate
approaches to the application of the traditional IMPROVE algorithm for estimating light
extinction have been developed that reduce the bias in light extinction estimates. Parallel efforts
have better characterized light extinction by major contributors to PM, particularly for biomass
burning. Overall, recent studies published since the 2019 PM ISA support and extend the
evidence that contributed to the conclusion of a causal relationship between PM and visibility.
May 2022
5-4
-------
Appendix A
Table A-1 Study-specific details for epidemiologic studies of short-term PM2.5 exposure 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)
Wasatch Front, UT
(4 counties)
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 (ICD-9
410), HF (ICD-9 428),
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
May 2022
A-1
-------
Table A-1 (Continued): Study-specific details for epidemiologic studies of short-term PM2.5 exposure 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 did not
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, one
monitor, east side 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-Apr
9.80 May-Oct
Upper (Max):
79.2 all year;
79.20 Nov-Apr;
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
May 2022
A-2
-------
Table A-1 (Continued): Study-specific details for epidemiologic studies of short-term PM2.5 exposure 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
three 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, MO, and IL
(16 counties)
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 (ICD-9 428);
Cardiac dysrhythmia
(ICD-9 427); IHD (ICD-
9 410-414); Stroke
(ICD-9 433-437)
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
May 2022
A-3
-------
Table A-1 (Continued): Study-specific details for epidemiologic studies of short-term PM2.5 exposure and
cardiovascular effects.
Study/Location,
Years
Study Population Exposure Assessment
Mean and Upper
Percentile
Concentrations
(|jg/m3)a
Outcome
Confounders
Considered
Copollutant
Examination
Ye etal. (2018)
Atlanta, GA
(5 counties)
Aug 14, 1998-Dec 15,
2013
24 h avg from one monitor
(Jefferson Street). A priori
lag 0.
Mean (SD): 14.46
(7.69)
IQR: 9.28
75th: 18.21
CVD ED visits
(i.e., IHD, Dysrhythmia,
and CHF)
Temporal trends
and meteorology
(maximum
temperature, cubic
function of
minimum and dew
point temperature,
day of week,
holiday, season,
hospital
participation
period)
Correlation (r):
CO: 0.47
N02: 0.50
SO2: 0.24
Ozone: 0.44
WS Fe: 0.65
Copollutant
models with:
water-soluble Fe
Fisher etal. (2019)
Contiguous U.S.
1999-2010
HPFS
Men
40-75 yr in 1986
n = 51,529
Validated kriging models to Mean (SD): 12.9
estimate daily PM2.5 (7 4)
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.
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)
May 2022
A-4
-------
Table A-1 (Continued): Study-specific details for epidemiologic studies of short-term PM2.5 exposure 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 aged 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,
30-day hospital
readmissions
(ICD-9 codes:
401-405, 410-411,
413, 426-27, 428)
HF Temperature and
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.
Correlation (r):
NR
Copollutant
models with: NR
May 2022
A-5
-------
Table A-1 (Continued): Study-specific details for epidemiologic studies of short-term PM2.5 exposure 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
Continental U.S.
2000-2012
(low-income and/or
disabled)
n = 3,666,657 CVD
HAs
Daily average at ZIP code Mean (SD) 11.5 First HA for CVD (ICD- Air and dew-point Copollutant
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)
9 390-495); IHD (ICD- temperature (daily correlation (r):
9 410-414); CHF (ICD-
9 428); AMI 410.9; IS:
(ICD-9 434.91)
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; BASIC = Brain Attack Surveillance
in Corpus Christi: BC = black carbon; CHF = congestive heart disease; CMAQ = Community Multi-Scale Air Quality; C-V = cross-validation; CVD = cardiovascular disease(s);
ED = emergency department; h = hour; HA = hospital admission; HF = heart failure; HPFU = Health Professionals Follow-up Study; ICD-9 = International Classification of Disease
9th revision; IHD = ischemic heart disease; IQR = interquartile range; km = kilometer(s); LUR = land use regression; Ml = myocardial infarction; NA = not applicable; NR = not
reported; PE = prediction error; PM10 = particulate matter with a nominal mean aerodynamic diameter less than or equal to 10; r= correlation coefficient; REGARDS = REasons for
Geographic and Racial Differences in Stroke; RH = relative humidity; RMSS = root mean square standardized; SD = standard deviation; SPE = standardized prediction error;
STEMI = ST segment elevation myocardial infarction; USRDS = U.S. Renal Data System; WHO = World Health Organization; WS Fe = water-soluble iron.
May 2022
A-6
-------
Table A-2 Study-specific details for epidemiologic studies using accountability analyses or alternative
methods for confounder control to examine short-term PM2.5 exposure 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
May 2022
A-7
-------
Table A-2 (Continued): Study-specific details for epidemiologic studies using alternative methods for confounder
control to examine short-term PM2.5 exposure and cardiovascular effects.
Study/Location
Years
Population
(Cohort)
Outcome
Exposure Assessment
PM2.5 Concentration
(Hg/m3)
Confounders
Copollutant
Examination
Qiu et al. (2020)
Connecticut,
Maine,
Massachusetts,
New Hampshire,
Rhode Island, and
Vermont
2000-2012
Medicare
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
BC = black carbon; BMI = body mass index; EPA = Environmental Protection Agency; NA = not applicable; PM25 = particulate matter with a nominal mean aerodynamic diameter
less than or equal to 2.5; r = correlation coefficient; SD = standard deviation; SPARCS = New York State Department of Health Statewide Planning and Research Cooperative
System; STEMI = ST segment elevated myocardial infarction.
May 2022
-------
Table A-3 Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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 [ICD-10
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
May 2022
A-9
-------
Table A-3 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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)
Jackson, MS
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
May 2022
A-10
-------
Table A-3 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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)
U.S. Nationwide
2003-2007 (baseline) to 2012
REGARDS
n = 17,126 Black and
White adults (> 45 yr
old)
Total CHD
(deaths and
nonfatal Ml
combined),
nonfatal 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
May 2022
A-ll
-------
Table A-3 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure and
cardiovascular effects.
Mean and
Upper
Percentile
Study/Location Exposure Concentration Confounders
Years Population (Cohort) Outcome Assessment (|jg/m3) Considered Copollutant Examination
Honda et al. (2017)
Exposure: 1980-2010
Outcome: 1993-1998
(recruitment) to 2010
income,
employment
status, insurance
status, history of
high cholesterol,
history of
cardiovascular
disease, history
of diabetes,
clinical trial study
arm, and WHI
clinical site
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
Correlation (r): PM10:
0.56; PM10-2.5: 0.03
Copollutant models
with: NR
Duan et al. (2019b)
Pittsburgh and Chicago, U.S.
SWAN
n = 417 Black and
White women (mean
age 51 yr at baseline)
cIMT by
ultrasound and
plaque burden
(i.e., four
levels, with 0
for no plaque
to 3 for a
plaque taking
up > 50%
diameter of the
artery)
Annual average 360
days prior to clinic visit
calculated from
monitors located within
20 km of residential
address
Daily PM2.5 values
retrieved from U.S.
AQS
(Green et al.. 2016:
Ostro et al.. 2014)
Mean: 16.5
(baseline)
75th: 17.1
(baseline)
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:
Copollutant models:
with O3
May 2022
A-12
-------
Table A-3 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure and
cardiovascular effects.
Mean and
Upper
Study/Location
Percentile
Exposure
Concentration
Confounders
Years
Population (Cohort)
Outcome
Assessment
(Hg/m3)
Considered
Copollutant Examination
Duan et al. (2019a)
SWAN
Mean cIMT,
5-yr avg 360 days prior
Mean (SD):
Age, race,
Correlation (r):
Detroit, Ml; Oakland, CA;
n = 1,188 women
max cIMT, IAD
to each clinic visit
14.9 (1.9)
education, SES,
Os: 0.56
Pittsburgh, PA; Chicago, IL;
and Newark, NJ
Mean age 59.6 yr
and plaque
presence or
calculated from
monitors located within
75th: 16.1
Figure 2
BMI, and CVD
risk factors
Copollutant models :
with O3
Exposure: 1999-2005
severity
(i.e., four
20 km of residential
address
(i.e., smoking,
total cholesterol,
Outcome: 2009-2013
levels, with 0
for no plaque
to 3 for a
plaque taking
Daily PM2.5 values
retrieved from U.S.
AQS (Green et al..
2016: Ostro et al..
HDL-c,
triglyceride,
menopause
status, hormone
up > 50%
2014)
use, fasting
diameter of the
glucose,
artery)
antidiabetic
medication, and
antihypertensive
medication)
Keller et al. (2018)
MESA Air
CAC
Spatiotemporal
Mean (SD):
Age, sex,
Correlation (r):
Baltimore, MD
n = 1,005
progression
prediction models used
15.9 (0.80)
race/ethnicity,
NR
Jul 2000-Aug 2002 through
2012
(Agatston
units)
to long-term average
PM2.5 concentration
site, scanner
type, adiposity,
Copollutant models: NR
from recruitment to
physical activity,
Exposure: Feb and Jun 2012
exam based on each
participants residential
historv (Keller et al..
2015). C-V R2 = 0.84.
smoking,
employment,
total cholesterol,
high-density
lipoprotein level,
triglyceride level,
statin use, an
index of
neighborhood
SES, education,
and income
May 2022
A-13
-------
Table A-3 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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
(first 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
May 2022
A-14
-------
Table A-3 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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. (C-V
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
(C-V R2 = 0.86) (Pi et
al.. 2016)
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; AQS = Air Quality System; BMI = body mass index; CAC = coronary artery calcium;
CATHGEN = Catheterization Genetics study; CHF = congestive heart failure; cIMT = carotid intimal-medial thickness; C-V = cross-validation; DBP = diastolic blood pressure;
ECG = electrocardiogram; GEOS-Chem = Goddard Earth Observing System with global chemical transport model; HDL = high-density lipoprotein; IAD = inter-adventitial diameter;
ICD = International Classification of Disease; IMPROVE = Interagency Monitoring of Protected Visual Environments; JHS = Jackson Heart Study; km = kilometer(s);
MESA = Multi-Ethnic Study of Atherosclerosis and Air Pollution; Ml = myocardial infarction; NHS = Nurses' Health Study; ONPHEC = Ontario Population Health and Environment
Cohort; Ox = redox weighted average of N02 and 03; PM2 5 = particulate matter with a nominal mean aerodynamic diameter less than or equal to 2.5; PM10 = particulate matter
with a nominal mean aerodynamic diameter less than or equal to 10; PM10-2 5 = particulate matter with a nominal mean aerodynamic diameter greater than 2.5 and less than or equal
to 10; r = correlation coefficient; REGARDS = REasons for Geographic and Racial Differences in Stroke; SBP = systolic blood pressure SD = standard deviation;
SES = socioeconomic states; SWAN = Study of Women's Health Across the Nation; yr = year(s).
May 2022
A-15
-------
Table A-4 Study-specific details for epidemiologic studies using accountability analyses or alternative
methods for confounder control to examine long-term PM2.5 exposure 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
three 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 one
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 one
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
May 2022
A-16
-------
Study/Location
Years
Population
(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
Zialer et al.
(2018)
Eastern U.S.
2000-2012
Medicare
N=3,892,984
Medicare
fee-for-
service
beneficiaries
and
ISM,620,778
Medicare
managed-
care
beneficiaries
Cardiovascular
stroke, heart
failure, heart
rhythm
disorders,
ischemic heart
disease,
peripheral
vascular
disease
U.S. EPA Air Quality
System: Ambient PM2.5
measurements in
operation between 1997
and 2012 and enumerated
which were in areas
designated as
nonattainment for PM2.5 in
2005. Monitors were only
included each year if the
annual percentage of valid
measurements for that
year was at least 67%.
Mean (SD) for 2002-
2004 in attainment
areas: 11.59 (1.88)
Mean (SD) for 2002-
2004 in non-attainment
areas: 14.48 (1.39)
Mean (SD) for 2010-
2012 in attainment
areas: 9.39 (1.65)
Mean (SD) for 2010-
2012 in non-attainment
areas: 11.13 (1.35)
Temperature, relative humidity,
dew point, which were measured
from climate monitors located
within 150 km of the monitoring site
Age, sex, race, rural/urban,
education, income, occupied
housing, migration rate, house
value among zip codes with
controls located within 6 miles of
the monitoring station
Smoking rate from the surrounding
county
Correlation (r): NA
Copollutant models
with: NA
EPA = Environmental Protection Agency; GEOS-Chem = Goddard Earth Observing System with global chemical transport model; HYSLPLIT = HYbrid Single-Particle Lagrangian
Integrated Trajectory; IQR = interquartile range; km = kilometer(s); NA = not applicable; PM2 5 = particulate matter with a nominal mean aerodynamic diameter less than or equal to
2.5; r= correlation coefficient; SD = standard deviation.
May 2022
A-17
-------
Table A-5 Study-specific details for epidemiologic studies of short-term PM2.5 exposure 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
globally; 107
cities in the U.S.;
25 cities in
Canada
(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.
Canada
(24 cities)
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 DOW
concentrations were calculated for each Temperature (lag
study city using ambient monitoring data g)
available from Canada's NAPS. Daily
PM2.5 concentrations were averaged
across monitors within a city when
multiple monitoring sites were present.
Year
Mean:
Warm
(April-September): £
Cold
(October-March): 6
Correlation ®: NA
Copollutant models
with: NA
May 2022
A-18
-------
Table A-5 (Continued): Study-specific details for epidemiologic studies of short-term PM2.5 exposure 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.
(2019)
North Carolina
Mar 2013-Feb
2015
18-64 yr of age Nonaccidental Hourly PM2 5 measurements from the
n = ggg out-of-hospital Wake County central site monitor were
sudden obtained though EPA's AQS data mart,
unexpected Daily concentrations were calculated by
deaths averaging hourly measurements over
24-h (midnight to midnight).
Temperature (lag
0)
RH (lag 0)
Mean: 10.93
75th: 13.22
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 CBSAs)
1999-2005
All ages
Nonaccidental Daily (24-h) mean PM2.5 concentrations DOW
from population-based monitors were
obtained from 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; EPA = Environmental Protection Agency; h
Research Network; MD = metropolitan division; NA = not applicable; NAPS = National Air Pollution Surveillance System; PM25 =
aerodynamic diameter less than or equal to 2.5; r = correlation coefficient; RH = relative humidity; yr = year(s).
= hour; MCC = Multi-City Multi-Country Collaborative
particulate matter with a nominal mean
May 2022
A-19
-------
Table A-6 Study-specific details for epidemiologic studies using alternative methods for confounder control to
examine short-term PM2.5 exposure 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
May 2022
A-20
-------
Table A-6 (Continued): Study-specific details for epidemiologic studies using alternative methods for confounder
control to examine short-term PM2.5 exposure 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
Black population and percent
of Hispanic population,
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.
(2018a)
U.S. (135 cities)
1999-2010
National
Center for
Health
Statistics
n = 7,277,274
Mortality (daily
deaths)
Obtained PM2.5 and Mean: 12.8 Meteorologic: Daily mean Correlation (r): NA
NO2 from U.S. 25th Percentile: 7.5 temperature, wind speed, and Copollutant models with: NO2
EPA's Air Quality
System
Technology
Transfer Network
75th Percentile: 16.1
sea-level pressure data
BMI = body mass index; EPA = Environmental Protection Agency; km = kilometer(s); NA = not applicable; PM25 :
less than or equal to 2.5; r = correlation coefficient; SD = standard deviation.
particulate matter with a nominal mean aerodynamic diameter
May 2022
A-21
-------
Table A-7 Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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 to
1998, based on the relationship
between monitored PM10 and PM2.5
concentrations, and primary
modeled data from 1999 to 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
May 2022
A-22
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure and mortality.
Study
Study
Population
Mortality
Outcomes
Exposure Assessment
Confounders in
Statistical Model
PM2.5
Concentrations
(Hg/m3)
Copollutant
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; survey 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
May 2022
A-23
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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
May 2022
A-24
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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.
PM2.5: 2000-2012
Follow-up: Dec
2000-Dec 2012
All cause
n =
20,744,214
deaths =
5,484,947
65+ yr of age
U.S. EPA Air Quality System (AQS)
monitors. Included monitors with at
least 8 calendar years of data,
having 9+ months 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
May 2022
A-25
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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
May 2022
A-26
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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: May
15, 1991- Dec
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 from high
school, proportion of
persons in low-income
families
Mean (SD):
(2.60)
75th: 9.07
95th: 11.97
Max: 20.00
7.37 Correlation (r): NA
Copollutant models
with: NA
May 2022
A-27
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure and mortality.
Study
Study
Population
Mortality
Outcomes
Exposure Assessment
Confounders in
Statistical Model
PM2.5
Concentrations
(Hg/m3)
Copollutant
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
1991
1996
2001
0.37-20.00
0.37-20.00
0.37-18.50
Correlation (r): NA
Copollutant models
with: O3, NO2, Ox
Range (min-max):
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 Mean (SD): 5.9
status, visible minority (2.0)
identity, Indigenous 75^- 7 1
identity, marital status,
Correlation (r): NA
Copollutant models
with: O3, NO2, Ox
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
95th: 9.7
Max: 17.2
May 2022
A-28
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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: May
= 684,400
Cerebrovascular
15, 2001-Dec 31,
deaths (non-
Respiratory
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: 9.28
(2.28)
1981-1990: 9.54
(2.13)
1991-2001: 9.69
(2.02)
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
May 2022
A-29
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure and mortality.
PM2.5
Study Mortality Confounders in Concentrations Copollutant
Study Population Outcomes Exposure Assessment Statistical Model (|jg/m3) Examination
Zhang et al. Ontario Health
(2021) Study
Ontario, Canada n = 88,615
PM2.5: 2000-2016 deaths =
7,488
30+ yr of age
Nonaccidental
Cardiovascular
Respiratory
Follow-up:
2009-2017
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.:
(1.5)
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 to 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
May 2022
A-30
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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 sulfate 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)
Ontario, Canada
PM2.5: 2000-2016
Follow-up:
2001-2016
ONPHEC
n = 5,264,985
deaths =
305,335
35-85 yr of
age
Cardiovascular 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
Sulfate: 0.73
Ammonium: 0.85
Nitrate: 0.9
Sea salt: 0.87
Mineral dust: 0.92
Black carbon: 0.97
Copollutant models
with: NA
May 2022
A-31
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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: May
15, 2001- Dec
mCCHS
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
~met 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)
octimataH iicinn onnnol Q\/orono raro/othnirih/ (O ~7\ * . ^
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
May 2022
A-32
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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: Jul
2004-Dec 2016
North Carolina All cause
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)
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 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
May 2022
A-33
-------
Table A-7 (Continued): Study-specific details for epidemiologic studies of long-term PM2.5 exposure 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; AQS = Air Quality System; 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 Multiscale Air Quality model;
EPA = Environmental Protection Agency; FRM = Federal Reference Method; GEOS-Chem = Goddard Earth Observing System-Chem; GWR = geographically weighted regression;
HF = heart failure; IMPROVE = Interagency Monitoring of Protected Visual Environments; IQR = interquartile range; mCCHS = Canadian Community Health Survey—Mortality
cohort; NA = not applicable; NAAQS = National Ambient Air Quality Standards; 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; PM = particulate matter; PM2 5 = particulate matter with a nominal
mean aerodynamic diameter less than or equal to 2.5; PM10 = particulate matter with a nominal mean aerodynamic diameter less than or equal to 10; PREMIER = Prospective
Registry Evaluating Myocardial Infarction: Events and Recovery; r= correlation coefficient; RH = relative humidity; SD = standard deviation; SE = standard error;
SES = socioeconomic status; TRIUMPH = Translational Research Investigating Underlying Disparities in Acute Myocardial Infarction; yr = year(s).
May 2022
A-34
-------
Table A-8 Study-specific details for epidemiologic studies using accountability analyses or alternative
methods for confounder control to examine long-term PM2.5 exposure and mortality.
PM2.5
Study/Location Population Exposure Concentrations
Years (Cohort) Outcome Assessment (|jg/m3) Confounders Copollutant Examination
Peterson et al.
(2020)
U.S.
1990-2010
National
Center for
Health
Statistics
n = 2,132
counties
Cardiovascular
mortality
Estimated annual
average PIVte.s-total
and component
concentrations
(sulfates, nitrates,
elemental carbon, and
organic carbon)
between 1990 and
2010 from CMAQ
Weighted annual
trend mass
concentration
(standard error):
0.134 (0.001)
County: median household
income, percent of
non-White population, and
population; age-standardized
annual COPD mortality rates
(to account for the
cumulative burden of
smoking and annual smoking
rates)
Correlation (r): NA
Copollutant models with: NA
Wei et al. (2020)
Massachusetts
2000-2012
Medicare
n = 1,503,572
All-cause Used predicted daily
mortality 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): 9.0
(1.9)
Range: 3.3-16.4
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
May 2022
A-35
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using alternative methods for confounder
control to examine long-term PM2.5 exposure 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
May 2022
A-36
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using alternative methods for confounder
control to examine long-term PM2.5 exposure 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
May 2022
A-37
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using alternative methods for confounder
control to examine long-term PM2.5 exposure and mortality.
PM2.5
Study/Location Population Exposure Concentrations
Years (Cohort) Outcome Assessment (|jg/m3) Confounders Copollutant Examination
Wu etal. (2019)
New England
(Vermont, New
Hampshire,
Connecticut,
Massachusetts,
Rhode Island,
and Maine)
2000-2012
Medicare
n = 3,300,000
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 one 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
May 2022
A-38
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using alternative methods for confounder
control to examine long-term PM2.5 exposure and mortality.
Study/Location population
Years (Cohort)
Exposure
Assessment
PM2.5
Concentrations
Outcome
(Hg/m3)
Confounders
Copollutant Examination
Yitshak-Sade et Medicare Mortality
al. (2019b)
Highly spatially Range of Mean
resolved PM2.5 data Annual PM2.5
(1 x 1 km spatial Concentrations:
resolution) from a 6.5-14.5
hybrid satellite-based
model incorporating
daily satellite remote
sensing Aerosol Optic
Depth data and
classic land-use
regression
methodologies
Temperature
Individual: age,
race/ethnicity, sex, Medicare
eligibility
Correlation (r): NA
Copollutant models with: NA
Maine, New
Hampshire,
Vermont,
Massachusetts,
Rhode Island,
Connecticut,
New York, New
Jersey,
Delaware,
Pennsylvania,
n =
15,401,064
Maryland,
Washington, DC,
Virginia, and
West Virginia
2000-2013
May 2022 A-39
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using alternative methods for confounder
control to examine long-term PM2.5 exposure 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
May 2022
A-40
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using alternative methods for confounder
control to examine long-term PM2.5 exposure and mortality.
Study/Location population
Years (Cohort)
Outcome
Exposure
Assessment
PM2.5
Concentrations
(Hg/m3)
Confounders
Copollutant Examination
Fan and Wang Medicare
(2020)
U.S.
1999-2013
Mortality
n = 770
county-month
observation for
treated
n = 7504
county-month
observation for
controls
Ambient PM2.5
monitoring data from
EPA 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
May 2022
A-41
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using alternative methods for confounder
control to examine long-term PM2.5 exposure and mortality.
PM2.5
Study/Location Population Exposure Concentrations
Years (Cohort) Outcome Assessment (|jg/m3) Confounders Copollutant Examination
Schwartz et al. Medicare Mortality Used a validated Mean (SD): 10.3
(2021) n= prediction model (3.1)
U.S. 623,036,820 calibrated to Median: 9.8
measurements at
2000-2016 person-years O|m^onnniic cda IQR: 7.9, 12.0
of follow-up
almost 2000 U.S. EPA
AQS monitoring
stations using an
ensemble of machine
learners that provided
daily estimates for a 1
km grid of the
contiguous U.S.
May 2022
A-42
Individual: age, sex, ZIP Correlation (r): NA
code, Medicaid eligibility Copollutant models with: NA
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 (LDL-c) 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)
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using alternative methods for confounder
control to examine long-term PM2.5 exposure 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
May 2022
A-43
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using alternative methods for confounder
control to examine long-term PM2.5 exposure and mortality.
PM2.5
Study/Location Population Exposure Concentrations
Years (Cohort) Outcome Assessment (|jg/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
four 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
May 2022
A-44
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using alternative methods for confounder
control to examine long-term PM2.5 exposure and mortality.
Study/Location population
Years (Cohort)
Outcome
Exposure
Assessment
PM2.5
Concentrations
(Hg/m3)
Confounders
Copollutant Examination
Ziqler et al.
(2018)
Eastern U.S.
2000-2012
Medicare
N=3,892,984
Medicare fee-
for-service
beneficiaries
and
N=1,620,778
Medicare
managed-care
beneficiaries
All-cause U.S. EPA Air Quality
mortality System: Ambient
PM2.5 measurements
in operation between
1997 and 2012 and
enumerated which
were in areas
designated as
nonattainment for
PM2.5 in 2005.
Monitors were only
included each year if
the annual percentage
of valid
measurements for that
year was at least
67%.
Mean (SD) for
2002-2004 in
attainment areas:
11.59 (1.88)
Mean (SD) for
2002-2004 in non-
attainment areas:
14.48 (1.39)
Mean (SD) for
2010-2012 in
attainment areas:
9.39 (1.65)
Mean (SD) for
2010-2012 in non-
attainment areas:
11.13 (1.35)
Temperature, relative
humidity, dew point, which
were measured from climate
monitors located within 150
km of the monitoring site
Age, sex, race, rural/urban,
education, income, occupied
housing, migration rate,
house value among zip
codes with controls located
within 6 miles of the
monitoring station
Smoking rate from the
surrounding county
Correlation (r): NA
Copollutant models with: NA
Hiabee 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 to 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
May 2022
A-45
-------
Table A-8 (Continued): Study-specific details for epidemiologic studies using alternative methods for confounder
control to examine long-term PM2.5 exposure 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 (LDL-c) 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
AQS = Air Quality System; BMI = body mass index; EPA = Environmental Protection Agency; GEOS-Chem = Goddard Earth Observing System with global chemical transport model;
HYSLPLIT = HYbrid Single-Particle Lagrangian Integrated Trajectory; IQR = interquartile range; km = kilometer; NA = not applicable; PM2 5 = particulate matter with a nominal mean
aerodynamic diameter less than or equal to 2.5; r = correlation coefficient; SD = standard deviation; yr = year(s).
May 2022
A-46
-------
Table A-9 Study-specific details for epidemiologic studies of short-term PM2.5 exposure and COVID-19
outcomes.
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
(2020)
Queens County,
NY
Mar 1, 2020-Apr
20, 2020
Daily confirmed
COVID-19 cases
and deaths
Daily average PM2.5 collected and averaged from Lagged outcome and day trend
two stationary monitors.
Mean (SD): 4.73 (2.39)
Median: 4.1
Range: 0.65 to 11.15
Correlation (r): O3 = -0.82
Copollutant models with:
NA
AQS = Air Quality System; BMI = body mass index; EPA = Environmental Protection Agency; GEOS-Chem = Goddard Earth Observing System with global chemical transport model;
HYSLPLIT = HYbrid Single-Particle Lagrangian Integrated Trajectory; IQR = interquartile range; km = kilometer(s); LDL = low-density lipoprotein; NA = not applicable;
PM2.5 = particulate matter with a nominal mean aerodynamic diameter less than or equal to 2.5; r= correlation coefficient; SD = standard deviation.
May 2022
A-47
-------
Table A-10 Study-specific details for epidemiologic studies of long-term PM2.5 exposure and COVID-19
outcomes.
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, Intensive Care
Unit beds, liquid asset poverty rate, total
health care and social services workers,
total essential workers, fraction of total
health care 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 two 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
May 2022
A-48
-------
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
Liang et al.
(2020)
U.S.
(3,122 U.S.
counties)
Jan 22, 2020 to
Jul 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
Mendv et al.
COVID-19
(2021)
hospitalizations
University of
n =1,128 COVID
Cincinnati
cases
Hospital System
n = 310
Mar 13, 2020-Jul
hospitalizations
5, 2020
Stieb et al.
SARS-CoV-2
(2020)
infections
Canada
n = 73,390
(111 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 and 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 <
lowest 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
May 2022
A-49
-------
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
Wu et al. (2020b)
U.S.
(3,089 counties
Up to Jun 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 < high school education, median house
value, % owner-occupied housing, % of
population with obesity, % current smokers,
number of hospital beds per unit population,
average daily temperature, relative humidity
for summer (June to September) and winter
(December to February) for each county,
days since issuance of stay-at-home order
for each state.
Correlation (r): NA
Copollutant models with: NA
COPD = chronic obstructive pulmonary disease; COVID-19 = coronavirus disease 2019; ICU = intensive care unit; NA = not applicable; PM2 5 = particulate matter with a nominal
mean aerodynamic diameter less than or equal to 2.5; r= correlation coefficient; SARS-CoV-2 = severe acute respiratory syndrome coronavirus 2; SD = standard deviation.
May 2022
A-50
-------
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
Jun 2013-Nov 2013
PM2.5 concentrations collected from two stationary
monitors placed in a single Low SES community,
and a High SES community.
Mean (SD):
Low SES: 11.3(2.90)
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-Brvant 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
May 2022
A-51
-------
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
< High School: 2003: 11.3, 2010: 8.2
Household Income:
> $75,000: 2003:11.1, 2010:7.9
cn
0
I
<0
cn
0
0
0
2003:
11.1,
2010:
(JO
O
CO
cn
I
<0
cn
0
0
0
0
2003:
11.2,
2010:
B.O
w
K)
0
I
<0
CO
cn
0
0
0
2003:
11.2,
2010:
B.1
Correlation (r): NA
Copollutant models with: NA
< $20,000: 2003: 11.4, 2010: 8.2
Tanzer et al. (2019)
Pittsburgh, PA
Apr 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 Real-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
NC
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
EJ = environmental justice; km = kilometer(s); NA = not applicable; NR = nor reported; NPMs = Neighborhood PM Monitors; PM = particulate matter; PM2.5 = particulate matter with
a nominal mean aerodynamic diameter less than or equal to 2.5; r= correlation coefficient; RAMP = Real-time Affordable Multi-Pollutant; SD = standard deviation;
SES = socioeconomic status.
May 2022
A-52
-------
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. Massa-
(2019a)
Massachusetts
2001-2011
Case-crossover
chusetts
Department
of Public
Health
n = 179,986
Cardio- Daily average PM2.5
vascular estimated from a model
mortality incorporating aerosol
optical depth and
monitored PM2.5 at a
1 x 1 km grid
Mean: 10.2
Max: 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)
Death Total
Daily average PM2.5
OR (95% CI):
Residential greenness,
Correlation (r): O3:
North Carolina
Records Mortality
concentrations estimated
Income
proximity to water bodies,
0.48
2002-2013
n = 775,338
from the CMAQ
downscaler model on a
< $41,500: 1.01 (1.01, 1.01)
median household
income, and classification
Copollutant models
with: NA
12x12 grid
>$41,500: 1.01 (1.00, 1.01)
of urbanicity
Mean (SD): 11.4 (5.7)
Education
Max: 70.8
< 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)
CI = confidence interval; CMAQ = Community Multi-Scale Air Quality; CV = cardiovascular; km = kilometer(s); NA = not applicable; PM2.5 = particulate matter with a nominal mean
aerodynamic diameter less than or equal to 2.5; r = correlation coefficient; SD = standard deviation.
May 2022
A-53
-------
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
2001-2015
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.
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
May 2022
A-54
-------
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
Nonaccidental
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.
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
Max: 19.02
PM2.5 more detrimental in states
with high percent of income
inequality.
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
Max: 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
May 2022
A-55
-------
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.
Canadian
(2019)
Ontario
Ontario,
Population
Canada
and Health
April 1, 2001 -
Cohort
March 31,
n -
2015
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
Max: 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, > 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 income.
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
May 2022
A-56
-------
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)
NC, U.S.
Duke
University
Medical Center
Wake,
Durham, and
Orange
Counties
2001-2010
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)
Odds ratios were
adjusted for age, sex,
BMI, race, and smoking
status
Correlation (r): NA
Copollutant models
with: NA
Wvatt et al.
(2020b)
U.S.
3,132 counties
1990-2010
National
Center for
Health
Statistics
Cardiovascular
mortality
Annual average PM2.5
concentrations were
estimated using
CMAQNR
In the 1990s, counties with highest
social deprivation benefited least,
but by 2010, counties with highest
social deprivation benefited the
most by a reduction in PM2.5.
Age, baseline year
PM2.5 and CMR for each
county
Correlation (r): NA
Copollutant models
with: NA
May 2022
A-57
-------
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
nonaccidental
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)
Nonaccidental
<$25,000: 1.61 (1.23, 2.12)
$25,000-$49,999: 1.42 (1.18, 1.72)
$50,000-$74,999: 1.14 (0.94, 1.38)
$75,000-$99,999: 1.24 (0.97, 1.60)
>$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, 1.97)
$50,000-$74,999: 1.10 (0.73, 1.68)
$75,000-$99,999: 1.11 (0.63, 1.98)
>$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
>$100,000: 4.48 (1.69, 11.83)
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)
0.84 (0.45, 1.57)
1.33 (0.63, 2.79)
1.47 (0.45, 4.74)
Correlation (r): NA
Copollutant models
with: NO2
AMI = acute myocardial infarction; AOD = aerosol optical depth; BMI = body mass index; CATHGEN = Catheterization Genetics study; CHF = congestive heart disease;
CI = confidence interval; CMAQ = Community Multi-Scale Air Quality; CTM = chemical transport model; GEOS-Chem = Goddard Earth Observing System with global chemical
transport model; hr = hazard ratio; IQR = interquartile range; km = kilometer(s); PM2.5 = particulate matter with a nominal mean aerodynamic diameter less than or equal to 2.5;
r= correlation coefficient; SD = standard deviation; SDI = social deprivation index; SES = socioeconomic status.
May 2022
A-58
-------
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
and 2012 on a 1 * 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
Overall: 15.7 (0.77)
Copollutant models with: NA
300 m buffer surrounding an
2001-2014
individual's residence for the year prior
Black: 16.1 (0.75)
to enrollment in the study
White: 15.7 (0.73)
Kellv et al. (2020)
PM2.5 concentrations estimated from
Population-weighted average (range from models)
Correlation (r): NA
U.S.
nine different exposure models which
NH-White: 9-10.3
Copollutant models with: NA
2011
can be described as either
Hispanic: 9.8-11.4
geophysical process,
interpolation-based, Bayesian
NH-Other: 9.4-11.5
statistical regression,
NH-Black: 10.1-12.1
satellite-AOD-based, or machine
learning models
Lee (2019)
PM2.5 concentrations estimated from
Mean (SD): 8.09 (3.25)
Correlation (r): NA
California
both stationary monitors and satellite
25th-75th percentiles = 5.77-9.76
Copollutant models with: NA
2016
May 2022
A-59
-------
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)
Max: 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
May 2022
A-60
-------
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)
U.S.
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
AOD = aerosol optical depth; inMAP = Intervention Model for Air Pollution; km = kilometer(s); NA = not applicable; NR = not reported; PM2.5 = particulate matter with a nominal mean
aerodynamic diameter less than or equal to 2.5; PM10 = particulate matter with a nominal mean aerodynamic diameter less than or equal to 10; r= correlation coefficient;
SD = standard deviation.
May 2022
A-61
-------
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
(H9/m3)
Select Results
Covariates in
Statistical Model
Copollutant
Examination
Yitshak-Sade et
al. (2019a)
Massachusetts
2001-2011
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
Max: 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 Residential 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
Temperature and
day of the week
Correlation (r): NA
Copollutant models
with: NA
CV = cardiovascular; km = kilometer(s); IRD = Index of Racial Dissimilarity; NA = not applicable; NR = not reported; PM2.5 = particulate matter with a nominal mean aerodynamic
diameter less than or equal to 2.5; RRS = racial residential segregation.
May 2022
A-62
-------
Table A-16 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
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
and 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
loss using 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
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
Max: 19.02
PM2.5 more detrimental in states Income share of top 10%, %
with high percent of population
of Black race.
Black, total population,
median household income,
median age, % college
degree or higher
Correlation (r):
NA
Copollutant
models with:
NA
May 2022
A-63
-------
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
Juarez et al.
(2020)
Southeastern
U.S.
2002-2009
Southern Cardio- Annual average PM2.5
Community metabolic concentrations estimated
Cohort Study disease from a continuous, spatial
n = 72 215 surface 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
Hypertension 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
Lim et al. (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)
Max: 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
May 2022
A-64
-------
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
Lipfert and
Wvzqa (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
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, NA
marital status, education, Copollutant
county-level income, region moc|els with:
of county, urbanization, and
survey year
May 2022
A-65
-------
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
Son et al. (2020) Death
North Carolina Records
2002-2013 n = 775,338
Total Mortality Daily average PM2.5
concentrations estimated
from the CMAQ downscaler
model on a 12 ><12 grid
Mean (SD): 11.4 (5.7)
Max: 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
AARP = American Association of Retired Persons; AMI = acute myocardial infarction; BMI = body mass index; CMAQ = Community Multi-Scale Air Quality; CI = confidence interval;
CV = cardiovascular; km = kilometer(s); HR = hazard ratio; IRD = Index of Racial Dissimilarity; NA = not applicable; NIH = National Institutes of Health; PM2.5 = particulate matter
with a nominal mean aerodynamic diameter less than or equal to 2.5; PM10 = particulate matter with a nominal mean aerodynamic diameter less than or equal to 10; PM10-
25 = particulate matter with a nominal mean aerodynamic diameter greater than 2.5 and less than or equal to 10; r = correlation coefficient; RRS = racial residential segregation;
SD = standard deviation; SES = socioeconomic status; WHI = Women's Health Initiative.
May 2022
A-66
-------
A.1 Advances in Estimating Light Extinction
Equation 4-1, which across studies has been referred to as the original IMPROVE equation was
originally the basis for estimating light extinction to track progress in reducing haze in visibility protected
areas for the 1999 Regional Haze Rule (Malm ct al.. 1994):
bext ~ 3 xf[RH) x [Ammonium Sulfate] + 3 xf[RH) x [Ammonium Nitrate] +
4 x [Organic Mass] + 10 x [Elemental Carbon] + 1 x [Fine Soil] + 0.6 x [Coarse Mass]
+ Rayleigh scattering (Eq. A-l)
Light extinction (bext) is in units of per Mm; [Ammonium Sulfate], [Ammonium Nitrate], [Organic Mass],
[Elemental Carbon], [Fine Soil], and [Coarse Mass] are the concentrations in (.ig/nr1 of ammonium sulfate,
ammonium nitrate, organic matter, elemental carbon, fine soil, and coarse mass, respectively: /(/W) is the
relative-humidity-dependent water growth function for both ammonium sulfate and ammonium nitrate,
and the various coefficients are empirically derived mass scattering and absorption coefficients originally
proposed by (Malm et al.. 1994). Organic mass is derived from organic carbon measurements by
multiplying measured organic carbon by a factor of 1.4 to account for the non-carbon elements that also
contribute to organic mass. Details on PM size distribution assumptions, component mass scattering
extinction efficiencies, hygroscopicity, and other assumptions used to develop Equation A-l are discussed
in detail by Malm et al. (1994) and reviewed in Section 9.2.2 of the 2009 PM ISA (U.S. EPA. 2009) and
Section 13.2.3 of the 2019 PM ISA (U.S. EPA. 2019).
Generally good performance was attributed to Equation A-l, but it underestimated high light
extinction values and overestimated low light extinction values. Over the years, changes to the original
IMPROVE equation have been adopted or proposed. Equation A-2 was developed to address bias at the
low and high light extinction values of Equation A-l (Pitchford et al.. 2007). Equation A-2 has been
referred to as the revised IMPROVE equation.
bext ~ 2.2 x fs(RH) x [Small Ammonium Sulfate] + 4.8 xfi(RH) x [Large Ammonium Sulfate] +
2.4 x/s [RH] x [SmallAmmonium Nitrate] + 5.1 * fi(RH) x [Large Ammonium Nitrate]
+ 2.8 x [Small Organic Mass] + 6.1 x [Large Organic Mass] +
10 x [Elemental Carbon] + 1 x [Fine Soil] + 1.7 x fss(RH) x [Sea Salt] +
0.6 x [Coarse Mass] + Rayleigh Scattering (Site Specific) + 0.33 x [NO2 (ppb)] (Eq. A-2)
Term definitions are the same as for Equation A-l .fs and f. are the relative-humidity-dependent water
growth functions of ammonium sulfate and ammonium nitrate in the small and large modes, respectively,
May 2022
A-67
-------
and fss is the water growth function for sea salt. Small and large ammonium sulfate, ammonium nitrate,
and organic mass are used to refer to the splitting of the concentrations of each of those three species into
two size modes.
Equation A-2 is the result of five revisions to Equation A-l: (1) the addition of a sea salt term, (2)
a change in the ratio of organic mass to organic carbon mass from 1.4 to 1.8, (3) a change in the Rayleigh
scattering term to a site-specific term based on elevation and annual temperature, (4) introduction of split-
component extinction efficiency terms to represent two particle size modes for sulfate, nitrate, and
organic mass along with a new water growth term for sulfate and nitrate, and (5) addition of a NO2 light
absorption term for monitors where NO2 concentration data are available. Sulfate, nitrate, and organic
mass are each split into a small mode and a large mode, with the fraction of the large mode estimated by
dividing the component concentration by 20 (ig/m3 and the remaining component mass attributed to the
small mode if the PM2 5 concentration is under 20 (ig/m3 and attributing all component mass to the large
mode if the concentration is greater than 20 (ig/m3. This approach is consistent with an assumption that
lower concentrations are associated with fresher emissions and smaller particle sizes, and higher
concentrations with more aged PM and larger particle sizes. Because of this particular feature, Equation
A-2 is sometimes referred to as the split-component algorithm. Further explanation of this calculation is
available at (http://vista.cira.colostate.edu/Improve/the-improve-algorithm/). Light extinction and
hygroscropicity are based on literature values, but component splitting between size modes is based on
empirical observations (Prcnni et al.. 2019). Details on component mass scattering extinction efficiencies
and hygroscopicities, and other assumptions used to develop Equation A-2 are discussed by Pitchford et
al. (2007) and in Section 9.2.3 of the 2009 PM ISA (U.S. EPA. 2009).
Good performance of Equation A-2 was reported for a wide range of PM composition and sample
loadings through 2003 (Prenni et al.. 2019). However, Equation 4-2 was based on data from 1995 to
2003, when PM concentrations were higher than they are now, and PM composition and source
contributions have also changed since then (Prenni et al.. 2019). By the time of publication of the 2019
PM ISA, new results indicated that the Equation A-2 had not been generally successful in decreasing the
bias in light extinction estimates associated with Equation A-1. For example, Equation A-2 was evaluated
by Lowenthal and Kumar (2016). who recommended a further increase in the ratio of organic mass to
organic carbon from 1.8 to 2.1, as well as the introduction of a relative-humidity-dependent water growth
function for organic mass. The basis for these recommendations was discussed in detail by Lowenthal and
Kumar (2016) and summarized in the 2019 PM ISA (U.S. EPA. 2019). Implementation of their
recommendations results in Equation A-3, which is identical in form to Equation A-2, except for the
insertion of the water growth terms /s[RH)om and f{RH) for small and large organic mass, respectively:
May 2022
A-68
-------
bext ~ 2.2 x fs(RH) x [Small Ammonium Sulfate] + 4.8 xfi(RH) x [Large Ammonium Sulfate] +
2.4 x/s (/?W] x [SmallAmmonium Nitrate] + 5.1 * f.[RH) x [Large Ammonium Nitrate]
+ 2.8 x/s(/?ffjoM x [Small Organic Mass] + 6.1 x/i(/?f/]op,i x [Large Organic Mass] +
10 x [Elemental Carbon] + 1 x [F/ne So/7] + 1.7 x fss[RH) x [Sea Sa/t] +
0.6 x [Coarse Mass] + Rayleigh Scattering [Site Specific) + 0.33 x [/VO2 (ppb]\ (Eq. A-3)
The 2011 IMPROVE report (Hand etal.. 2011) and the 2019 PM ISA (U.S. EPA. 2019) also
recognized concerns with Equation A-2, and returned to the use of Equation 4-1, but with the following
changes that had been incorporated into Equation A-2: (1) the sea salt term was added, (2) the factor used
to compute organic mass concentration from organic carbon measurements increased from 1.4 to 1.8, and
(3) the site-specific term based on elevation and mean temperature was substituted for the constant value
10/Mm for Rayleigh scattering (U.S. EPA. 2019; Hand etal.. 2011). These modifications resulted in
Equation A-4, sometimes referred to as the modified original IMPROVE equation:
bext ~ 3 xf[RH) x [Ammonium Sulfate] + 3 xf[RH) x [Ammonium Nitrate] +
4 x [Organic Mass] + 10 x [Elemental Carbon] + 1 x [Fine Soil] + 1.7 xf[RH) x [Sea Salt] +
0.6 x [Coarse Mass] + Rayleigh scattering (Eq. A-4)
More detailed discussions of Equations A-l through A-4, as well as additional studies that have evaluated
their performance are reviewed are in Sections 9.2.2 and 9.2.3 of the 2009 PM ISA (U.S. EPA. 2009) and
Section 13.2.3 of the 2019 PM ISA (U.S. EPA. 2019).
The observations of persistent and potentially increasing bias in the established method for
estimating light extinction from national monitoring network data are a concern and, as a result, it
continues to be investigated. Since publication of the 2019 PM ISA, a comparison of direct measurements
of light extinction using an integrating nephelometer with estimates of reconstructed light extinction
based on Equation A-2 at 11 monitoring locations from 2002 to 2018 showed that the relationship
between measured and reconstructed light extinction is changing (Prenni et al.. 2019). As large decreases
in sulfate and organic mass occurred over this 16-year period, the difference between measured and
reconstructed light extinction increased at the five eastern monitoring locations, indicating that Equation
A-2 increasingly underestimated light extinction overtime. Multiple linear regressions of light extinction
against PM components resulted in increasing regression coefficients over time for ammonium sulfate
and particulate organic matter at these locations.
As PM concentrations have decreased, an increasingly larger portion of PM mass has been
allocated to the smallest of the two size modes of Equation A-2 because, as described above, species mass
is apportioned between size modes by dividing the decreasing species concentration by the same fixed
factor of 20 (ig/m3 each year (Prenni et al.. 2019). Prenni et al. (2019) used particle size distribution data
May 2022
A-69
-------
from intensive field studies and from model simulations to demonstrate that changes in particle size
distributions have not actually occurred as PM concentrations have decreased, and they concluded that a
part of the apparent decreasing trend in light extinction by ammonium sulfate and particulate organic
matter is likely an artifact of Equation A-2, leading to calculations of changes in mass scattering
efficiencies that have not occurred in the atmosphere. Estimating the fraction of the large size mode by
dividing the component concentration by five times the annual median component mass calculated for
ammonium sulfate, ammonium nitrate, and organic mass for each monitor and year instead of by the 20
(ig/m3 factor previously used for all sites and years, reduced the bias, especially in later years (Prcnni et
al.. 2019).
Other sources of bias associated with both Equations A-l and A-2 that could affect reconstructed
light extinction have also been documented and reviewed by Prenni et al. (2019). Among the potential
sources of biases that have been investigated are (1) the assumption of a constant ratio for converting
measured organic carbon mass to total organic mass, in spite of spatial and seasonal variability and
observed changes during atmospheric aging; (2) the assumption that organic mass is not hygroscopic
while there is increasing evidence to the contrary; and (3) the assumption that ammonium sulfate is the
only form of atmospheric sulfate even though recent research provides evidence that it is not fully
neutralized in many U.S. locations (Prenni et al.. 2019). While these could also lead to increased bias in
light extinction estimates, the results of Prenni et al. (2019) indicate that a better understanding and
correction of the potential biases is necessary before additional revision of Equation A-2 could be
effective.
May 2022
A-70
-------
A.2 Quality Assurance Summary
The use of quality assurance (QA) and peer review help ensure that EPA conducts high-quality
science assessments that can be used to help policymakers, industry, and the public make informed
decisions. Quality assurance activities performed by EPA ensure that environmental data are of sufficient
quantity and quality to support the Agency's intended use. The ISA for Particulate Matter is classified as
a Highly Influential Scientific Assessment (HISA), which is defined by the Office of Management and
Budget (OMB) as a scientific assessment that is novel, controversial, or precedent-setting, or has
significant interagency interest. OMB requires a HISA to be peer reviewed before dissemination. To meet
this requirement, EPA engages the Clean Air Scientific Advisory Committee (CASAC) as an independent
federal advisory committee to conduct peer reviews. Both peer-review comments provided by the
CASAC panel and public comments submitted to the panel during its deliberations about the external
review draft were considered in the development of this ISA. Agency-wide, EPA Quality System
provides the framework for planning, implementing, documenting, and assessing work performed by the
Agency, and for carrying out required quality assurance and quality control (QA/QC) activities.
Additionally, the Quality System covers the implementation of EPA Information Quality Guidelines. This
ISA follows all Agency guidelines to ensure a high-quality document. Within EPA, Quality Assurance
Project Plans (QAPPs) are developed to ensure that all Agency materials meet a high standard for quality.
U.S. EPA has developed a Program-level QAPP (PQAPP) for the ISA Program to describe the technical
approach and associated QA/QC procedures associated with the ISA Program (PQAPP ID# L-HEEAD-
0030253-QP-1-5). In addition, QAPP (L-HEEAD-0030768-QP-1-0) was applied to the PM ISA Project.
All QA objectives and measurement criteria detailed in the PQAPP and QAPP have been employed in
developing this ISA. Quality assurance checks were conducted on numerical entries used in the
appendices, and at a minimum, the numbers obtained from every tenth reference cited in the appendices
were verified against the original source by an independent scientist for accuracy. Furthermore, publicly
available databases (e.g., National Emissions Inventory, Air Quality System database) from which data
were used in analyses were verified to have their own QA processes in place. U.S. EPA QA staff are
responsible for the review and approval of all quality-related documentation. Because this is a HISA, U.S.
EPA QA staff performed a Technical System Audit on the 2019 PM ISA in August 2019 and September
2020, and the Supplement to the 2019 PM ISA in March 2022. These audits verified that the appropriate
QA/QC procedures and reviews were adequately performed and documented.
May 2022
A-71
-------
References
Aaron. CP: Chervona. Y: Kawut. SM: Diez Roux. AY: Shen. M: Bluemke. DA: Van Hee. VC: Kaufmaa JD: Barr.
RG. (2016). Particulate matter exposure and cardiopulmonary differences in the multi-ethnic study of
atherosclerosis. Environ Health Perspect 124: 1166-1173. http://dx.doi.org/10.1289/ehp. 1409451
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:
Martia 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
May 2022
R-l
-------
Berrocal. VJ: Gelfand. AE: Holland. DM. (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
Bertazzon. S: Johnson. 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. RP: Pervez. S: van Ponkelaar. A: Martin. RV. (2021).
Ambient PM2.5 exposure and rapid spread of COVIP-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 Ponkelaar. 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 Pisease: 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. JP. (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 Ponkelaar. 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. P: Philip. S: van Ponkelaar. 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 Ponkelaar. 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
May 2022
R-2
-------
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
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: Lvapustin. 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
May 2022
R-3
-------
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
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
May 2022
R-4
-------
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/10.1080/10962247.2Q18.1452801
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. (2016). Long-and short-term
exposure to air pollution and inflammatory/hemostatic markers in midlife women. Epidemiology 27: 211-
220.
https://iournals.lww.com/epidem/Fulltext/2016/03000/Long and Short term Exposure to Air Pollution
and.8.aspx
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/10.1198/iasa.201 l.ap09392
Han. I: Guo. YC: 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. J: Copeland. S: Day. D: Dillner. A: Indresand. H: Malm. W: McDade. C: Moore. C. Jr: Pitchford. M:
Schichtel. B. (2011). IMPROVE (Interagency Monitoring of Protected Visual Environments): Spatial and
seasonal patterns and temporal variability of haze and its constituents in the United States. Fort Collins, Co:
Colorado State University, Cooperative Institute for Research in the Atmosphere.
http://vista.cira.colostate.edu/Improve/spatial-and-seasonal-patterns-and-temporal-variabilitv-of-haze-and-
its-constituents-in-the-united-states-report-v-iune-2011/
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
Hemmingsea 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: Lykkesfeldt. 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
May 2022
R-5
-------
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
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: Coogan. 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: 3561. 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
ChemPhys 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
May 2022
R-6
-------
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.vpmed.2013.06.009
Krall. JR: Chang. HH: Waller. LA: Mulholland. JA: Winquist. A: Talbott. EO: 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.0009096
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
Lambert. A: Hallar. AG: Garcia. M: Strong. C: Andrews. E: Hand. JL. (2020). Dust impacts of rapid agricultural
expansion on the great plains. Geophys Res Lett 47: e2020GL090347.
http://dx.doi.org/10.1029/2020GL09Q347
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-Q544-9
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: Pockerv. P: 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. P: 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 COVIP-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. P: Peuauet. PJ: Puan. Y: Whitsel. EA: Pou. 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. P: Whitsel. EA: Puan. Y: Lin. HM: Ouibrera. PM: Smith. R: Peuauet. PJ: Prineas. RJ: Zhang. ZM: Anderson.
G. (2009). Ambient particulate air pollution and ectopy-the environmental epidemiology of
May 2022
R-7
-------
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
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. (2013a). National Particle Component Toxicity
(NPACT) Initiative: Integrated epidemiologic and toxicologic studies of the health effects of particulate
matter components [HEI], (177). Boston, MA: Health Effects Institute.
https://www.healtheffects.org/svstem/files/RR177-Lippmann.pdf
Lippmann. M: Chen. LC: Gordon. T: Ito. K: Thurston. GD. (2013b). 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
Lowenthal. DH: Kumar. N. (2016). Evaluation of the IMPROVE Equation for estimating aerosol light extinction. J
Air Waste Manag Assoc 66: 726-737. http://dx.doi.org/10.1080/10962247.2Q16.1178187
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
May 2022
R-8
-------
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
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://web.archive.Org/web/20170711012213/http://www.nhlbi.nih.gov/about/documents/factbook/2012/c
hapter4 (accessed August 23, 2017).
NYC (City of New York). (2022). COVID-19 Data. Trends and totals. Available online at
https://www 1 .nvc. gov/site/doh/covid/covid-19-data-totals.page
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. JP: 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 Ponkelaar. 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
May 2022
R-9
-------
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: Donkelaar. AY: 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. D: 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: Dockerv. DW: 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/151.3 Pt 1.669
Pope. CA: Lefler. JS: Ezzati. M: Higbee. JD: Marshall. JD: 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
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
May 2022
R-10
-------
Schulz. AJ: Mentz. GB: Sampson. N: Ward. M: Dvonch. JT: De Maio. R: Israel. BA: Reves. 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: 1209.
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
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: 2523. 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: Halonen. 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
May 2022
R-ll
-------
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 (Final report,
2004) [EPA Report], (EPA 600/P-99/002aF-bF). Washington, DC.
http://cfpub.epa.gov/ncea/cfm/recordisplav.cfm?deid=87903
U.S. EPA (U.S. Environmental Protection Agency). (2009). Integrated science assessment (ISA) for particulate
matter (Final report, Dec 2009) [EPA Report]. (EPA/600/R-08/139F). Washington, DC.
http://cfpub.epa.gov/ncea/cfm/recordisplav.cfm?deid=216546
U.S. EPA (U.S. Environmental Protection Agency). (2011). Policy assessment for the review of the particulate
matter national ambient air quality standards [EPA Report]. (EPA 452/R-l 1-003). Research Triangle Park,
NC. http://nepis.epa. gov/exe/ZvPURL.cgi?Dockev=P100AUMY.txt
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://nepis.epa.gov/Exe/ZvPURL.cgi?Dockev=P100R5VE.txt
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
U.S. EPA (U.S. Environmental Protection Agency). (2022). Policy assessment for the reconsideration of the
National Ambient Air Quality Standards for particulate matter [EPA Report]. (EPA-452/R-22-004).
Research Triangle Park, NC: U.S. Environmental Protection Agency, Office of Air Quality Planning and
Standards, Health and Environmental Impacts Division, https://www.epa.gov/naaas/particulate-matter-pm-
standards-policv-assessments-current-review-0
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
USAFacts. (2022). US COVID-19 cases and deaths by state. Available online at
https://usafacts.org/visualizations/coronavirus-covid-19-spread-map/ (accessed April 14, 2022).
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
May 2022
R-12
-------
van Donkelaar. A: Martig 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.5b02076
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
Weaver. AM: Mcguinn. L: Neas. L: Mirowskv. J: Devlin. RB: Dhingra. 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: Pominici. F: Schwartz. JP. (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
May 2022
R-13
-------
Wvatt. LH; Peterson. GCL: Wade. TJ: Neas. LM: Rappold. AG. (2020b). The contribution of improved air quality
to reduced cardiovascular mortality: Declines 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. JD: Paciorek. CJ: Laden. F: Hart. JE: Puett. RC: Liao. D: 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. JD: Paciorek. CJ: Laden. F: Hart. JE: Puett. RC: Liao. D: Suh. HH. (2014b). Spatio-temporal modeling of
particulate air pollution in the conterminous United States using geographic and meteorological predictors -
Supplementary material [Supplemental Data]. Environ Health 13.
Ye. D: 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
Yitshak-Sade. M: James. P: Kloog. I: Hart. JE: Schwartz. JD: Laden. F: Lane. KJ: Fabian. MP: Fong. KC:
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: Lin. S: Hopke. PK: Thurstoa SW: van Wiingaarden. 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
Zigler. CM: Choirat. C: Dominici. F. (2018). Impact of National Ambient Air Quality Standards nonattainment
designations on particulate pollution and health. Epidemiology 29: 165-174.
http://dx.doi.org/10.1097/EDE.000000000000Q777
May 2022
R-14
------- |